Best Practices in Geometric Morphometrics for Taxonomy: A Guide for Precision Biology and Drug Discovery

Abstract

This article provides a comprehensive guide to the application of geometric morphometrics (GM) in modern taxonomy, with a special focus on implications for biomedical and drug discovery research. It covers foundational principles, from landmark selection to Procrustes analysis, and details robust methodological workflows for species discrimination in complex groups. The content addresses common troubleshooting scenarios and optimization techniques for challenging specimens, and concludes with rigorous validation protocols and comparative analyses against traditional methods. Aimed at researchers and drug development professionals, this guide serves as a critical resource for employing GM to achieve high-precision taxonomic identification, which is foundational for accurate biodiversity assessment and the discovery of biologically active compounds.

Understanding Geometric Morphometrics: Core Principles and Taxonomic Significance

Geometric morphometrics (GM) has revolutionized the quantitative analysis of biological forms by preserving complete geometric information throughout statistical analyses. This technical guide examines GM's fundamental principles, contrasting it with traditional morphometric approaches while providing detailed methodologies for taxonomic applications. We explore how coordinate-based data analysis overcomes limitations of linear measurement systems through Procrustes superimposition, which separates shape from size, position, and orientation. Within taxonomy research, GM has proven particularly valuable for distinguishing cryptic species and identifying quarantine-significant pests where traditional morphological characters show limited diagnostic power. This whitepaper synthesizes current protocols, visualization techniques, and analytical frameworks to establish best practices for implementing GM within systematic biology research.

Traditional morphometrics primarily relied on linear distances, ratios, and angles to quantify morphological variation. While useful for basic comparisons, these approaches discarded crucial geometric information about the spatial relationships between anatomical structures. Geometric morphometrics represents a paradigm shift by analyzing the complete configuration of landmarks, thus preserving the geometry of biological forms throughout statistical analyses [1].

The fundamental advantage of GM lies in its ability to statistically analyze shape variables independent of size, position, and orientation through Procrustes superimposition. This mathematical framework allows researchers to test hypotheses about form variation while visualizing results directly in morphological space. For taxonomic applications, this approach has demonstrated particular efficacy in discriminating closely related species where traditional characters show continuous variation or high phenotypic plasticity [2] [3].

Across biological disciplines, GM has resolved taxonomic uncertainties in diverse groups including fossil sharks [1], lepidopteran pests [2], thrips [4], leaf-footed bugs [3], and shrews [5]. The method's reproducibility and statistical rigor make it particularly valuable for quarantine decisions where rapid, accurate identifications are essential for biosecurity [2] [3] [4].

Fundamental Principles and Data Types

Landmarks and Semilandmarks

The foundation of GM rests on capturing biological forms through coordinated points:

• Homologous landmarks represent discrete anatomical loci that correspond across specimens (e.g., tooth cusps, suture intersections, setal bases) [1] [5]. These Type I landmarks reflect true biological homology and provide the most reliable data for taxonomic comparisons.

• Semilandmarks quantify information along curves and surfaces where discrete landmarks are insufficient. By sliding along tangent vectors to minimize bending energy, semilandmarks capture outline geometry while allowing statistical comparison [1]. For example, Pagliuzzi et al. used eight semilandmarks along the ventral margin of fossil shark tooth roots where no homologous points could be detected [1].

Table 1: Landmark Types in Geometric Morphometrics

Type	Definition	Taxonomic Application	Example
Type I (Homologous)	Discrete anatomical points at tissue intersections	Primary data for phylogenetic comparisons	Landmark #11 on thrips head: anterior base of occipital setae [4]
Type II (Mathematical)	Points of maximum curvature or extremal positions	Supplement Type I landmarks in sparse regions	LM10 on astragalus: peak point of medial protuberance [6]
Semilandmarks	Points along curves and surfaces	Capturing outline morphology without homologous points	Eight equidistant points along shark tooth root ventral margin [1]
Sliding Semilandmarks	Semilandmarks optimized by minimizing bending energy	Complex biological shapes with smooth contours	Pronotum outlines in Acanthocephala bugs [3]

The Procrustes Framework

Generalized Procrustes Analysis (GPA) standardizes raw landmark coordinates by translating, scaling, and rotating configurations to optimize fit [5] [6]. This process removes non-shape variation through three mathematical operations:

• Translation: Centering each configuration at the origin (0,0) by subtracting centroid coordinates
• Scaling: Scaling configurations to unit centroid size (the square root of summed squared distances from landmarks to their centroid)
• Rotation: Rotating configurations to minimize the sum of squared distances between corresponding landmarks

The resulting Procrustes coordinates represent pure shape variables that can be analyzed using multivariate statistics while preserving their geometric relationships [5]. This framework enables direct visualization of shape differences as actual morphological changes rather than abstract numerical outputs.

Experimental Protocols and Workflows

Data Acquisition and Imaging

Successful GM analysis requires consistent, high-quality specimen imaging:

• 2D Photography: Standardized orthogonal views with scale references for relatively flat structures [3] [6]. Smith-Pardo et al. used high-resolution images of slide-mounted thrips for head and thorax analysis [4].

• 3D Surface Scanning: For complex morphological structures, 3D scanners capture comprehensive surface topology [7]. Darkling beetle studies used six scanning orientations to ensure complete surface reconstruction [7].

• Micro-CT Scanning: Internal structures and minute morphological features can be digitized through computed tomography [8]. This approach has revolutionized analysis of craniodental morphology in shrews [5] and other small mammals.

Table 2: Research Reagent Solutions for Geometric Morphometrics

Tool/Category	Specific Examples	Function in GM Workflow
Imaging Equipment	Canon 600D DSLR [6], Shining 3D EinScan Pro 2X 3D scanner [7], micro-CT scanners	Digital capture of specimen morphology
Digitization Software	TPSDig2 [1] [3] [4], 3D Slicer [7]	Landmark and semilandmark placement on digital specimens
Shape Analysis Platforms	MorphoJ [2] [3] [4], R geomorph package [3] [4] [7]	Statistical shape analysis and visualization
Data Processing Utilities	TPSUtil [6], Deformetrica [8]	File format conversion, landmark-free analysis

Landmarking Protocols

Consistent landmark application is critical for reproducible results:

• Landmark Definition: Precisely define anatomical criteria for each landmark position
• Digitization Sequence: Apply landmarks in consistent order to minimize error
• Repeatability Assessment: Conduct multiple digitizations of subset to quantify operator error
• Validation: For taxonomic studies, include specimens of known identity to test landmark efficacy

For example, in Chrysodeixis moth identification, researchers used seven forewing venation landmarks that consistently discriminated between invasive C. chalcites and native C. includens [2].

Analytical Workflow

The following diagram illustrates the standard GM analytical pipeline from raw data to taxonomic interpretation:

Analytical Methods in Geometric Morphometrics

Core Statistical Approaches

• Principal Component Analysis (PCA): Identifies major axes of shape variation within the sample without a priori groupings. In thrips taxonomy, the first three PCs accounted for 73% of head shape variation, effectively separating T. australis and T. angusticeps [4].

• Canonical Variate Analysis (CVA): Maximizes separation among predefined groups, ideal for testing species boundaries. CVA successfully discriminated 11 Acanthocephala bug species based on pronotum shape [3].

• Procrustes ANOVA: Tests for shape differences between groups while accounting for allometric effects. Studies of bovid astragali found no significant size effect on shape (0.99% prediction, p=0.1634), enabling pure shape-based taxonomic discrimination [6].

• Mahalanobis Distances: Measures multivariate divergence between group means, accounting for within-group covariance. Permutation tests of Mahalanobis distances provided statistical support for thrips species separations [4].

Advanced Methodological Extensions

• 3D Geometric Morphometrics: Darkling beetle studies demonstrated how 3D GM captures taxonomic differences in prothorax and pterothorax morphology that 2D approaches might miss [7].

• Functional Data Geometric Morphometrics (FDGM): Represents landmark data as continuous curves rather than discrete points, potentially capturing more subtle shape variations [5].

• Landmark-Free Methods: Techniques like Deterministic Atlas Analysis (DAA) eliminate manual landmarking, enabling comparisons across highly disparate taxa [8].

The following diagram contrasts traditional morphometrics with modern geometric approaches:

Taxonomic Applications and Case Studies

Fossil Shark Teeth Identification

Pagliuzzi et al. directly compared traditional and geometric morphometrics on the same sample of 120 lamniform shark teeth. Both methods recovered the same taxonomic separation, but GM captured additional shape variables overlooked by traditional approaches [1]. This demonstrates GM's capacity to extract more comprehensive morphological information from the same specimens, particularly valuable for fossil material where other characters are unavailable.

Invasive Pest Biosecurity

GM has become instrumental in agricultural biosecurity for distinguishing invasive species from morphologically similar natives:

• Chrysodeixis Moths: Wing GM discriminated invasive C. chalcites from native C. includens, providing a rapid identification method superior to time-consuming genitalia dissection or DNA analysis [2].

• Thrips Species: Head and thorax shape analysis separated quarantine-significant from non-significant Thrips species, creating identification tools for port inspectors [4].

• Leaf-Footed Bugs: Pronotum shape variation successfully discriminated Acanthocephala species of quarantine concern, enabling reproducible identifications where traditional keys are inadequate [3].

Paleontological and Archaeological Taxonomy

• Bovid Astragali: GM completely separated bovine and ovine astragali (100% classification), with caprine samples largely distinct (97.2%), providing powerful tools for zooarchaeological identification [6].

• Fossil Shrews: Craniodental GM supported taxonomic classification of fossil Soricidae, revealing shape associations with dietary specialization [5].

Best Practices for Taxonomic Research

Experimental Design Considerations

• Sample Sizes: Balance statistical power with practical constraints; studies cited typically used 40-150 specimens per group [1] [6]

• Landmark Density: Distribute landmarks to adequately capture morphology without oversampling redundant information

• Validation Samples: Always include specimens of known identity to test classification accuracy

• Measurement Error: Quantify digitization error through repeated measurements and incorporate into statistical models

Visualization and Interpretation

Effective visualization communicates shape differences intuitive to taxonomists:

• Thin-Plate Spline Deformation Grids: Show continuous shape change between reference and target forms [9]

• Wireframe Graphs: Connect landmarks with lines to maintain anatomical context during shape comparison

• Principal Component Plots: Visualize specimen distribution in morphospace with minimum spanning trees or confidence ellipses

Geometric morphometrics represents a fundamental advancement over traditional measurement approaches by preserving complete geometric information throughout analysis. The Procrustes framework provides a mathematically rigorous method for analyzing shape independent of size, while various statistical tools enable hypothesis testing about taxonomic boundaries. As GM methodologies continue evolving with 3D imaging, landmark-free approaches, and functional data analysis, their applications in taxonomy will expand accordingly. Properly implemented, GM offers taxonomists powerful tools for discriminating cryptic species, resolving complexes, and providing quantitative support for systematic decisions.

The Role of Landmarks and Semilandmarks in Capturing Biological Form

Geometric morphometrics (GM) has revolutionized the quantitative analysis of biological form by providing powerful methods to quantify and statistically analyze shape variation. This approach fundamentally relies on the precise capture of anatomical geometry using landmarks and semilandmarks, which serve as the primary data points for shape analysis. Landmarks are defined as discrete, anatomically homologous points that correspond across specimens in a biological study, while semilandmarks are used to capture the geometry of curves and surfaces between these fixed landmarks. Together, these points enable researchers to quantify complex biological shapes with mathematical precision, preserving the geometric relationships among structures throughout statistical analysis.

The application of GM has expanded dramatically across biological disciplines, with Google Scholar showing an increase from approximately 50 results in 1998 to around 76,000 in 2024 for the keywords "geometric" and "morphometrics" [10]. In taxonomy research specifically, GM has become an indispensable tool for discriminating between closely related species and understanding morphological evolution, particularly when dealing with structures that exhibit continuous curvature and complex geometries that cannot be adequately captured by traditional linear measurements alone. The power of GM lies in its ability to separate biologically meaningful shape variation from other sources of variation such as size, position, and orientation through statistical procedures like Procrustes superimposition.

Theoretical Foundations: Defining Landmarks and Semilandmarks

Landmark Types and Biological Significance

Landmarks in geometric morphometrics are classified based on their anatomical definition and biological significance. Type I landmarks are defined by local biological features, such as the intersection of sutures or small foramina, and represent discrete anatomical points that are clearly homologous across specimens. Type II landmarks are defined by local geometry, such as the point of maximum curvature on a structure, while Type III landmarks are extremal points that may represent the furthest extension of a structure in a particular direction. The classification is crucial because Type I landmarks generally have the highest biological homology, making them most valuable for taxonomic comparisons across divergent groups.

In practice, landmark configurations must adequately represent the biological form under investigation. For example, in a study of fossil shark teeth, researchers used 7 homologous landmarks placed at key positions such as the apex of the crown and the extremities of the root to capture the overall tooth shape [1]. These landmarks were carefully selected to represent homologous positions across different species, enabling meaningful taxonomic comparisons. The strategic placement of landmarks ensures that the resulting shape variables capture biologically meaningful variation relevant to taxonomic questions.

Semilandmarks: Capturing Curves and Contours

Semilandmarks address a fundamental limitation of traditional landmarks: their inability to adequately capture information from curves and surfaces where discrete homologous points are sparse. Semilandmarks are points placed along curves and surfaces between traditional landmarks, allowing for the quantification of homologous regions rather than just discrete points. Unlike traditional landmarks, semilandmarks are considered "deficient" in terms of homology because their specific locations along a curve are not defined by unique biological features, but rather by their relative positions between fixed landmarks [10].

The application of semilandmarks is particularly valuable for structures with smooth contours or complex surfaces. In the shark tooth study mentioned previously, researchers supplemented their 7 traditional landmarks with 8 semilandmarks placed along the curved profile of the ventral margin of the tooth root where no homologous points could be detected [1]. This approach allowed them to capture the complete shape of the tooth, including curved regions that would otherwise be poorly represented. Similarly, in plant biology, semilandmarks are frequently employed to capture the contours of leaves and flowers where homologous points are limited [10].

Table 1: Comparison of Landmark Types in Geometric Morphometrics

Landmark Type	Definition	Biological Homology	Example Applications
Type I	Defined by local biological features (e.g., suture intersections)	High	Cephalometric points in skulls [11], foramina in bones
Type II	Defined by local geometry (e.g., point of maximum curvature)	Moderate	Tooth cusps, leaf tips [10]
Type III	Extremal points (e.g., furthest extensions)	Lower	Wing tips, leaf marginal points [10]
Semilandmarks	Points along curves/surfaces between fixed landmarks	Relative (based on sliding algorithm)	Tooth roots [1], leaf contours [10]

Applications in Taxonomy Research

Taxonomic Discrimination with Landmark-Based Approaches

Geometric morphometrics has proven particularly valuable in taxonomic research where morphological differences may be subtle yet biologically significant. The precision offered by landmark-based approaches enables researchers to detect minimal morphological differences that are often difficult to observe through qualitative assessment alone. In a compelling example from paleontology, landmark-based GM successfully discriminated between isolated teeth of different lamniform shark genera, validating qualitative taxonomic identifications while capturing additional shape variables that traditional morphometric methods did not consider [1].

The taxonomic power of GM stems from its ability to quantify entire shapes rather than isolated measurements. When applied to the same sample of shark teeth, GM recovered the same taxonomic separation identified by traditional morphometrics while providing a larger amount of information about tooth morphology [1]. This comprehensive shape capture makes GM particularly effective for classifying morphologically similar taxa where traditional characters may overlap. The method has been successfully applied across diverse taxonomic groups, from fish and mammals to plants, demonstrating its broad utility in systematic biology.

Handling Morphological Complexities in Taxonomic Studies

Many biological structures used in taxonomy present challenges for quantitative analysis due to their complex geometries and limited homologous points. Semilandmarks specifically address this limitation by enabling the capture of homologous curves and surfaces. In plant taxonomy, for instance, landmarks and semilandmarks have been extensively applied to analyze leaf and flower structures, which often lack discrete homologous points but exhibit characteristic shapes that are taxonomically informative [10].

The combination of landmarks and semilandmarks creates a more complete representation of biological form, which is particularly important for taxonomic studies focusing on structures with complex curvatures. A review of GM applications in plant science found that leaves and flowers were among the most frequently analyzed structures, with researchers using both landmark-based and semilandmark-based approaches to capture shape variations with taxonomic significance [10]. This approach has enabled more precise discrimination between closely related plant species that may be difficult to distinguish using traditional characters alone.

Table 2: Research Applications of Landmarks and Semilandmarks in Taxonomy

Research Domain	Biological Structure	Landmark Approach	Taxonomic Utility
Paleontology [1]	Shark teeth	7 landmarks + 8 semilandmarks	Discrimination of fossil lamniform genera
Entomology [11]	Drosophila wings	13 landmarks	Species identification and developmental studies
Ichthyology [11]	Zebrafish skeleton	25 landmarks	Skeletal development and phenotypic analysis
Botany [10]	Leaves, flowers	Landmarks + semilandmarks for contours	Species discrimination and adaptive morphology
Biomedical [12]	Human arm shape	Landmarks + semilandmarks	Nutritional status classification

Experimental Protocols and Methodologies

Data Acquisition and Landmark Digitization

The initial phase of any geometric morphometric study involves careful data acquisition and landmark digitization. High-quality imaging is paramount, with researchers using various modalities including computed tomography (CT), surface scanning, or standard photography depending on the specimen and research question. For 2D analyses, consistent orientation and lighting are critical, while 3D analyses require complete capture of the specimen's geometry. In a comprehensive study of 322 mammalian crania, researchers used both CT and surface scans, addressing modality differences through Poisson surface reconstruction to create watertight, closed surfaces for all specimens [8].

Landmark digitization follows standardized protocols using specialized software. For 2D images, tools like TPSdig2 are commonly employed, while 3D datasets may require more sophisticated visualization and annotation software. The process demands careful training to ensure consistency, particularly when multiple researchers are involved. In the shark tooth study, researchers used TPSdig2 to digitize landmarks and semilandmarks on isolated teeth, ensuring consistent placement across all specimens [1]. For semilandmarks, an additional step involves defining curves between fixed landmarks along which semilandmarks are initially placed before sliding procedures.

Landmark Detection Algorithms and Automation

Manual landmark placement remains time-consuming and potentially susceptible to operator bias, especially with large datasets. Consequently, automated and semi-automated landmark detection methods have emerged as valuable tools for improving efficiency and repeatability. These approaches typically use machine learning algorithms trained on manually landmarked datasets to predict landmark positions in new specimens [11].

One automated method employs a multi-resolution tree-based approach using Extremely Randomized Forests for landmark detection [11]. This method extracts multi-resolution features around each pixel and uses ensemble machine learning to predict whether a pixel corresponds to a landmark position (classification approach) or its distance to the nearest landmark (regression approach). The algorithm has been successfully applied to diverse datasets including cephalometric radiographs, zebrafish skeletons, and Drosophila wings, achieving recognition performances competitive with existing approaches while being generic and fast [11]. Another emerging approach, Large Deformation Diffeomorphic Metric Mapping (LDDMM), offers a landmark-free alternative that uses control points and momentum vectors to capture shape variation without predefined landmarks [8].

Statistical Analysis of Landmark Data

Following landmark digitization, the raw coordinate data undergoes Procrustes superimposition to remove the effects of size, position, and orientation, isolating pure shape variation. This process involves three mathematical operations: translation (centering configurations at the origin), scaling (normalizing to unit size), and rotation (aligning configurations to minimize distances between corresponding landmarks) [10]. The resulting Procrustes coordinates represent shape variables that can be analyzed using multivariate statistical methods.

For semilandmarks, an additional step called "sliding" is required to minimize the artificial variance introduced by their initial placement. Semilandmarks are allowed to slide along tangents to curves or surfaces until they minimize bending energy or Procrustes distance between specimens, effectively optimizing their positions to best represent the biological shape variation [1]. The aligned landmark and semilandmark coordinates then serve as input for various multivariate analyses, including Principal Component Analysis (PCA) for exploring major shape trends, Canonical Variate Analysis (CVA) for group discrimination, and Partial Least Squares (PLS) for analyzing covariation between structures.

Current Methodological Advances and Challenges

Landmark-Free Methods and Emerging Alternatives

While landmark-based approaches remain the gold standard in geometric morphometrics, emerging landmark-free methods offer promising alternatives, particularly for analyses across highly disparate taxa where homologous points may be limited. Methods such as Deterministic Atlas Analysis (DAA) using Large Deformation Diffeomorphic Metric Mapping (LDDMM) enable shape comparison without predefined landmarks by quantifying the deformation required to match specimens to a computed atlas shape [8]. These approaches generate control points and momentum vectors that capture shape variation, effectively bypassing the need for manual landmark identification.

Comparative studies between traditional landmark-based and landmark-free approaches reveal both strengths and limitations. In a comprehensive analysis of 322 mammal crania, DAA produced comparable but varying estimates of phylogenetic signal, morphological disparity, and evolutionary rates when compared to high-density geometric morphometrics [8]. The landmark-free approach showed particular promise for large-scale studies across disparate taxa due to enhanced efficiency, though challenges remained in certain groups like Primates and Cetacea. This suggests that landmark-free methods may serve as complementary approaches rather than replacements for traditional landmark-based morphometrics, especially in taxonomically broad studies.

Addressing Methodological Challenges in Taxonomic Research

Several methodological challenges persist in the application of landmarks and semilandmarks to taxonomic research. One significant issue involves the handling of incomplete specimens common in paleontological and museum collections. In the shark tooth study, researchers addressed this by excluding incomplete specimens from analysis, as missing data would prevent reliable statistical comparisons [1]. Alternative approaches include estimation of missing landmarks using reconstruction algorithms or focusing analyses on regions common to all specimens.

Another challenge concerns the selection of appropriate landmarks for taxonomic questions. When comparing highly disparate taxa, the number of discernible homologous landmarks decreases, potentially limiting biological inferences [8]. Semilandmarks partially address this issue by capturing homologous curves and surfaces, though their sliding algorithms introduce mathematical complexities. Additionally, the integration of geometric morphometric data with phylogenetic frameworks requires specialized approaches to account for evolutionary relationships when assessing taxonomic boundaries based on shape differences.

Table 3: Essential Research Tools for Geometric Morphometrics

Tool Category	Specific Software/Solutions	Primary Function	Application Context
Digitization Software	TPSdig2 [1]	Landmark/semilandmark placement	2D coordinate capture
3D Analysis	Deformetrica [8]	Landmark-free shape analysis	3D surface and volume data
Automated Landmarking	Cytomine [11]	Machine learning-based detection	High-throughput studies
Statistical Analysis	MorphoJ, R geomorph	Procrustes analysis & statistics	Multivariate shape analysis
Data Integration	GIS and phylogenetic tools	Spatial and evolutionary context	Comparative taxonomy

Best Practices for Taxonomic Applications

Optimizing Landmark Schemes for Taxonomic Discrimination

Developing effective landmark schemes requires balancing anatomical coverage with biological homology. For taxonomic studies, landmark configurations should include sufficient Type I landmarks to establish firm homologies, supplemented by Type II and III landmarks and semilandmarks to capture comprehensive shape information. The specific landmark scheme should be tailored to the taxonomic question and the anatomical structures under investigation. In practice, pilot studies testing different landmark configurations can help identify the most informative scheme for discriminating between taxonomic groups.

Documentation and standardization of landmark protocols are particularly important for taxonomic research to ensure reproducibility and facilitate comparisons across studies. Detailed descriptions of landmark definitions, along with visual guides illustrating their placement, should be included in methodological sections. When semilandmarks are employed, researchers should specify the curves along which they were placed, the initial spacing, and the sliding criterion used (e.g., minimum bending energy vs. Procrustes distance). This transparency enables other researchers to replicate methods and build upon existing work.

Integrating Landmark Data with Other Taxonomic Characters

While geometric morphometrics provides powerful tools for quantifying shape variation, effective taxonomy typically integrates multiple lines of evidence. Landmark-based shape data should be considered alongside traditional characters, genetic data when available, and ecological information to develop robust taxonomic hypotheses. The strength of GM lies in its ability to detect and quantify subtle shape differences that may not be apparent through qualitative observation alone, providing statistical support for taxonomic decisions.

For studies specifically focused on classification, such as developing identification tools for closely related species, linear discriminant analysis applied to shape coordinates has proven effective [12]. However, researchers must be cautious about applying classification rules derived from one sample to new specimens, as shape spaces are sample-dependent. When classifying out-of-sample individuals, careful consideration must be given to registration methods and template selection to ensure proper alignment to the reference shape space [12]. This is particularly relevant for taxonomic identification tools intended for field use or automated applications.

Landmarks and semilandmarks represent fundamental tools in geometric morphometrics, enabling precise quantification of biological form for taxonomic research. When applied according to best practices, these methods provide powerful approaches for discriminating between closely related taxa, understanding morphological evolution, and developing identification tools. The integration of traditional landmarks with semilandmarks allows researchers to capture both discrete homologous points and continuous geometrical features, creating comprehensive representations of biological shapes.

As methodological advances continue to emerge, including automated landmark detection and landmark-free approaches, the taxonomic applications of geometric morphometrics are likely to expand further. However, these technological developments must be grounded in rigorous biological understanding, with careful attention to homology and anatomical correspondence. By adhering to established best practices while embracing innovative approaches, researchers can leverage the full power of landmarks and semilandmarks to address fundamental questions in taxonomy and systematic biology.

Procrustes superimposition constitutes a foundational step in geometric morphometrics (GMM), enabling the precise isolation of shape variation by removing the confounding effects of position, orientation, and scale. This technical guide details the core principles, mathematical formulations, and practical protocols for implementing Procrustes methods within taxonomic research. By providing a standardized framework for quantifying pure shape, these analyses empower taxonomists to discriminate between closely related species, identify cryptic morphological variation, and test evolutionary hypotheses with enhanced statistical power and visual clarity. Framed within best practices for taxonomy, this whitepaper serves as a comprehensive resource for researchers seeking to integrate robust shape analysis into their investigative toolkit.

Taxonomy, the science of defining biological groups, relies heavily on morphology for discrimination. Geometric morphometrics (GMM) has emerged as a superior framework for quantifying phenotypic differences, with Procrustes superimposition at its core [13] [14]. Shape is formally defined as the geometric properties of an object that remain after normalizing for differences in location, orientation, and scale [15]. This definition is critical for taxonomy, as it allows researchers to compare organisms based solely on biologically relevant morphological variation, rather than arbitrary differences arising from how a specimen was placed or measured.

The power of GMM, and Procrustes analysis specifically, lies in its ability to preserve geometric relationships throughout the analysis. Unlike traditional morphometrics, which relies on linear distances and ratios, GMM uses the relative positions of anatomical landmarks to capture the geometry of a structure. This enables both powerful statistical quantification and intuitive visualization of shape changes, for instance, via deformation grids [14]. In taxonomy, this is particularly effective when applied to structures like the marmot mandible [13], insect pronotum [15], or leaf outlines [14], facilitating the identification of evolutionarily significant units and the resolution of complex taxonomic challenges.

The Mathematical Foundation of Procrustes Superimposition

Procrustes superimposition aims to optimally align two or more landmark configurations using similarity transformations (translation, rotation, and scaling) to minimize the sum of squared distances between corresponding landmarks [16] [17]. The core mathematical procedure for aligning multiple configurations is known as Generalized Procrustes Analysis (GPA) [18].

The Core Procrustes Procedure

Consider a set of ( k ) landmark configurations, each represented by a ( n \times p ) matrix of coordinates for ( n ) landmarks in ( p ) dimensions. The GPA procedure follows these steps:

• Translation: Each configuration is translated so that its centroid (the mean of its landmark coordinates) is at the origin of the coordinate system. This is achieved by centering the configuration matrix ( \mathbf{X}i ): ( \mathbf{X}{c,i} = \mathbf{X}i - \mathbf{1}n \mathbf{x}{0,i}^T ) where ( \mathbf{1}n ) is a vector of ones and ( \mathbf{x}_{0,i}^T ) is the centroid of the ( i )-th configuration [16].

• Scaling: Each translated configuration is scaled to unit centroid size. Centroid size (CS) is defined as the square root of the sum of squared distances of all landmarks from their centroid: ( CS(\mathbf{X}i) = \sqrt{\sum{j=1}^{n} \lVert \mathbf{x}{j,i} - \mathbf{x}{0,i} \rVert^2} ) The scaled configuration is ( \mathbf{X}{s,i} = \mathbf{X}{c,i} / CS(\mathbf{X}_i) ) [18].

• Rotation: The scaled configurations are rotated to minimize the Procrustes distance relative to a reference shape (often the mean shape). For two configurations ( \mathbf{X}s ) and ( \mathbf{Y}s ), the optimal rotation matrix ( \mathbf{R} ) is found by maximizing ( tr(\mathbf{R}^T \mathbf{X}s^T \mathbf{Y}s) ). The solution involves the singular value decomposition (SVD): ( \mathbf{Y}s^T \mathbf{X}s = \mathbf{U} \mathbf{\Lambda} \mathbf{V}^T ) The optimal rotation is then ( \mathbf{R} = \mathbf{V} \mathbf{U}^T ) [16] [17].

These steps are applied iteratively in GPA until the mean shape and the sum of squared Procrustes distances stabilize.

The Procrustes Distance and the Consensus Shape

Following GPA, the variation that remains is Procrustes shape variance. The distance between two optimally superimposed shapes ( \mathbf{X}p ) and ( \mathbf{Y}p ) is the Procrustes distance, defined as the square root of the sum of squared differences between their corresponding landmark coordinates [16]: ( D{Proc}(\mathbf{X}p, \mathbf{Y}p) = \sqrt{\sum{j=1}^{n} \lVert \mathbf{x}{p,j} - \mathbf{y}{p,j} \rVert^2} )

The iterative process of GPA produces a consensus (mean) shape—the average of all aligned configurations. This consensus serves as a central reference for describing shape variation within the sample and is crucial for visualizing differences from the mean [18].

Figure 1: The Generalized Procrustes Analysis (GPA) Workflow. This iterative process removes non-shape variation to produce aligned coordinates for analysis.

A Protocol for Taxonomic Research: Analyzing Leaf Morphology

To illustrate a practical application, we outline a simplified protocol for a taxonomic study of leaf morphology, adapted from Viscosi & Cardini [14]. This protocol demonstrates a hierarchical design to assess variation from the population level down to measurement error.

Experimental Workflow and Data Acquisition

• Plant Material: A balanced design is recommended. For example, collect two leaves from each of 22 randomly selected trees from two geographic populations of a sessile oak species.
• Landmarking: Press, dry, and scan leaves. Digitize 11 Type I and II landmarks on the right half of each leaf blade to capture key anatomical points (e.g., apex, base, sinus tip). To assess measurement error, repeat the entire digitization process two weeks after the first round [14].
• Data Preparation: The collected landmark coordinates are stored in a 3D array for analysis. The dataset should be organized hierarchically: Population > Tree > Leaf > Replicate.

Statistical Analysis of Shape Variation

The aligned Procrustes coordinates are analyzed to partition variance across hierarchical levels and test for significant group differences.

• Procrustes ANOVA: A multivariate ANOVA is performed on the shape data to partition the sum of squares and mean squares across the factors: Population, Tree (nested within Population), Leaf (nested within Tree), and Measurement Error [14].
• Hypothesis Testing: The significance of each factor is tested using permutation tests (e.g., 1000+ permutations) to obtain p-values. This determines if population differences are larger than differences among trees, and so on.
• Discrimination Accuracy: The ability of shape to correctly classify leaves into their source populations can be estimated using linear discriminant analysis and cross-validation. The effect of allometry (size-related shape change) can be investigated by correlating Procrustes coordinates with centroid size and re-estimating discrimination accuracy after removing the allometric component [14].

Table 1: Key Software and Tools for Procrustes-Based Geometric Morphometrics

Software/Package	Language	Primary Function	Application in Taxonomy
geomorph [13] [19]	R	Comprehensive GM analysis: GPA, Procrustes ANOVA, modularity tests	Standard toolkit for morphological divergence studies.
Morpho [19]	R	Shape analysis and visualization: Procrustes registration, outlier detection	Processing 3D landmark data, e.g., from skulls.
Momocs [13] [19]	R	Outline and landmark analysis	Analyzing leaf outlines or insect wings.
morphospace [19]	R	Building and visualizing morphospaces	Creating publication-ready ordination plots.
alignProMises [17]	R	Advanced Procrustes alignment with priors	Aligning high-dimensional data (e.g., neuroimaging).
vegan [20]	R	Ordination and ecological analysis	Comparing ordinations via procrustes() and protest().

Advanced Considerations and Best Practices

Handling Complex Articulating Structures

For rigid structures, standard GPA is sufficient. However, taxonomic studies often involve complex articulating structures (e.g., fish skulls, arthropod exoskeletons) where arbitrary differences in the resting position of elements confound biological shape variation [18]. In such cases, local superimposition techniques are required. These methods involve:

• Independent GPA: Performing separate Procrustes superimpositions on landmark subsets defining each rigid articulating element.
• Recombination: Concatenating the independently superimposed coordinates into a common shape variable [18]. The "matched local superimpositions" approach places the superimposed subsets into an anatomically realistic reference configuration, preserving biological interpretability [18].

Measurement Error and Statistical Power

A critical but often neglected preliminary analysis is the assessment of measurement error [13]. This is easily done by replicating the digitization process and conducting a Procrustes ANOVA. In the leaf morphology example, measurement error was found to be "completely negligible," providing confidence that the observed shape variation was biological in origin [14]. Furthermore, statistical power in morphometrics is strongly influenced by sample size. Studies with small sample sizes per group may fail to detect biologically meaningful differences and are more susceptible to sampling error [13]. Power analysis should be conducted during the experimental design phase.

Table 2: Summary of Key Quantitative Findings from Case Studies

Study System	Sample Size	Key Finding (Procrustes ANOVA)	Taxonomic Implication
Sessile Oak Leaves [14]	2 pops, 22 trees/pop, 2 leaves/tree	Measurement error negligible; individual tree variation > small population differences.	Confirms species identity; highlights high individual plasticity.
Tetropium Beetles [15]	42 specimens, 9 species	Pronotum shape effectively distinguishes most of the 9 beetle species.	GM is a valid tool for identification of cryptic and quarantine species.
Marmot Mandibles [13]	Large sample, multiple species	-	A large sample enables robust exploration of interspecific morphological variation.

Figure 2: Decision Framework for a Taxonomic GM Study. This workflow integrates best practices from data collection to final interpretation.

Procrustes superimposition is more than a statistical pre-processing step; it is the cornerstone of rigorous shape analysis in modern taxonomy. By providing a mathematically sound method for isolating shape from other sources of geometric variation, it enables the precise quantification and visualization of morphological differences essential for discriminating species, identifying cryptic diversity, and understanding evolutionary patterns. Adherence to best practices—including careful landmark selection, assessment of measurement error, appropriate sample sizes, and the use of specialized methods for complex structures—ensures that taxonomic conclusions drawn from shape data are both robust and biologically meaningful. As geometric morphometrics continues to evolve, Procrustes-based methods will remain integral to the taxonomist's toolkit for exploring and documenting the phenotypic dimension of biodiversity.

Taxonomy, the science of classification, lays the foundational framework for studying biodiversity and its conservation [13]. In this context, Geometric Morphometrics (GMM) has emerged as a powerful methodology for quantifying biological shape, enabling rigorous comparison of phenotypic differences among populations and species [13] [21]. Unlike traditional measurement approaches that treat form as a set of isolated linear distances, GMM captures the geometric configuration of homologous landmarks, thereby preserving the spatial relationships throughout analysis [21]. Principal Component Analysis (PCA) serves as a critical statistical technique within this framework, allowing researchers to visualize and interpret the major patterns of shape variation across specimens in a reduced-dimensional space, known as a morphospace [21] [22].

The power of PCA in taxonomic research lies in its ability to transform complex, correlated landmark coordinates into a new set of uncorrelated variables—the principal components [21]. Each component describes an axis of continuous shape variation within the sample, ordered by the amount of variance they explain [23]. This transformation enables the identification of the most significant patterns of morphological disparity, which may reflect underlying phylogenetic relationships, ecological adaptations, or allometric growth patterns [13]. When integrated with other data sources— molecular, ecological, and behavioral—GMM and PCA create a powerful, integrative approach for detecting evolutionarily significant units and delineating taxonomic boundaries [13].

Theoretical Foundation: From Biological Form to Morphospace

The Procrustes Superimposition Framework

Before PCA can be applied, raw landmark coordinates must be processed to remove non-shape information. This is achieved through Procrustes Superimposition, a method that compares shapes by fitting landmark configurations using optimization criteria [21]. The process consists of three mathematical steps:

• Translation: The centroid (the center of gravity) of each landmark configuration is translated to a common origin (0,0). The centroid coordinates are calculated as the average of the x-coordinates and the average of the y-coordinates across all landmarks for an individual specimen [21].
• Scaling: Configurations are scaled to a standard size, typically to unit centroid size. Centroid size is defined as the square root of the summed squared distances of each landmark from the centroid [21].
• Rotation: The configurations are rotated to minimize the deviation between them and a reference, usually the mean shape [21].

This process results in Procrustes coordinates, which represent shape variables isolated from differences in position, size, and orientation [21]. These coordinates reside in a curved, non-Euclidean space. To apply conventional multivariate statistics like PCA, they are projected onto a linear tangent space, where standard statistical methods can be used to test hypotheses about shape [21].

Mathematical Basis of Principal Component Analysis

PCA is a statistical technique for reducing the dimensionality of complex datasets while preserving maximal variance [21]. It identifies new, orthogonal axes—the principal components (PCs)—that are linear combinations of the original Procrustes-aligned coordinates.

The computation proceeds as follows:

• The mean shape (consensus) is calculated by averaging the x and y coordinates of each landmark across all specimens in the sample [21].
• The covariance matrix of the Procrustes coordinates is computed, describing the variance and covariation among all landmark coordinates [23].
• The eigen decomposition of this covariance matrix yields eigenvalues and eigenvectors. The eigenvectors define the directions of the new PC axes in morphospace, while the eigenvalues represent the variance explained by each corresponding PC [23] [21].

Table 1: Key Outputs of a Principal Component Analysis

Output	Description	Interpretation in Morphometrics
Eigenvalues	A value for each PC axis indicating the variance it accounts for [23].	Higher eigenvalues indicate more important axes of shape variation.
% Variance	The percentage of total shape variance explained by each PC [23].	Determines the relative importance of each PC; typically, the first 5 PCs are the most informative [23].
Cumulative %	The running total of variance explained by successive PCs [23].	Helps assess how much total shape information is captured by the first N PCs.
PC Scores	The coordinates of each specimen along the PC axes [23].	Used to plot specimens in morphospace (e.g., PC1 vs. PC2 scatterplot).
PC Coefficients (Eigenvectors)	The loadings describing how original variables contribute to each PC [23].	Used to visualize the hypothetical shape changes associated with movement along a PC axis.

Practical Protocol: Executing a PCA in Morphometric Software

Software Implementation in MorphoJ

MorphoJ provides a user-friendly platform for conducting PCA on geometric morphometric data. The workflow is as follows [23]:

• Data Preparation: Import the Procrustes-aligned coordinate data, typically from a TPS file or similar format.
• Covariance Matrix Calculation: In the Project Tree window, select the CovMatrix object derived from your dataset.
• Run PCA: Navigate to the Variation menu and select Principal Component Analysis (PCA). MorphoJ will compute the PCA from the covariance matrix [23].
• Interpret Outputs:
  • Graphical Output:
    • PC Shape Changes: Visualize the shape transformation associated with positive and negative extremes of a PC axis using transformation grids, warped outline drawings, or wireframe graphs [23].
    • Eigenvalues: View a scree plot showing the relative percentage of variance for each principal component [23].
    • PC Scores: Generate a scatterplot of specimens in morphospace (e.g., PC1 vs. PC2) to visualize group differences and variation [23].
  • Text Output: Access the Results tab for numerical data, including a table of eigenvalues (with variance percentages) and a table of PC coefficients (eigenvectors) [23].

Case Study: Marmot Mandibles in Taxonomic Research

To illustrate a real-world application, consider a taxonomic study of North American marmot mandibles using a large sample size [13]. After digitizing landmarks on the mandibles, the researcher would:

• Perform a Generalized Procrustes Analysis (GPA) to superimpose all specimens.
• Run a PCA on the Procrustes coordinates to explore the major patterns of mandibular shape variation.
• Visualize the results by coloring the PC scores plot by known species membership.
• Interpret the PC axes by examining the wireframe diagrams that show how landmark configurations change from the negative to the positive end of each axis.

This analysis might reveal that PC1 corresponds to the relative length and robustness of the mandible, effectively separating different marmot species, while PC2 might be associated with the shape of the angular process, potentially revealing differences related to age or population-level variation [13].

Figure 1: A workflow for a geometric morphometric analysis using PCA, from specimen preparation to biological interpretation.

Visualization and Interpretation of Results

Visualizing Shape Changes in Morphospace

The primary visualization tools in a PCA are the scores plot and the shape deformation diagrams.

• The Scores Plot (Morphospace): This scatter plot (e.g., PC1 vs. PC2) displays each specimen as a point. The spatial arrangement of points reveals morphological clusters, which may correspond to taxonomic groups, and gradients, which may represent continuous patterns of allometry or ecophenotypic variation [23]. Outliers can also be identified for further investigation [13].
• Shape Deformation Diagrams: To interpret the morphological meaning of a PC axis, one visualizes the shape change from the negative extreme (e.g., -0.1 units) through the mean shape (0) to the positive extreme (e.g., +0.1 units) [23]. In MorphoJ, this is achieved by right-clicking on the shape visualization and selecting options like Transformation Grid, Warped Outline Drawing, or Wireframe Graph [23]. The wireframe graph is particularly effective as it connects landmarks with lines, making it easier to see the stretching, compression, and twisting of the biological form.

Limitations and Considerations

While PCA is a powerful exploratory tool, taxonomists must be aware of its limitations:

• Lack of Direct Taxonomic Classification: PCA is an unsupervised method that describes variation without a priori group information. It reveals patterns but does not directly assign specimens to taxonomic groups [22]. For classification, supervised methods like Linear Discriminant Analysis (LDA) or Random Forest may be more appropriate [22].
• Statistical vs. Biological Components: PCA components are derived statistically to maximize explained variance, which may not always align with biologically or taxonomically meaningful units [21]. Interpretation requires careful biological knowledge.
• Sensitivity to Outliers and Error: The accuracy of PCA can be compromised by measurement error and outliers. Preliminary analyses, including the assessment of measurement error through replicability studies, are therefore fundamental for robust results [13].

Table 2: Comparison of Multivariate Methods Used in Morphometric Taxonomy

Method	Type	Primary Goal	Key Strength	Key Limitation
Principal Component Analysis (PCA)	Unsupervised	Dimensionality reduction; visualization of major variation [21] [22].	Excellent for exploring continuous shape variation and identifying major trends [21].	Does not use group labels; components may not reflect taxonomic boundaries [22].
Linear Discriminant Analysis (LDA)	Supervised	Classification and group separation [22].	Maximizes separation between pre-defined groups; useful for prediction [22].	Requires a priori groups; prone to overfitting with small sample sizes [22].
Random Forest (RF)	Supervised (Machine Learning)	Classification with complex data [22].	Handles missing data well; high predictive accuracy; no strict data assumptions [22].	"Black box" nature can make interpretation of shape changes less straightforward [22].

Figure 2: Interpreting a morphospace and its corresponding shape changes. The morphospace plot is a visualization of the PC scores, while the linked tables describe the actual morphological transformations captured by each principal component axis.

Table 3: Research Reagent Solutions for Geometric Morphometric Studies

Tool / Resource	Function / Description	Application in Taxonomy
2D Digital Imaging	Capture of specimen images for landmark digitization [13].	Provides low-cost, rapid data acquisition; effective for structures that are largely flat [13].
Landmarking Software (e.g., tpsDig2)	Allows precise digitization of homologous landmarks on specimen images [23].	Creates the primary coordinate data for shape analysis.
Morphometric Software (e.g., MorphoJ, geomorph R package)	Performs Procrustes superimposition, PCA, and other statistical shape analyses [13] [23].	Core platform for processing landmark data and visualizing shape variation.
Comparative Reference Collection	A curated set of verified specimens for training statistical models [22].	Essential for establishing morphological baselines and validating taxonomic identifications.
R Statistical Environment	A programming language with specialized packages (e.g., geomorph, Momocs) for advanced analyses [13].	Offers maximum flexibility and power for custom analyses and automation.

Principal Component Analysis remains a cornerstone of geometric morphometrics, providing an indispensable method for visualizing and interpreting the complex, multivariate nature of biological shape variation. By reducing dimensionality while preserving essential morphological patterns, PCA allows taxonomists to generate hypotheses about group differences, continuous variation, and the morphological facets that contribute most to diversity [21]. However, its application must be guided by rigorous preliminary analyses—including checks for measurement error and outliers—and a clear understanding that its statistically derived components require biological interpretation [13].

Ultimately, PCA is most powerful when used as part of an integrative taxonomic framework [13]. The morphological patterns revealed in the morphospace should not be the sole criterion for taxonomic decisions but rather a key line of evidence to be weighed alongside molecular, ecological, and behavioral data. This multi-pronged approach, facilitated by robust tools like PCA, ensures that our understanding of biodiversity is both quantitatively rigorous and biologically meaningful.

The accurate delineation of species boundaries represents a foundational challenge in biology with far-reaching implications for biodiversity assessment, conservation planning, and public health strategies. Historically, taxonomic classifications have relied predominantly on diagnostic phenotypic characters, an approach that has proven insufficient for detecting cryptic species—genetically distinct lineages that are morphologically indistinguishable through traditional observation [24]. The emergence of integrative taxonomy has revolutionized systematic biology by combining multiple lines of evidence, including molecular data, geometric morphometrics, and ecological analyses, to achieve more robust species delimitation [24]. Within this integrative framework, quantitative shape analysis through geometric morphometrics has emerged as a powerful methodology for capturing and quantifying subtle morphological variation that often eludes conventional descriptive techniques.

The significance of precise species delimitation extends beyond theoretical systematics into applied domains such as epidemiology and vector control. For instance, in the case of Chagas disease vectors like Triatoma species, failure to distinguish between cryptic species can severely compromise disease management efforts, as different vector species may exhibit varying ecological preferences, behavioral patterns, and vectorial capacities [24]. Geometric morphometrics provides the methodological rigor necessary to extract complex shape data that can be statistically linked to genetic divergences and ecological gradients, thereby offering insights into evolutionary processes such as adaptive radiation, character displacement, and niche specialization.

Geometric Morphometrics: A Methodological Framework

Theoretical Foundations and Data Acquisition

Geometric morphometrics (GM) represents a paradigm shift from traditional measurement-based approaches by preserving the complete geometry of morphological structures throughout statistical analysis. Unlike classical morphometrics, which relies on linear distances, ratios, or angles, GM utilizes landmarks and semi-landmarks:

• Type I landmarks are defined by discrete anatomical loci (e.g., suture intersections, muscle attachment points)
• Type II landmarks represent points of maximum curvature or other local geometric features
• Type III landmarks are defined as extreme points (e.g., tips of spines or processes)
• Semi-landmarks capture the outline of structures without discrete landmarks, allowing for the quantification of curves and surfaces [24]

The acquisition of shape data begins with the generation of high-quality images of taxonomically informative structures, such as insect heads, pronota, wings, or genitalia. Specimens should be positioned in a standardized orientation to minimize measurement error, with careful attention to lighting conditions, scale calibration, and resolution optimization. For each specimen, the two-dimensional or three-dimensional coordinates of predefined landmarks are digitized using specialized software, creating a comprehensive dataset of geometric information that forms the basis for subsequent statistical analyses.

Analytical Workflow and Statistical Procedures

The analytical pipeline in geometric morphometrics involves a sequence of statistical procedures designed to isolate shape variation from other sources of morphological difference:

• Generalized Procrustes Analysis (GPA): This procedure removes the effects of position, orientation, and scale by superimposing landmark configurations through translation, rotation, and scaling, effectively isolating pure shape variables for subsequent analysis [24].

• Procrustes ANOVA: A specialized variance partitioning method that tests for significant differences in mean shape among predefined groups (e.g., haplogroups, species, populations) while accounting for measurement error and directional asymmetry [24].

• Discriminant Function Analysis: A multivariate technique that maximizes separation among groups and assesses the classificatory power of shape variables, often expressed as percentages of correct assignment [24].

• Thin-plate spline visualization: A deformation-based method that produces graphical representations of shape differences between groups, allowing for intuitive interpretation of morphological variation [24].

The following diagram illustrates the complete experimental workflow for geometric morphometrics in species delimitation studies:

Figure 1: Experimental workflow for geometric morphometrics in species delimitation studies, showing the sequence from specimen collection through data acquisition, statistical analysis, and integration with complementary data types.

Case Study: Triatoma pallidipennis Haplogroups

Experimental Design and Morphometric Analysis

A recent investigation of Triatoma pallidipennis, a Chagas disease vector in Mexico, exemplifies the power of geometric morphometrics for discriminating cryptic species [24]. Researchers analyzed four haplogroups previously identified through molecular phylogenetics, focusing on two morphological structures: the head and pronotum. The experimental protocol involved:

• Specimen sourcing: Haplogroup I (Morelos, Oaxaca, eastern Puebla), Haplogroup II (southern Morelos, southwestern Mexico State, eastern Guerrero), Haplogroup III (Mexico State), and Haplogroup V (Colima, Jalisco) [24]
• Image acquisition: Standardized digital photographs of heads and pronota
• Landmark configuration: 23 landmarks for head structures, 18 landmarks for pronotal structures
• Statistical framework: Procrustes ANOVA with pairwise comparisons, discriminant function analysis

The analysis revealed significant differences in head shape among almost all haplogroups, with deformation grids showing anterior displacement of the antenniferous tubercle and posterior displacement of pre-ocular landmarks as the most distinctive shape variables [24]. In contrast, pronotum shape showed less discriminatory power, with pairwise comparisons revealing significant differences among only three haplogroups, suggesting that cephalic morphology possesses higher taxonomic value for differentiating these putative cryptic species.

Integration with Ecological Niche Modeling

To strengthen the species delimitation framework, the morphometric analysis was integrated with ecological niche modeling (ENM) using the MaxEnt algorithm [24]. Occurrence records for each haplogroup were combined with bioclimatic variables to characterize environmental niches and predict potential distribution areas. The ENM results demonstrated:

• Niche divergence: Ecological niche models for each haplogroup failed to predict the climatic suitability areas of other haplogroups
• Environmental segregation: Significant differences in environmental preferences among haplogroups, particularly along temperature and precipitation gradients
• Evolutionary implications: The combined evidence supports ecological speciation as a potential mechanism driving divergence in this complex

The following table summarizes the key quantitative findings from the Triatoma pallidipennis study:

Table 1: Summary of morphometric and ecological findings for Triatoma pallidipennis haplogroups [24]

Haplogroup	Geographic Distribution	Head Shape Differentiation	Pronotum Shape Differentiation	Ecological Niche Distinctness
I	Morelos, Oaxaca, eastern Puebla	Significant differences from II, III, V	Significant difference from V only	Unique combination of bioclimatic variables
II	Southern Morelos, southwestern Mexico State, eastern Guerrero	Significant differences from I, III	No significant differences from other groups	Predicts unique suitability areas
III	Mexico State	Significant differences from I, II, V	Significant differences from I, V	Non-overlapping environmental space
V	Colima, Jalisco	Significant differences from I, III	Significant differences from I, III	Distinct precipitation requirements

The Researcher's Toolkit: Essential Methodologies and Reagents

Successful implementation of geometric morphometrics in species delimitation requires specialized materials, software tools, and analytical frameworks. The following table details essential components of the research pipeline:

Table 2: Essential research reagents, tools, and methodologies for geometric morphometrics in taxonomy

Category	Specific Tool/Method	Application/Function	Technical Specifications
Imaging Equipment	Stereo microscope with digital camera	High-resolution image acquisition of morphological structures	Minimum 5MP resolution, standardized magnification
Landmarking Software	tpsDig2	Precise digitization of landmark coordinates	Supports Type I, II, III landmarks and semi-landmarks
Morphometric Analysis	MorphoJ	Comprehensive geometric morphometrics analysis	Implements GPA, Procrustes ANOVA, regression, discrimination
Statistical Framework	Procrustes ANOVA	Hypothesis testing for shape differences	Partitioning of variance components with permutation tests
Visualization Methods	Thin-plate spline	Graphical representation of shape changes	Vector deformation grids based on bending energy
Ecological Modeling	MaxEnt	Predictive species distribution modeling	Uses presence-only data with environmental layers
Molecular Integration	Mitochondrial gene sequencing (e.g., nad4)	Independent genetic evidence for species boundaries	Provides phylogenetic framework for morphometric comparisons

Interpretation and Evolutionary Implications

The relationship between morphological disparity, genetic divergence, and ecological specialization provides critical insights into evolutionary processes. The following diagram illustrates the conceptual framework linking these elements in the context of species delimitation:

Figure 2: Conceptual framework illustrating the relationships between genetic divergence, morphological variation, and ecological niche differentiation in species delimitation.

The case study of Triatoma pallidipennis demonstrates how geometric morphometrics can reveal patterns of morphological variation that align with genetic haplogroups and ecological differentiation [24]. The unequal distribution of taxonomically informative variation across morphological structures (e.g., head vs. pronotum) highlights the importance of functional integration and module-specific evolution in shaping organismal form. Structures under stronger functional constraints may exhibit less variation across recently diverged lineages, while those involved in ecological interactions may show more rapid differentiation in response to selective pressures.

From an evolutionary perspective, the concordance between morphometric disparity, genetic distance, and niche divergence provides compelling evidence for ecological speciation in the T. pallidipennis complex [24]. The fact that different haplogroups occupy distinct environmental spaces with limited niche overlap suggests that adaptation to local ecological conditions has driven morphological divergence, particularly in cephalic structures that may be linked to feeding efficiency, host preference, or other ecologically relevant functions.

Geometric morphometrics has transformed the role of morphological data in species delimitation by providing rigorous quantitative frameworks for analyzing shape variation in an evolutionary context. When integrated with molecular phylogenetics and ecological niche modeling, morphometric approaches can effectively discriminate cryptic species, reveal patterns of adaptive divergence, and provide insights into speciation mechanisms. The Triatoma pallidipennis case study exemplifies this integrative approach, demonstrating how shape analysis of taxonomically informative structures can corroborate genetic evidence and illuminate the ecological dimensions of evolutionary divergence.

Future advances in geometric morphometrics will likely focus on three-dimensional imaging techniques, automated landmark placement through machine learning algorithms, and more sophisticated models of morphological integration and modularity. As these methodological innovations mature, geometric morphometrics will continue to strengthen its position as an essential component of integrative taxonomy, providing critical evidence for species boundaries while illuminating the evolutionary processes that generate and maintain biological diversity.

A Step-by-Step GM Workflow: From Specimen to Statistical Analysis

In taxonomic research utilizing geometric morphometrics (GMM), the integrity of the entire analytical pipeline is contingent upon the initial stages of sample preparation and image acquisition [13]. Proper execution of this first stage is a critical prerequisite for generating high-fidelity shape data that can reliably capture phenotypic variation among species and populations [25]. This guide details the established best practices for preparing biological specimens and acquiring their images for subsequent landmark-based analysis, a foundational step for studies ranging from mammalian crania to minute insect taxa [13] [4]. The protocols outlined herein are designed to minimize measurement error, control for extraneous sources of variance, and ensure the resulting data are robust for testing taxonomic hypotheses [13].

Specimen Selection and Preparation

The process begins with the careful selection and preparation of specimens to ensure that the observed morphological variation reflects genuine biological differences rather than preparation artifacts or ontogenetic stage.

Table 1: Specimen Preparation Guidelines for Common Taxonomic Groups

Taxonomic Group	Preparation Concern	Recommended Action	Rationale
Small Insects (e.g., Thrips) [4]	Positioning for consistent view	Slide-mounting	Facilitates a perfectly lateral or dorsal view, standardizing orientation for landmarking.
Vertebrate Skulls (e.g., Marmots) [13]	Asymmetry and missing data	Assess for damage & completeness; use paired landmarks if possible [13].	Incomplete specimens can bias analyses; symmetry can be leveraged to increase landmark count.
Bone Elements [25]	Surface debris and texture	Gentle cleaning; avoid reflective coatings.	Ensures clear visualization of anatomical structures without obscuring morphology.
Live Animal Faces (e.g., Cats) [26]	Postural effects and movement	Standardize camera angle & use high-speed shutter.	Controls for non-rigid shape changes induced by head orientation relative to the camera.

The overarching goal of specimen preparation is to standardize posture and orientation to the greatest extent possible. For durable specimens like bones and slide-mounted insects, this involves physical manipulation and mounting [13] [4]. For live animals or soft tissues, standardization is achieved through controlled imaging conditions [26].

Image Acquisition Protocols

High-resolution image acquisition is the cornerstone of generating reliable 2D geometric morphometric data. The following protocol provides a general framework that can be adapted for specific research contexts.

Figure 1: A generalized workflow for high-resolution image acquisition in geometric morphometric studies.

Equipment and Configuration

A consistent and well-documented imaging setup is non-negotiable for producing comparable data across a sampling session. Key considerations include:

• Camera and Lens: Use a high-resolution digital single-lens reflex (DSLR) or mirrorless camera with a macro lens capable of capturing fine morphological detail [4]. The lens should be fixed to avoid changes in focal length.
• Stabilization: Mount the camera on a stable tripod and use a remote shutter release or timer to prevent motion blur.
• Lighting: Implement uniform, diffuse lighting to eliminate shadows and specular highlights that can obscure morphological features [26]. Ring lights or a dual-light setup from fixed angles are optimal.
• Scale and Orientation: Include a micrometer or scale bar of known dimensions within every image frame. For 3D structures, a fiduciary marker or an object of known shape can aid in assessing orientation and potential distortion [13].

Table 2: Camera Configuration for High-Resolution Morphometric Imaging

Parameter	Setting	Justification
Aperture (f-stop)	f/8 - f/11	Balances sufficient depth of field with image sharpness.
ISO	Lowest native setting (e.g., 100)	Minimizes digital noise in the image.
Shutter Speed	As required for exposure	Fast enough to prevent motion blur; use a tripod.
File Format	RAW + JPEG	RAW retains maximum data for processing; JPEG for quick review.
White Balance	Manual or custom setting	Prevents inconsistent color casts between images.
Focus	Manual	Ensures consistency and prevents autofocus from shifting between shots.

Specimen-Specific Imaging Workflows

Imaging of Slide-Mounted Insects: As demonstrated in studies of Thrips taxonomy, specimens are slide-mounted to achieve a consistent 2D orientation [4]. High-resolution digital images are then captured using a microscope-equipped camera system. Post-capture image enhancement (e.g., increasing contrast and sharpening) may be performed uniformly across all images to improve landmark visibility [4].

Imaging of Live Subjects: For studies of facial expression in non-human animals, standardizing the position of the subject relative to the camera is critical [26]. This involves controlling the camera angle and distance, and capturing images when the subject is in a neutral, reproducible posture to minimize the confounding effects of head orientation on 2D shape.

Quality Control and Metadata Management

Following image acquisition, a rigorous quality control process is essential before proceeding to landmark digitization.

• Check for Artifacts: Inspect each image for blur, glare, poor focus, or obstructions of key anatomical structures.
• Verify Standardization: Ensure consistency in scale, orientation, and lighting across all images in the dataset.
• Assess Repeatability: For a subset of specimens, acquire multiple images to later quantify measurement error associated with the imaging process itself [13].

The Scientist's Toolkit: Essential Research Reagents and Materials

Item	Function in Sample Prep & Imaging
Slide-Mounting Media	Secures and preserves small specimens (e.g., insects) in a standardized orientation for imaging [4].
Macro Lens	Allows for high-resolution, close-up photography of small biological structures.
Calibrated Scale Bar/Micrometer	Provides a spatial reference within the image, allowing for size calibration and ensuring scale consistency.
Tripod & Remote Shutter	Eliminates camera shake, ensuring image sharpness and consistency across the image set.
Diffuse Lighting Setup	Provides even, shadow-free illumination to reveal true morphological form without obscuring highlights [26].
Specimen Positioning Stage	Allows for precise and repeatable rotation and translation of the specimen in front of the camera.

Comprehensive metadata must be recorded for every image, including specimen identifier, date of acquisition, camera settings (aperture, ISO, shutter speed), lens used, lighting setup, and scale. This documentation is critical for replicability and for troubleshooting should inconsistencies in the data be discovered later.

Meticulous sample preparation and high-resolution image acquisition form the bedrock of any rigorous geometric morphometric study in taxonomy. By standardizing specimens, controlling imaging conditions, and implementing thorough quality control, researchers can generate high-quality shape data that accurately represents underlying biological variation. This careful attention to initial stages mitigates the introduction of spurious variance and measurement error, thereby ensuring the validity and reliability of all subsequent statistical comparisons and taxonomic inferences [13].

In taxonomic research, the accurate quantification of morphological shape is paramount for distinguishing between species, understanding evolutionary relationships, and identifying cryptic diversity. Geometric morphometrics (GMM) has emerged as a primary method for assessing essential morphological variables because it provides a quantitative and unbiased approach to morphological comparison [27]. The intricate relationship between morphogenetic and evolutionary factors underscores the need for such multivariate methods in biological and ecological research [27]. Within this framework, landmarking forms the very foundation of shape analysis; the selection of homologous points and curves is a critical step that directly influences the validity, reliability, and biological interpretability of all subsequent analyses. This guide provides an in-depth examination of landmarking strategies, focusing on their application within taxonomy. It details the typology of landmarks, practical protocols for their digitization, and the analytical workflows that transform raw coordinate data into robust taxonomic insights. Proper landmarking is not merely a technical procedure but a hypothesis-driven exercise in defining biological homology across specimens, making it a cornerstone of best practices in modern taxonomic research using GMM.

Landmark Typologies: Defining Biological and Mathematical Homology

Landmarks are discrete, homologous points that can be precisely located and reliably measured across all specimens in a study. The operational classification of landmarks is fundamental, as it determines the type of shape information captured and its biological relevance. While a traditional three-type system exists, a more nuanced six-type classification is often utilized in applied studies to better reflect the different operational origins of points situated on curves [27]. For the purposes of most taxonomic work, the following three core types are most relevant.

Table 1: Core Types of Landmarks in Geometric Morphometrics

Landmark Type	Definition	Basis for Homology	Examples	Reliability in Taxonomy
Type I (Anatomical)	Points of clear biological or anatomical significance, corresponding to specific, discrete features [27].	Ontogenetic and evolutionary homology.	The junction between bones or sutures, the tip of the nose, the corner of the eye [27].	High; considered the most reliable due to clear homology across specimens.
Type II (Mathematical)	Points defined by local geometric properties, such as maxima or minima of curvature [27].	Local geometry of the form.	The point of maximum curvature along a bone, the deepest point in a notch [27].	Moderate; useful for capturing shape information where anatomical landmarks are sparse.
Type III (Constructed)	Points defined by their relative position or constructed based on other landmarks [27].	Geometric relationship to other landmarks.	The midpoint between two Type I landmarks, extreme points at the ends of structures [27].	Lower; most susceptible to error but necessary for outlining complex shapes.

The process of landmarking and classifying landmarks relies heavily on biological interpretation, and a significant limitation of these techniques is that the labeling and analysis processes are often semi-manual or manual [27]. Despite this, landmarking analysis remains the primary technique in GMM. For taxonomic studies, a strategy that prioritizes Type I landmarks, uses Type II landmarks to supplement shape description, and employs Type III landmarks sparingly to capture overall geometry is considered a best practice. This ensures that the resulting shape variables are grounded in biological homology, which is essential for meaningful evolutionary and taxonomic inference.

Beyond Points: Capturing Curves with Semi-Landmarks

Many biologically significant morphological structures, such as mandible outlines, feather shapes, or leaf margins, are better defined by curves than by discrete points. To quantitatively analyze these structures, GMM employs semi-landmarks—points placed at defined intervals along curves and between two fixed landmarks [10]. Semi-landmarks are considered "deficient" in the sense that their initial placement is not based on ontogenetically conserved features; instead, their homology is established during the analysis through a process of "sliding" that minimizes their bending energy or procrustes distance relative to a mean shape [10]. This allows for the capture of continuous shape variation along contours that lack sufficient Type I landmarks.

The use of semi-landmarks is particularly powerful in taxonomy for differentiating groups based on subtle outline differences. For instance, studies have successfully used semi-landmarks on head and pronotum outlines to compare haplogroups of Triatoma pallidipennis, a Chagas disease vector, revealing differences that supported the delimitation of cryptic species [28]. Similarly, methodological comparisons have shown that semi-landmark-based methods (such as bending energy alignment and perpendicular projection) perform as well as other outline analysis methods like Elliptical Fourier Analysis in classifying specimens by age based on feather shape [29].

Experimental Workflow: From Image Acquisition to Shape Data

A standardized workflow is crucial for generating high-quality, reproducible landmark data. The following protocol, adaptable for most taxonomic studies involving 2D specimens, is summarized in the diagram below.

Figure 1: A generalized workflow for landmark and semi-landmark digitization and processing in taxonomic geometric morphometrics.

Detailed Protocol Steps

• Image Acquisition & Standardization: Specimens should be photographed on a solid, contrasting background. The camera must be fixed in position with its lens perpendicular to the plane of the specimen to avoid perspective distortion. For fish or similar specimens, the body axis should be positioned horizontally [27]. Image resolution should be sufficient to clearly identify all anatomical landmarks.
• Background Removal & Pre-processing: Use image editing software or AI-based tools to remove the background, isolating the specimen. This simplifies subsequent digitization. Ensure images are scaled appropriately if absolute size is a variable of interest.
• Landmark Digitization: Using software such as tpsDig2 [27], digitize all Type I landmarks first, as these are the most reliable. Follow with Type II and Type III landmarks. It is critical to maintain the same order of digitization for all specimens.
• Curve Definition and Semi-Landmark Placement: In your digitizing software, define curves between fixed landmarks. Subsequently, place a sufficient number of semi-landmarks along these curves to capture their geometry adequately. The number of points can be optimized, but studies suggest classification rates are not highly dependent on the exact number used [29].
• Sliding Semi-Landmarks: After initial digitization, semi-landmarks must be "slid" to remove the arbitrariness of their initial placement. This is typically done by iteratively moving the points along the tangent direction of the curve to minimize either the bending energy of the thin-plate spline or the Procrustes distance between each specimen and a reference (usually the sample mean) [29]. This step is automatically performed by software like tpsRelw [27] or R packages such as geomorph [13].
• Generalized Procrustes Analysis (GPA): This is the foundational step for all subsequent shape analysis [27]. GPA superimposes all landmark configurations by translating them to a common origin, scaling them to unit Centroid Size, and rotating them to minimize the sum of squared distances between corresponding landmarks. The output is a set of Procrustes shape coordinates, which represent the pure shape of each specimen, free from the confounding effects of size, position, and orientation [27] [10].

The Scientist's Toolkit: Essential Software and Reagents

Successful landmark-based analysis requires a suite of specialized software tools. The following table details key solutions used in the field.

Table 2: Essential Research Software Tools for Landmark-Based Geometric Morphometrics

Tool Name	Type/Category	Primary Function in Landmarking	Key Feature for Taxonomy
tpsDig2 [27]	Standalone Application	Digitizing landmarks, curves, and semi-landmarks from 2D images.	The industry standard for manual digitization; provides precise control over point placement.
tpsUtil [27]	Standalone Application	Managing and creating data files for the TPS series.	Used to build the master TPS file that links all images and their landmark data.
tpsRelw [27]	Standalone Application	Performing relative warps analysis and sliding semi-landmarks.	Critical for the sliding semi-landmark step prior to Procrustes superimposition.
MorphoJ [27]	Standalone Application	Performing Procrustes superimposition, statistical analysis, and visualization.	User-friendly GUI for a wide range of multivariate analyses like PCA, CVA, and discriminant analysis.
R Package: geomorph [13]	Programming Library	Comprehensive GMM analysis within the R environment.	Enables reproducible analysis pipelines, advanced statistical modeling, and high-quality graphing.
R Package: Momocs [27]	Programming Library	Outline acquisition, manipulation, and analysis.	Specialized for outline and Fourier analyses, complementing landmark-based approaches.

Analytical Pathways: From Shape Coordinates to Taxonomic Discrimination

Once Procrustes shape coordinates are obtained, a suite of multivariate statistical techniques can be applied to test taxonomic hypotheses. The analytical pathway, from raw shape data to taxonomic interpretation, is visualized below.

Figure 2: Core analytical pathways for interpreting landmark-based shape data in taxonomic research.

• Principal Component Analysis (PCA): Used to identify the major independent axes of shape variation within the entire dataset without a priori group information. It is excellent for exploring data structure, detecting outliers, and visualizing the primary patterns of morphological disparity [27].
• Canonical Variate Analysis (CVA): A multiple-group form of discriminant analysis used when specimens are assigned to pre-defined groups (e.g., species or populations). CVA finds axes (canonical variates) that maximize the between-group variance relative to the within-group variance, providing the best possible visual separation among groups [29]. The statistical significance of group differences in mean shape can be tested using Procrustes ANOVA [28].
• Discriminant Function Analysis (DFA): Often used in conjunction with CVA, DFA constructs functions to classify specimens into groups based on their shape. The cross-validation rate of correct assignment, where specimens are classified based on functions derived from all other specimens, provides a less biased estimate of the distinctness of the groups [29]. This is a key metric in taxonomy for assessing the strength of morphological separation.
• Thin-Plate Spline (TPS) Visualization: This technique provides a powerful visual representation of shape change. The TPS deformation grid graphically warps a shape from one state (e.g., the mean shape of one species) to another (e.g., the mean shape of another species), allowing for intuitive biological interpretation of the localized shape differences captured by the landmarks [27].

The strategic selection of homologous points and curves is the critical bridge between raw morphological form and quantitative shape data in taxonomy. A rigorous approach that combines a deep understanding of landmark typologies with a standardized digitization protocol and appropriate statistical analysis is fundamental to generating robust, reproducible, and biologically meaningful results. As geometric morphometrics continues to evolve, with advancements in software and methodology, the technical limitations associated with morphological analysis are expected to decrease [27]. However, the intellectual rigor applied during the landmarking stage will remain irreplaceable. By adhering to these best practices, taxonomists can leverage the full power of GMM to delimit species, uncover cryptic diversity, and elucidate the evolutionary processes that have shaped the biodiversity we see today.

In taxonomic research, the precise quantification of morphological shape is indispensable for distinguishing between species, understanding evolutionary relationships, and defining taxonomic groups. Geometric Morphometrics (GM) provides a powerful statistical framework for analyzing the geometry of biological forms. This guide details the core pre-processing stage of a GM analysis: the procedures of Generalized Procrustes Analysis (GPA) and the subsequent creation of shape variables. These steps are critical as they transform raw landmark coordinates into a set of variables that purely represent shape, free from the confounding effects of position, scale, and orientation [30]. The rigor applied in this stage directly impacts the validity of all subsequent statistical analyses and taxonomic conclusions.

Theoretical Foundation of Shape and Procrustes Superimposition

The Concept of Shape in Morphometrics

In geometric morphometrics, shape is formally defined as all the geometric information that remains when the effects of location, scale, and rotation are removed from an object [30]. An object's shape is represented by the configuration of landmarks—discrete, anatomically homologous points that can be precisely located across all specimens in a study [30] [11].

Landmarks are typically categorized into three types:

• Type I: Landmarks defined by the intersection of distinct biological structures (e.g., junction of three bones).
• Type II: Landmarks defined by a local property, such as a point of maximum curvature.
• Type III: Landmarks defined by extremal points or constructed points (e.g., the midpoint of a structure).

The raw data for a GM analysis is a set of landmark configurations, where each configuration consists of the (x, y) or (x, y, z) coordinates of k landmarks for a single specimen.

The Principle of Generalized Procrustes Analysis (GPA)

Generalized Procrustes Analysis is the statistical procedure used to remove the non-shape information from landmark data. It achieves this by superimposing all landmark configurations onto a common coordinate system through an iterative process that minimizes the sum of squared distances between corresponding landmarks across all specimens [30]. This process is often referred to as Procrustes superimposition.

The core steps of GPA are:

• Translation: Each configuration is centered at its centroid (the mean of the x and y coordinates for that configuration), moving all specimens to a common origin.
• Scaling: Each configuration is scaled to unit Centroid Size, a measure of size calculated as the square root of the sum of squared distances of all landmarks from their centroid [30].
• Rotation: Each configuration is rotated to align with a reference configuration (often the mean shape) to minimize the Procrustes distance.

The outcome of GPA is a set of Procrustes-aligned coordinates. The residuals from the superimposition, known as Procrustes residuals, form the basis of the shape variables used in multivariate analysis [25] [30].

Methodological Protocol for GPA and Shape Variable Creation

The following diagram illustrates the complete workflow from raw landmark data to the creation of shape variables, detailing the key stages of Generalized Procrustes Analysis.

Detailed Step-by-Step Procedure

Step 1: Translation

• Objective: Remove differences in location.
• Action: For each specimen's landmark configuration, calculate its centroid. The centroid is the average position of all its landmarks, found by centroid_x = mean(x_coordinates) and centroid_y = mean(y_coordinates). Subtract the centroid coordinates from each landmark's coordinates, effectively moving the entire configuration so that its centroid is at the origin (0,0).

Step 2: Scaling

• Objective: Remove differences in size.
• Action: Calculate the Centroid Size (CS) for each translated configuration. CS is defined as the square root of the sum of squared distances from each landmark to the configuration's centroid: CS = sqrt( Σ (x_i - centroid_x)² + (y_i - centroid_y)² ). Divide the coordinates of each landmark by its configuration's Centroid Size. This scales all specimens to a uniform size.

Step 3: Rotation

• Objective: Remove differences in orientation.
• Action: A target (reference) configuration is selected. For the first iteration, this can be the first specimen or an arbitrarily chosen one. Each specimen is then rotated around its centroid to minimize the Procrustes distance—the square root of the sum of squared differences between corresponding landmarks—between the specimen and the target.

Step 4 and 5: Iteration and Consensus

• Objective: Achieve an optimal overall fit for all specimens.
• Action: After all specimens have been aligned to the initial reference, a new mean shape (consensus configuration) is calculated from all the aligned specimens. This consensus becomes the new reference, and the rotation step (Step 3) is repeated. This iterative process continues until the change in the sum of Procrustes distances between successive iterations falls below a predefined threshold, indicating convergence [30].

Outputs and Shape Variable Creation

The primary outputs of GPA are:

• Procrustes-Aligned Coordinates: The coordinates of all landmarks for all specimens in a common shape space.
• Consensus Configuration: The average landmark configuration representing the mean shape of the sample.
• Procrustes Residuals: The differences between each specimen's aligned coordinates and the consensus. These residuals are the shape variables that encapsulate pure shape variation.

Because the aligned coordinates lie on a curved multidimensional space (a hypersphere), they are projected onto a linear tangent space for subsequent statistical analysis, such as Principal Component Analysis (PCA) [30]. The Procrustes coordinates in this tangent space serve as the shape variables for exploring patterns of morphological variation in taxonomy.

The Scientist's Toolkit: Essential Research Reagents and Materials

The following table catalogs the key software and data components required for executing a GPA and shape variable analysis.

Table 1: Essential Research Reagents and Materials for Geometric Morphometric Pre-processing

Item Name	Type/Format	Primary Function in GPA & Shape Analysis
Landmark Data (TPS File)	Digital data file (e.g., .TPS, .NTS)	Standard format for storing 2D or 3D landmark coordinates collected from multiple specimens; serves as the primary input for analysis [31].
R Statistical Environment	Software platform	A free, open-source computing environment for statistical analysis and graphics, which is widely used for geometric morphometrics [31].
geomorph R Package	R software library	A comprehensive R package that provides functions for every step of a GM analysis, including GPA (gpagen), statistical testing, and visualization [31].
MorphoJ	Standalone software	A user-friendly, cross-platform program dedicated to GM, offering a graphical interface for performing GPA, PCA, and other multivariate analyses [31].
PAST	Standalone software	A free software package for paleontological and general statistical analysis, which includes a suite of tools for geometric morphometrics [31].
TPS Dig2	Standalone software	A widely used program for the manual digitization of landmarks from 2D digital images [30].

Analysis of Symmetry and Asymmetry

In taxonomic studies of structures with symmetric organization, such as bilaterally symmetric leaves or flowers, the GPA framework can be extended to decompose total shape variation into symmetric and asymmetric components [30]. This is a critical step, as conflating the two can obscure true taxonomic signals.

• Symmetric Component (Among-Individual Variation): Represents the shape differences between the average shapes of different specimens. This is typically the variation of primary interest for taxonomy.
• Asymmetric Component (Within-Individual Variation): Represents the deviations from perfect symmetry within a single individual. This can be further divided into directional asymmetry (a consistent bias towards one side) and fluctuating asymmetry (small, random deviations), the latter often being used as a measure of developmental stability [30].

The analysis involves digitizing landmarks on both sides of the symmetric structure and using specialized GPA protocols that model the object's symmetry. Principal Component Analysis can then be applied separately to the symmetric and asymmetric components to visualize and quantify their respective patterns [30].

Advanced Considerations and Automation

The Challenge of Landmark Detection

Traditional manual landmarking is time-consuming, subject to observer bias, and limits the number of landmarks and specimens that can be practically analyzed [25] [11]. This has driven research into automated landmark detection systems. Machine learning approaches, such as those based on Random Forests, have shown promise by using multi-resolution image features to train models that predict landmark positions in new images, significantly speeding up data acquisition [11].

Landmark-Free and Fully Automated Phenotyping

For complex morphological structures where homology is difficult to establish, or for analyzing entire surfaces without a priori assumptions, novel "landmark-free" methods are emerging. One such approach is morphVQ (Morphological Variation Quantifier), which uses descriptor learning and functional maps to establish correspondence between entire 3D surface models of biological specimens [25]. This method quantifies shape variation using Latent Shape Space Differences (LSSDs), providing a comprehensive and automated alternative to traditional landmark-based GM that can capture more subtle morphological details [25].

In the domain of modern taxonomy, Geometric Morphometrics (GMM) has established itself as an indispensable methodology for quantifying and analyzing biological form. This in-depth technical guide focuses on the crucial stage of multivariate statistical analysis, which enables researchers to extract meaningful information from shape data. Within the framework of a broader thesis on best practices for GMM in taxonomy, this section details the application of Principal Component Analysis (PCA), Canonical Variate Analysis (CVA), and Discriminant Function Analysis for distinguishing between taxa, elucidating phylogenetic relationships, and understanding ecological adaptations [32] [13]. These methods transform raw landmark coordinates into powerful statistical evidence for taxonomic decisions, moving beyond qualitative descriptions to robust, quantitative hypothesis testing.

Theoretical Foundations and Workflow

Multivariate analyses in geometric morphometrics operate on shape variables derived from a Generalized Procrustes Analysis (GPA). GPA superimposes landmark configurations by optimizing their position through the sequential removal of non-shape information related to location, scale, and orientation [30]. The resulting Procrustes coordinates reside in a curved shape space, which is linearized via projection onto a tangent space. This tangent space coordinates are the data upon which conventional multivariate statistical procedures are applied. The core objective is to reduce the high dimensionality of the shape data (multiple landmark coordinates) and to test for significant group differences in a morphospace.

The following diagram illustrates the standard analytical workflow from raw images to statistical interpretation, highlighting the role of PCA, CVA, and DFA.

Core Analytical Methods

Principal Component Analysis (PCA)

Function: PCA is an unsupervised exploratory technique used to visualize the major patterns of shape variation within the entire dataset without prior group classification. It identifies the primary axes of variation (Principal Components) that account for the greatest proportions of total shape variance.

Protocol for Taxonomic Application:

• Input Data: The analysis uses the covariance matrix of the Procrustes-aligned coordinates from the tangent space.
• Eigenanalysis: The covariance matrix is decomposed into its eigenvalues and corresponding eigenvectors.
• Output Interpretation:
  • Principal Components (PCs): Each PC is an eigenvector representing a specific pattern of landmark covariation. Shape change along a PC is visualized by warping the consensus (mean) shape configuration along the positive and negative directions of that axis.
  • Percent Variance: The eigenvalue for each PC indicates the amount of total shape variance it explains. The first PC (PC1) explains the maximum variance, followed by PC2, and so on [30].
  • Scatterplots: Specimens are plotted in a morphospace defined by the first few PCs, allowing for the visual assessment of group separation, outliers, and continuous trends.

Taxonomic Context: PCA is fundamental for initial data exploration, assessing the existence of natural groupings, and identifying major morphological trends that may correspond to taxonomic divisions or allometric patterns [13].

Canonical Variate Analysis (CVA)

Function: CVA is a supervised technique that maximizes the separation among pre-defined groups (e.g., species, populations) relative to the variation within them. It finds linear combinations of the original variables (canonical variates) that best discriminate among the known groups.

Protocol for Taxonomic Application:

• Input Data: Procrustes shape variables and a grouping variable (e.g., species assignment).
• Calculation: CVA derives canonical axes that maximize the ratio of between-group to within-group variance.
• Output Interpretation:
  • Canonical Variates (CVs): Specimens are scored on each CV, and these scores are plotted to visualize group discrimination.
  • Mahalanobis Distances: The squared distance between group centroids in the canonical space, which can be used to test for the significance of group differences, often via permutation tests [32].
  • Classification: A leave-one-out cross-validation is typically performed to estimate the misclassification error rate, providing a realistic measure of the power of shape for distinguishing the groups.

Taxonomic Context: CVA is a powerful tool for hypothesis testing, specifically for validating the distinctiveness of described species or populations. It is extensively used in taxonomic revisions to quantify and test morphological differences between putative taxa [32] [13].

Discriminant Function Analysis (DFA)

Function: DFA (or Linear Discriminant Analysis, LDA) is closely related to CVA and is used to assign unknown specimens to pre-defined groups. It creates functions based on linear combinations of variables that best separate the groups and provides a classification rule.

Protocol for Taxonomic Application:

• Model Training: The discriminant functions are computed from a training dataset with known group membership.
• Classification: The derived functions are applied to assign new, unknown specimens to one of the pre-defined groups.
• Output Interpretation:
  • Classification Matrix: A table comparing original group membership versus predicted group membership.
  • Posterior Probabilities: The probability of a specimen belonging to each group, with the specimen assigned to the group with the highest probability.

Taxonomic Context: DFA is the method of choice for developing diagnostic keys and for the practical identification of specimens in ecological, archaeological, or forensic contexts [32] [33]. It operationalizes the findings of a morphometric study for applied use.

Case Study in Taxonomy: Sinibotia Fish Species

A study on Sinibotia fish species provides a clear example of the integrated application of these methods in a taxonomic context. The research aimed to clarify species boundaries within this genus, which is characterized by high morphological similarity and close phylogenetic relationships [32].

Table 1: Summary of Morphometric Analysis of Sinibotia Species

Species Analyzed	Sampling Location	Key Morphological Traits for Discrimination	Major Findings
S. superciliaris	Tuo River, Zizhong County	Snout length, nasal snout distance, head depth, body depth, caudal fin length, dorsal fin length	MM and GM yielded highly consistent results. MM quantified linear size differences effectively, while GM better captured and visualized complex overall shape variations.
S. reevesae	Tuo River, Zizhong County
S. robusta	Li River, Pingle County
S. pulchra	Li River, Pingle County
S. zebra	Lipu River, Pingle County

The study successfully used CVA and Discriminant Function Analysis to differentiate the species, with morphological variations primarily reflected in snout length, nasal snout distance, head depth, body depth, caudal fin length, and dorsal fin length [32]. The combined evidence from MM and GM was concluded to significantly contribute to species identification, understanding of phylogenetic relationships, and ecological adaptations.

The Scientist's Toolkit: Essential Software and Reagents

Successful multivariate analysis in geometric morphometrics relies on a suite of specialized software tools. The following table details key solutions for data digitization, processing, and statistical analysis.

Table 2: Essential Software Tools for Geometric Morphometric Analysis

Tool Name	Function/Best Use	Availability
TPS Dig2 [34]	Digitizing landmarks on 2D digital images. The standard starting point for many 2D GMM studies.	Free
MorphoJ [35]	Integrated software for a wide range of GMM analyses, including PCA, CVA, regression, and modularity tests. User-friendly.	Free
geomorph (R package) [36]	A comprehensive package for the collection and analysis of geometric morphometric data within the R environment. Highly flexible for advanced users.	Free (R)
StereoMorph (R package) [34]	Digitizing landmarks and curves, and for generating 3D models using multiple 2D images.	Free (R)
PAST [34]	Paläontological Statistics software; a general-purpose statistical package with strong support for morphometric analyses, including PCA and CVA.	Free

Advanced Considerations and Machine Learning

Beyond traditional methods, modern approaches are enhancing taxonomic morphometrics. A critical preliminary step is the assessment of measurement error through repeated digitizations, which is fundamental for data accuracy but often neglected [13]. Furthermore, the field is witnessing a paradigm shift with the integration of machine learning (ML) classifiers. For instance, studies on fruit fly morphometrics have demonstrated that Support Vector Machine (SVM) and Artificial Neural Network (ANN) models can achieve predictive accuracies over 95%, significantly outperforming traditional methods and offering powerful new candidates for developing automated species identification systems [33].

Geometric morphometrics (GM) has revolutionized the quantitative analysis of biological form by preserving geometric relationships throughout the statistical process [10]. This approach overcomes fundamental limitations of traditional morphometrics, which relied on linear measurements, ratios, and angles that were often highly autocorrelated and failed to capture complex shape information [10]. By using Cartesian coordinates of homologous points (landmarks), curves, and contours, GM enables researchers to analyze pure shape variation after removing differences in position, orientation, and scale through Procrustes superimposition [10] [37].

The application of GM spans diverse biological disciplines, from taxonomy and systematics to ecology and evolutionary biology [10]. This article examines three specific case studies demonstrating how GM techniques are applied in entomology, paleontology, and biomedical research, framed within best practices for taxonomic research. Each case study highlights specific methodological considerations, experimental protocols, and analytical frameworks that ensure robust, reproducible results.

Methodological Foundation of Geometric Morphometrics

Core Principles and Workflow

Geometric morphometrics relies on the operational definition of shape as "the geometric information that remains after removing differences in position, orientation, and scale" [10]. The standard GM workflow involves several key stages: (1) image acquisition, (2) landmark digitization, (3) Procrustes superimposition, and (4) multivariate statistical analysis [13] [10].

Procrustes superimposition is a critical step that registers objects to a common coordinate system by translating centroid positions to the origin, scaling to unit centroid size, and rotating to minimize distances between corresponding landmarks [10] [37]. The resulting Procrustes coordinates represent shape variables that can be analyzed using standard multivariate techniques like Principal Component Analysis (PCA) and Canonical Variate Analysis (CVA) [10] [38].

Software and Analytical Tools

Several specialized software packages support GM analyses, with increasing integration into open programming environments like R [19]. Key packages include:

• geomorph: Comprehensive toolkit for GM analysis [13]
• Morpho: Provides algorithms for GM and computational morphology [19]
• Momocs: Specializes in outline analysis [19]
• morphospace: New R package for building and visualizing morphospaces [19]

These tools enable researchers to execute the entire GM pipeline while providing advanced ordination and visualization capabilities [19].

Case Study 1: Entomology - Cryptic Species Discrimination in Macrostylid Isopods

Research Context and Objectives

Deep-sea macrostylid isopods present a significant taxonomic challenge due to their remarkably low morphological variation despite high genetic diversity [38]. Traditional taxonomic approaches relying on linear measurements and character ratios have proven insufficient for discriminating among closely related species [38]. This case study evaluated the efficacy of GM techniques for distinguishing five macrostylid species from Icelandic waters where conventional methods struggled.

Experimental Protocol

Table 1: Specimen Information for Entomology Case Study

Species	Number of Specimens	Sex	Collection Projects
Macrostylis spinifera	41 total across species	Female only	BIOICE, IceAGE, PolySkag
M. sp. aff. spinifera	-	-	-
M. subinermis	-	-	-
M. longiremis	-	-	-
M. magnifica	-	-	-

Specimen Preparation and Imaging: Researchers selected 41 female specimens (subadult and adult) from five Macrostylis species [38]. Only females were used as they are more abundant in collections and harder to distinguish using traditional morphology [38]. Each pleotelson (posterior body segment) was photographed in dorsal view using a Leica M165C stereomicroscope with a Leica DMC5400 camera [38]. Images were saved in TIFF format using Leica Application Suite (LAS X) [38].

Landmarking Protocol: The pleotelson was selected as it represents an important diagnostic character in macrostylid taxonomy [38]. Three homologous landmarks and 66 semilandmarks were digitized using tpsDig software [38]:

• Landmark 1: Point where lateral pleotelson outline meets the 7th pereonite
• Landmark 2: Midpoint of the posterior apex of the pleotelson
• Landmark 3: Maximum curvature point where uropod inserts into pleotelson
• Semilandmarks: 66 points placed along curves between landmarks 1 and 2 to capture lateral and posterior margins

Data Processing and Analysis: Raw coordinate data underwent Procrustes superimposition to remove non-shape variation [38]. The resulting Procrustes coordinates were analyzed using Principal Component Analysis (PCA) to visualize pleotelson shape variation and Canonical Variate Analysis (CVA) with permutation tests (10,000 iterations) to assess interspecific shape differences [38].

Key Findings and Taxonomic Significance

The GM analysis successfully discriminated among macrostylid species based on pleotelson shape variation [38]. The PCA created a morphospace where specimens clustered by species, with closer points indicating similar shapes and distant points indicating dissimilar shapes [38]. The CVA further confirmed significant interspecific shape differences in the pleotelson [38].

This study demonstrated that GM could detect subtle morphological differences invisible to traditional taxonomic approaches, providing taxonomists with a powerful tool for identifying and classifying cryptic species in challenging groups like macrostylid isopods [38].

Case Study 2: Paleontology - Biological Profiling of Prehistoric Hand Stencils

Research Context and Objectives

Prehistoric hand stencils provide direct impressions of artists' hands but characterizing the biological profile (sex and age) of these individuals remains challenging [37]. Previous studies used traditional morphometrics (e.g., Manning Index based on digit ratios), but these approaches have significant limitations [37]. This study investigated whether GM could analyze hand stencils despite substantial variation in finger positions in archaeological specimens [37].

Experimental Protocol

Table 2: Experimental Design for Paleontology Case Study

Variable	Specification
Sample Size	70 living adults (35 female, 35 male)
Hands Scanned	Left hands only (more common in archaeological record)
Scanning Method	HP Officejet Pro 8600 Plus contact scanner (300 dpi JPEG)
Landmarks	32 2D conventional landmarks on anatomical reference points

Specimen Preparation and Imaging: Researchers collected 2D left-hand scans from 70 living adults of known biological sex and age (balanced sample of 35 females and 35 males, all over 20 years old) [37]. Each participant was scanned in three standardized positions to mimic archaeological variability:

• Position 1 (Closed): Fingers fully extended and adducted (close together but not touching)
• Position 2 (Natural): Fingers fully extended and semi-spread apart
• Position 3 (Fully open): Fingers fully extended and maximally abducted [37]

This design resulted in 210 total images (3 positions × 70 individuals) [37].

Landmarking Protocol: Thirty-two 2D landmarks were digitized from each scan using TPSdig2 software [37]. Landmarks were placed on key anatomical reference points of the hand to enable detailed size and shape analysis [37].

Data Processing and Analysis: Landmark coordinates underwent Generalized Procrustes Analysis to remove translation, rotation, and scaling effects [37]. Researchers then computed:

• Centroid Size: To assess the impact of hand position on object size
• Procrustes Distances: To compare intra-individual vs. inter-individual variation
• Allometric Analysis: Multivariate regression of shape on size with 100 permutations
• Principal Component Analysis: To visualize shape variation by biological sex [37]

Key Findings and Taxonomic Significance

The analysis revealed that intra-individual variation (different positions of the same hand) was significantly larger than inter-individual variation (differences between individuals) [37]. Mean Procrustes distances between positions 1-2, 2-3, and 1-3 were 0.132, 0.191, and 0.292 respectively, while mean inter-individual distances for the same positions were 0.122, 0.142, and 0.165 [37].

This finding demonstrates that relative finger position creates substantial morphological variation that can overshadow biologically informative signals like sexual dimorphism [37]. For taxonomic applications, this highlights the critical importance of standardizing specimen orientation and position during data acquisition, particularly when working with natural historical collections or archaeological artifacts where control over original positioning is impossible [37].

Case Study 3: Biomedicine - Mammalian Cranial Evolution Analysis

Research Context and Objectives

This study addressed fundamental methodological challenges in large-scale evolutionary morphology by comparing traditional landmark-based GM with emerging landmark-free approaches [39]. While GM is considered the gold standard for evolutionary shape analysis, manual landmarking is time-consuming, prone to observer bias, and limited when comparing morphologically disparate taxa with few homologous points [39]. The research evaluated Deterministic Atlas Analysis (DAA), a landmark-free method, for analyzing cranial shape across 322 mammalian species spanning 180 families [39].

Experimental Protocol

Table 3: Experimental Design for Biomedicine Case Study

Method	Specimens	Modalities	Analysis Type
Manual Landmarking	322 mammals, 180 families	CT and surface scans	Geometric morphometrics
Deterministic Atlas Analysis (DAA)	322 mammals, 180 families	Poisson surface reconstruction	Landmark-free morphometrics

Specimen Preparation and Imaging: The dataset included 322 crown and stem placental mammals representing 180 families [39]. Specimens were obtained from mixed imaging modalities (CT scans and surface scans), creating challenges for comparative analysis [39]. Researchers addressed this by standardizing data using Poisson surface reconstruction to create watertight, closed surfaces for all specimens [39].

Landmarking Protocol: The traditional GM approach used manual landmarking and semilandmarking techniques with homologous anatomical points [39]. The landmark-free DAA approach used Large Deformation Diffeomorphic Metric Mapping (LDDMM) to compute deformations between a dynamically generated atlas shape and each specimen [39]. Control points guided shape comparison without predefined landmarks [39].

Data Processing and Analysis: For traditional GM, raw landmark coordinates underwent Procrustes superimposition [39]. For DAA, momentum vectors ("momenta") representing deformation trajectories were analyzed using kernel Principal Component Analysis (kPCA) [39]. Researchers compared methods using:

• Euclidean distances and Mantel tests to assess matrix correlation
• PROTEST (Procrustes randomization test) to quantify agreement between methods
• Heatmaps based on thin-plate spline deformations to visualize shape differences
• Macroevolutionary analyses of phylogenetic signal, morphological disparity, and evolutionary rates [39]

Key Findings and Taxonomic Significance

After standardizing mesh topology, both methods showed significant improvement in correspondence, though differences remained, particularly for Primates and Cetacea [39]. Both approaches produced comparable but varying estimates of phylogenetic signal, morphological disparity, and evolutionary rates [39].

The study demonstrated that landmark-free methods like DAA offer substantial efficiency advantages for large-scale studies across disparate taxa [39]. However, researchers noted several challenges that must be addressed before widespread adoption, including sensitivity to initial template selection and kernel width parameters [39]. For taxonomic research, this highlights the potential for automated approaches to expand analytical scope while emphasizing the continued importance of methodological validation.

Cross-Disciplinary Best Practices in Geometric Morphometrics

Standardized Experimental Workflow

Based on the case studies, a robust GM workflow for taxonomic research should include:

Figure 1: Standardized GM Workflow for Taxonomic Research

Research Reagent Solutions

Table 4: Essential Materials and Software for Geometric Morphometrics Research

Category	Specific Tools	Function	Application Context
Imaging Equipment	Leica M165C stereomicroscope, HP Officejet Pro 8600 Plus scanner, CT scanners	High-resolution image acquisition	Specimen digitization across scales
Landmarking Software	tpsDig, tpsUtil	Digitize landmarks and semilandmarks	Coordinate data collection
Analytical Packages	geomorph, Morpho, Momocs, morphospace (R packages)	Statistical shape analysis	Multivariate analysis and visualization
Visualization Tools	MorphoJ, morphospace package	Create morphospaces and shape models	Results interpretation and presentation

Methodological Considerations for Taxonomic Research

The case studies reveal several critical considerations for implementing GM in taxonomic research:

Specimen Positioning and Standardization: The paleontology study demonstrated that positional variation can overshadow biological signals [37]. Taxonomists must standardize imaging protocols and consider positional effects when interpreting results.

Landmark Selection and Homology: The entomology study used biologically homologous landmarks complemented by semilandmarks to capture outline information [38]. Careful landmark selection that reflects conserved developmental patterns is essential for meaningful comparisons.

Method Validation: The biomedicine study emphasized the importance of validating novel methods against established approaches [39]. This is particularly relevant with emerging automated techniques that promise efficiency but require careful benchmarking.

Statistical Power and Error Assessment: All case studies employed rigorous statistical frameworks including permutation tests, Procrustes distances, and multivariate regression [37] [39] [38]. Preliminary analyses of measurement error, statistical power, and outliers are fundamental for robust taxonomic conclusions [13].

Geometric morphometrics provides a powerful framework for quantitative shape analysis across biological disciplines. The case studies in entomology, paleontology, and biomedicine demonstrate both the versatility of GM approaches and the critical importance of methodological rigor in taxonomic research. By implementing standardized workflows, validating methods, and maintaining careful attention to anatomical homology, researchers can leverage GM to uncover subtle patterns of morphological variation that inform taxonomy, systematics, and evolutionary biology. As automated methods continue to develop, their integration with traditional GM approaches promises to further expand the scope and scale of morphological research.

Overcoming Common Challenges: A Troubleshooting Guide for Reliable GM

Addressing Missing Data and Incomplete Specimens in Landmarking

The foundation of robust taxonomy research lies in high-quality, complete morphological datasets. However, the reality of working with biological specimens—including fossils, rare taxa, or damaged samples—often introduces the significant challenge of missing data [40]. In geometric morphometrics (GM), a suite of tools for quantifying biological shape, most methods are highly intolerant of such gaps [40]. The presence of missing landmarks can compromise entire analyses, leading to biased results, reduced statistical power, and ultimately, an inaccurate understanding of trait diversification and evolutionary relationships [40]. This whitepaper provides an in-depth technical guide to addressing missing data within the context of a GM workflow, framing best practices that ensure the integrity and reliability of taxonomic research.

The Problem of Missing Data in Morphometrics

Missing data in landmark-based studies typically arises from incomplete, broken, distorted, or otherwise damaged specimens [40]. In taxonomy, these problematic specimens are often the most critical to include; fossil lineages and rare taxa, which are frequently poorly represented in collections, are precisely the materials needed to fully capture morphological variation within a clade [40]. Excluding them can introduce systematic bias and limit the scope of scientific inquiry.

Most multivariate morphometric methods, both linear and geometric, require a complete dataset where every specimen has a value for every landmark [40]. When data is missing, researchers must choose a strategy to handle the incompleteness. The strategic approach taken can profoundly impact the outcome of the analysis, influencing the perceived patterns of shape variation and divergence.

Strategic Approaches to Missing Data

Researchers generally have three overarching strategies for dealing with missing data in their datasets [40]. The following table summarizes these core strategies and their implications.

Table 1: Strategic Approaches for Handling Incomplete Specimens in Morphometric Analyses

Strategy	Description	Best Use Cases	Key Limitations
Trait Removal	Removing the measurement(s) missing data from all specimens in the dataset [40].	Missing data is restricted to one or a few traits that are unlikely to have a major impact on overall shape characterization [40].	Severely limits the dataset to a small number of traits; discards useful information from other landmarks [40].
Specimen Removal	Removing the incomplete specimen from the dataset entirely [40].	Few specimens are damaged, and they originate from species or populations that are well-represented by other, complete individuals in the dataset [40].	Risks losing rare or unique morphological information from critical taxa (e.g., fossils, rare species), potentially biasing the results [40].
Data Estimation	Estimating the missing data using statistical methods or interpolation techniques to "fill in the gaps" [40].	Incomplete specimens are essential to the study and cannot be excluded without compromising the scientific question.	The effectiveness of different estimation methods can vary across and even within datasets; requires careful method selection [40].

The decision-making workflow for navigating these strategic choices is visualized below.

Data Estimation Methodologies

When data estimation is the chosen strategy, several techniques are available. It is critical to select a method based on the dataset's properties and the biological question.

Common Estimation Techniques

Thin-Plate Spline (TPS) Interpolation is a widely used method in geometric morphometrics. It uses the deformation between complete specimens (the reference) to estimate landmarks in an incomplete specimen (the target). However, one study found TPS to be one of the least reliable methods across diverse datasets, urging caution in its application [40].

Regression-Based Estimation involves predicting the coordinates of a missing landmark from the coordinates of other, non-missing landmarks in the same specimen, using a regression model built from a set of complete specimens.

Mean Substitution is a simpler method where the missing landmark in a specimen is replaced by the mean coordinate of that same landmark from all other complete specimens in the sample. This method can be a reasonable baseline but may reduce overall shape variance in the dataset.

Evaluating Estimation Method Effectiveness

The performance of different estimation methods is not universal. A comparative study recommended using the dataset of complete specimens to evaluate different methods via simulation before applying them to the real missing data [40]. This involves:

• Artificially removing known landmarks from complete specimens.
• Applying one or more estimation methods to predict the "missing" values.
• Comparing the estimated coordinates to the actual known coordinates to measure accuracy and bias.

This simulation-based approach allows researchers to identify the most effective estimation method for their specific dataset.

Table 2: Comparison of Common Missing Data Estimation Techniques

Estimation Technique	Methodology Overview	Relative Performance	Key Considerations
Thin-Plate Spline (TPS)	Interpolates missing points based on the bending energy of a theoretical metal plate deformed to match reference specimens [40].	One of the least reliable across datasets [40].	Common but can be unpredictable; requires validation.
Regression-Based Methods	Uses multivariate regression to predict a missing landmark's coordinates from the other, present landmarks in the specimen.	Highly variable; performance depends on the correlation structure of the dataset.	Can be powerful if strong correlations exist among landmarks.
Mean Substitution	Replaces a missing landmark with the mean coordinate of that landmark from all complete specimens in the sample.	Generally reduces variance and can bias results if used injudiciously.	Simple to implement but should be used as a baseline comparison only.

The Scientist's Toolkit: Research Reagent Solutions

Success in geometric morphometrics and the handling of missing data relies on a suite of specialized software tools. The following table details the essential digital "reagents" for a modern GM workflow.

Table 3: Essential Software Toolkit for Geometric Morphometrics and Data Estimation

Tool Name	Primary Function	Role in Addressing Missing Data
TPS Series (tpsDig2, tpsRelw, tpsUtil) [27]	Digitizing landmarks, managing TPS data files, and performing relative warps analysis.	The core software suite for landmark data acquisition and file management, often used as a platform for data estimation protocols.
MorphoJ [27]	A comprehensive Java application for multivariate statistical analysis of shape.	Performs a wide range of GM analyses and includes tools for missing data estimation, such as TPS interpolation.
R Statistical Environment with geomorph & LOST packages [40] [27]	Provides a powerful, scriptable environment for advanced statistical analysis and custom workflows.	The geomorph package is a standard for GM analysis. The LOST package is specifically designed for evaluating missing data estimation techniques in morphometrics [40].
ImageJ [27]	An open-source image processing program used for image acquisition and pre-processing.	Used to prepare specimen images (e.g., scaling, rotation, background removal) prior to landmark digitization.

An Integrated Experimental Protocol for Missing Data

The following workflow, adapted from a detailed protocol for fish morphology, provides a generalized, step-by-step guide for a GM analysis that incorporates the handling of missing data [27].

Step-by-Step Execution:

• Image Acquisition & Preparation: Capture high-resolution, standardized images of specimens. Use software like ImageJ to remove backgrounds, ensure consistent scale, and orient all specimens uniformly (e.g., body axis horizontal, head facing left) [27].
• Landmark Digitization: Using tpsDig2, place Type I (anatomical junctions), Type II (maxima of curvature), and Type III (constructed points) landmarks on every specimen [27]. Document any landmarks that cannot be digitized due to damage as "missing."
• Data Management & Validation: Use tpsUtil to compile the landmark data from all specimens into a single TPS data file. This file is the input for subsequent analyses.
• Handle Missing Data: This is the critical step. Based on the decision workflow (Diagram 1), choose and execute a strategy. For estimation, use the simulation approach with the LOST package in R to test methods before applying the best-performing one to the true missing values [40].
• Procrustes Superimposition: Perform a Generalized Procrustes Analysis (GPA) in MorphoJ or R to remove the effects of non-shape variation (position, scale, rotation). This creates a matrix of Procrustes shape coordinates for statistical analysis [27].
• Statistical Shape Analysis: Analyze the Procrustes coordinates using multivariate techniques like Principal Component Analysis (PCA) to identify major axes of shape variation, or Discriminant Function Analysis (DFA/Canonical Variate Analysis) to test for group differences [27].
• Visualization & Interpretation: Use thin-plate spline deformation grids in tpsRelw or MorphoJ to visualize the shape changes associated with statistical results (e.g., movement of landmarks along a principal component) [27]. Interpret these morphological changes in a taxonomic and evolutionary context.

The silent extinction of species is paralleled by a loss of taxonomic expertise, making it imperative to extract maximum information from every available specimen, even incomplete ones [41]. A deliberate, evidence-based approach to missing data is not a methodological footnote but a cornerstone of rigorous geometric morphometrics in taxonomy. By systematically evaluating and integrating incomplete specimens through robust estimation protocols, researchers can build more comprehensive and accurate representations of morphological diversity. This practice strengthens the foundational framework upon which our understanding of evolution, ecology, and biodiversity conservation is built.

In taxonomic research utilizing geometric morphometrics (GMM), the reliability of findings is the cornerstone of scientific validity. Measurement error—arising from random variation or systematic bias in data collection—can inflate variance, reduce statistical power, and potentially obscure true biological signals [42]. The "replication crisis" in science underscores that the failure to reproduce findings is often rooted in unaccounted methodological variability [43]. For GMM, which relies on the precise digitization of landmarks, this variability is frequently introduced by the human operator. Studies have demonstrated that inter-operator error can contribute between 19.5% and 60% of total shape variation and, in some cases, can even dominate the main patterns of biological variation in large datasets [43] [44]. Therefore, establishing and adhering to rigorous protocols for intra- and inter-operator repeatability testing is not merely a best practice but a fundamental requirement for ensuring the accuracy and credibility of taxonomic comparisons.

Core Concepts and the Impact of Measurement Error

Types of Measurement Error in GMM

In geometric morphometrics, measurement error can be categorized into two primary types:

• Random measurement error: This refers to unpredictable, non-systematic variation introduced during data acquisition. It increases the overall variance in a dataset, which can lead to a loss of statistical power, making it more difficult to detect genuine group differences [42].
• Systematic bias: This is a non-random error where measurements consistently deviate from the true value in a specific direction. It can be introduced by different operators or protocols and can lead to incorrect conclusions by being misinterpreted as biologically meaningful variation [42] [44].

Consequences for Taxonomic Research

The impact of unaddressed measurement error in GMM is profound. A landmark study on human MRI data revealed that inter-operator bias could account for over 30% of the total sample shape variation, an effect so substantial that it surpassed the well-established morphological differences between hundreds of male and female individuals [44]. Similarly, research on Patagonian lizards found that measurement error increased with the complexity of the quantified shape, and inter-operator error contributed significantly to total variation [43]. This highlights that even precise landmarks may not guarantee negligible errors in shape data, and the reliability of findings is inextricably linked to the consistency of the data collection protocol.

Experimental Protocols for Assessing Repeatability

A robust assessment of repeatability involves structured experiments designed to quantify the variability introduced by a single operator over time (intra-operator error) and between different operators (inter-operator error).

Protocol for Intra-Operator Repeatability Testing

This protocol evaluates the consistency of a single trained individual.

• Sample Selection: A subset of specimens (e.g., 10-20% of the total sample or a minimum of 10-20 specimens) is selected to represent the full morphological range of the study [44] [45].
• Repeated Digitization: The same operator digitizes all landmarks on the selected specimens multiple times. It is critical that these repeated sessions are performed on different days and in a randomized order to avoid the introduction of memory bias [42] [44].
• Data Management: All data must be collected blindly, meaning the operator should not be aware of specimen identities or group affiliations during digitization to prevent subconscious bias.

Protocol for Inter-Operator Repeatability Testing

This protocol assesses the impact of multiple individuals collecting data, a common scenario in collaborative research.

• Operator Training: All operators must be trained on the same landmark definitions and digitization protocol. This often involves joint sessions using practice specimens not included in the main study [46].
• Standardized Landmark Definitions: Landmarks must be clearly defined. Type I (discrete anatomical points) and Type II (geometric constructs like maxima of curvature) landmarks are generally more reliable than Type III (extremal points susceptible to sampling effects) [42].
• Data Collection: A common set of specimens is digitized independently by each operator. The use of 3D-printed replicas of a reference collection has been demonstrated as an effective solution for distributing identical specimens among international collaborators, ensuring the physical objects being measured are perfectly consistent [46].
• Standardized Imaging: For 2D GMM, the photography procedure must be rigorously standardized, including camera calibration, focal length, specimen positioning, and lighting conditions, to minimize parallax error and other imaging artifacts [46].

Quantitative Assessment of Measurement Error

Once repeatability data is collected, statistical analysis is used to quantify the magnitude of error. The following table summarizes the key metrics and methods used.

Table 1: Statistical Methods for Quantifying Measurement Error in GMM

Method	Data Type	Purpose	Interpretation
Procrustes ANOVA [42] [44]	Shape (Procrustes coordinates)	Partitions total variance into components due to individual specimens (biological signal) and measurement error.	A high variance component for "error" relative to "specimen" indicates poor repeatability.
Lin's Concordance Correlation Coefficient (CCC) [45]	Continuous data (e.g., landmark coordinates)	Assesses agreement between two sets of repeated measurements; values range from 0 (no agreement) to 1 (perfect agreement).	CCC > 0.99 indicates excellent agreement; CCC < 0.95 may signal concerning levels of error [45].
Intraclass Correlation Coefficient (ICC)	Continuous data	Similar to CCC, it measures reliability based on the proportion of total variance attributed to the subjects.	ICC > 0.9 is often considered a threshold for high reliability.
MANOVA on Replicate Means [42]	Shape (Procrustes coordinates)	Tests for systematic bias (e.g., between operators). A significant effect indicates the presence of non-random error.	A significant p-value suggests that operator bias is a source of systematic variation in the data.

Workflow for Data Analysis

The process of assessing measurement error follows a logical sequence, from data collection to final interpretation, ensuring that the biological signal is distinguishable from noise.

Diagram 1: Workflow for repeatability analysis in GMM. The process is iterative; if error is unacceptably high, protocols must be refined and testing repeated.

The Scientist's Toolkit: Essential Reagents and Materials

Implementing these protocols requires a set of key tools and resources. The following table details essential solutions for ensuring reliability in GMM studies.

Table 2: Research Reagent Solutions for GMM Repeatability Testing

Tool / Material	Function in Repeatability Testing	Examples & Notes
3D Printed Replicas [46]	Provides physically identical specimens for distribution among multiple operators, enabling direct assessment of inter-observer error without travel.	Created from 3D scans of key specimens; ideal for collaborative, international teams.
Standardized Imaging Chamber [46]	Controls lighting, focal length, and specimen position to eliminate parallax and other optical distortions as a source of error in 2D GMM.	Can be custom-built or purchased; includes a fixed camera mount and calibrated scale.
Landmarking Software	Facilitates precise digitization of landmarks and semi-landmarks on 2D images or 3D models.	tpsDig [38], Viewbox [45], MorphoJ [38].
R Packages for GMM	Provides a comprehensive suite of tools for Procrustes superimposition, statistical analysis (e.g., Procrustes ANOVA), and visualization.	geomorph [13] [45], Momocs [13].
Detailed Landmarking Protocol	A written document with visual guides that unambiguously defines the location and type of every landmark and semi-landmark.	The single most cost-effective tool for reducing inter-operator error [46] [43].

Recommendations for Best Practices in Taxonomy

Based on the reviewed literature, the following recommendations are crucial for taxonomists employing GMM:

• Pilot Testing is Non-Negotiable: Always conduct a formal repeatability analysis in a pilot phase before committing to full-scale data collection [42] [44].
• Prioritize Landmark Clarity: The use of more landmarks does not automatically lead to better performance. Focus on a set of well-defined, reliable landmarks, even if the number is smaller. Complex shapes are particularly susceptible to low repeatability [43].
• Use a Single Operator When Possible: For studies where the highest level of shape consistency is critical, having a single, well-trained operator digitize the entire dataset can eliminate inter-operator bias [44].
• Report Error Metrics: To enhance the transparency and credibility of taxonomic research, publications should routinely include a summary of repeatability assessments, such as the results of a Procrustes ANOVA or concordance coefficients [42] [43].

In the context of geometric morphometrics for taxonomy, assuming data reliability is a significant risk. Measurement error, particularly from inter-operator differences, is not a minor nuisance but a major source of variation that can compromise the integrity of research findings. By implementing the protocols outlined in this guide—systematically testing for intra- and inter-operator error, quantifying it using robust statistical tools, and adhering to standardized best practices—researchers can fortify their work against the replication crisis. A rigorous commitment to repeatability ensures that the morphological differences identified and used for taxonomic decisions are genuine biological signals, not artifacts of methodological inconsistency.

Optimizing Landmark Sets for Morphologically Conservative or Damaged Specimens

Geometric morphometrics (GM) has revolutionized the quantitative analysis of biological shape, providing powerful statistical methodologies for studying morphological evolution, taxonomy, and phenotypic variation [47]. A persistent challenge in GM research, particularly within taxonomy, is acquiring adequate sample sizes of ideal specimens, as museum collections often contain individuals with varying degrees of damage or pathological conditions [47]. Traditionally, such specimens are excluded from analyses over concerns that missing data or altered morphologies could distort shape variation assessments. However, emerging evidence suggests that strategic inclusion of these specimens can bolster sample sizes and even enhance the detection of dominant biological signals, provided that landmarking protocols are carefully optimized [47].

This technical guide synthesizes current best practices for optimizing landmark sets within the specific context of morphologically conservative taxa or damaged specimens. Optimizing landmark configurations is not merely a technical exercise; it is a fundamental step that determines the statistical power, biological validity, and interpretive value of a geometric morphometric study. By providing a structured framework for landmark selection, data collection, and analysis, this guide aims to empower researchers to make informed decisions that enhance the robustness and reproducibility of taxonomic research using geometric morphometrics.

Foundational Concepts and Challenges

The Sample Size Problem in Morphometrics

The pursuit of adequate sample sizes is a central concern in geometric morphometrics. While a minimum of 15–20 specimens per sample has been suggested to generate consistent estimates of mean shape, centroid size variance, and shape variance [47], achieving this threshold is often complicated by practical constraints. For vertebrate skeletal morphology studies relying on museum dry bone specimens, available specimens may be limited, and many may exhibit conditions considered deleterious to reliable shape data [47]. These conditions generally fall into three categories:

• Postmortem Damage: Breakage or missing elements occurring after death (e.g., shelf damage).
• Perimortem Damage: Unhealed injuries incurred at or near the time of death (e.g., bullet wounds).
• Antemortem Pathologies: Healed injuries and evidence of disease occurring during life (e.g., healed fractures, dental pathologies, osteoarthritis) [47].

The automatic exclusion of specimens exhibiting these conditions substantially reduces achievable sample sizes and may inadvertently omit demographic-specific shape variation from groups more likely to exhibit these conditions [47].

Impact of Specimen Quality on Shape Analysis

Research on crab-eating macaques (Macaca fascicularis) has demonstrated that the inclusion of damaged/pathologic specimens in larger datasets can strengthen statistical support for dominant biological predictors of shape, such as sexual dimorphism and allometry [47]. The normal variation present in numerous undamaged specimens appears to overwhelm unique individual variation resulting from damage or pathology. However, analyzing only the most severely affected specimens in isolation can confound statistical outputs for less influential principal components and predictors [47].

For small sample sizes bolstered with damaged specimens, analyses typically provide adequate assessment of major shape components but may identify finer-scale differences that require careful interpretation [47]. Consequently, optimization strategies must balance the benefits of increased sample size against potential noise introduced by non-normal morphologies.

Methodological Framework for Landmark Optimization

Landmark Types and Biological Homology

A landmark in geometric morphometrics is a point of biological correspondence located on each specimen in a study [10]. Landmarks are generally categorized as:

• Type I Landmarks: Defined by discrete local topological features (e.g., foramina, suture intersections).
• Type II Landmarks: Defined by local geometry (e.g., points of maximum curvature).
• Type III Landmarks: Defined by extremal points (e.g., furthest point along an axis).
• Semilandmarks: Points used to outline curves and surfaces between traditional landmarks [10].

For morphologically conservative or damaged specimens, Type I landmarks provide the most reliable foundation due to their clear homology, while semilandmarks require careful sliding procedures to minimize artifactual variation.

Strategic Landmark Set Design

Designing an optimized landmark set requires balancing comprehensive coverage with practical implementability, especially when working with damaged material.

Table 1: Landmark Configuration Strategies for Challenging Specimens

Strategy	Application Context	Implementation	Considerations
Modular Landmarking	Specimens with localized damage	Landmarking divided into cranium, mandible, or regional modules [47]	Enables exclusion of damaged modules while retaining use of intact regions
Hierarchical Landmarks	Mixed-quality specimen sets	Core (essential) vs. supplementary landmark classification	Maintains analyses with core landmarks when supplementary points are missing
Adaptive Semilandmarks	Irregular contours or damaged edges	Dynamic placement of semilandmarks based on available morphology [48]	Requires careful sliding algorithms to minimize arbitrary variation

Experimental Protocols for Validation

Protocol: Assessing Impact of Damaged Specimens

A systematic experimental approach is essential for validating the inclusion of damaged specimens in any specific study system.

Objective: To quantitatively evaluate how the inclusion of damaged/pathologic specimens influences the assessment of normal shape variation in a dataset.

Materials:

• 3D surface meshes of crania and mandibles from 100 adult Macaca fascicularis (or taxon of interest) [47]
• Surface scanner (e.g., HDI 120 blue-LED scanner)
• Geometric morphometrics software (e.g., Landmark Editor, Geomorph R package)

Methodology:

• Specimen Categorization: Classify specimens into quality tiers during scanning:
  • Tier 1: All landmarks present, no damage/pathology
  • Tier 2: Minor damage/pathology (e.g., slight alveolar recession)
  • Tier 3: Major damage/pathology (e.g., antemortem tooth loss, healed fractures) [47]
• Landmark Data Collection: Place fixed landmarks and semilandmarks on all specimens. For missing elements due to damage, mark coordinates as missing data. For antemortem conditions, retain data but note alterations [47].
• Dataset Construction: Create multiple datasets for comparative analysis:
  • Dataset 1: Only Tier 1 specimens (all landmarks present, no damage/pathology)
  • Dataset 2: Tiers 1 + 2 (includes minor damage/pathology)
  • Dataset 3: Tiers 1 + 2 + 3 (includes all specimens regardless of condition)
  • Dataset 4: Only Tier 3 specimens (severely damaged/pathologic) [47]
• Statistical Comparison: Perform standard GM tests (Procrustes ANOVA, PCA, regression) on all datasets to evaluate:
  • Strength of allometric patterns
  • Expression of sexual dimorphism
  • Covariation between structures
  • Variance explained by principal components [47]

Interpretation: Compare statistical outputs across datasets. If inclusion of damaged specimens (Datasets 2-3) strengthens support for dominant biological predictors without substantially altering major shape components, their inclusion is justified. If Dataset 4 (damaged-only) yields markedly different results, this suggests caution when analyzing such specimens without reference to normal variation [47].

Protocol: Evaluating Landmark Configurations with Reduced Sample Sizes

Objective: To determine how sample size reduction impacts mean shape estimation and shape variance for different landmark configurations.

Materials:

• Large intraspecific sample (n > 70) of high-quality specimens [48]
• Imaging equipment (e.g., Canon EOS 70D with macro lens)
• Landmarking software (e.g., tpsDIG2)

Methodology:

• Image Acquisition: Photograph all specimens in consistent orientations (lateral, ventral cranial views; lateral mandibular views) [48].
• Comprehensive Landmarking: Apply multiple landmark configurations:
  • Configuration A: Minimal landmark set (Type I only)
  • Configuration B: Balanced set (Type I + II with limited semilandmarks)
  • Configuration C: Dense set (Type I + II + extensive semilandmarks) [48]
• Sample Size Simulation: From the full dataset, randomly subsample without replacement at decreasing sample sizes (e.g., n = 50, 30, 20, 15, 10).
• Shape Analysis: For each subsample and landmark configuration, perform Generalized Procrustes Analysis and calculate:
  • Distance from true mean shape (full dataset)
  • Mean shape variance
  • Morphological disparity [48]

Interpretation: Determine which landmark configuration maintains the most accurate estimation of population mean shape at minimal sample sizes. Typically, balanced configurations (Configuration B) outperform both minimal and excessively dense configurations when samples are small [48].

Decision Framework and Workflow

The following diagram illustrates the systematic decision process for optimizing landmark sets and specimen inclusion, integrating the methodologies described in this guide:

Research Reagent Solutions

Table 2: Essential Materials and Software for Geometric Morphometrics

Item	Function/Application	Implementation Example
3D Surface Scanner (e.g., HDI 120 blue-LED scanner)	Creation of high-resolution 3D models from physical specimens [47]	Surface scanning crania and mandibles; exporting .ply files
Landmark Digitization Software (e.g., Landmark Editor v. 3.6)	Precise placement of 2D/3D landmarks and semilandmarks on digital models [47]	Placing 84 fixed and 104 semilandmarks on crania; 36 fixed and 74 semilandmarks on mandibles
R Package Geomorph (v. 4.0.5)	Comprehensive statistical analysis of shape data [48]	Performing Generalized Procrustes Analysis; principal component analysis
Image Processing Software (e.g., Geomagic Studio)	Mesh cleaning and preparation [47]	Filling small sections of missing data with "Mesh Doctor" and "Fill" functions
Photographic Equipment (e.g., Canon EOS 70D with macro lens)	Standardized 2D image capture for 2DGM [48]	Photographing specimens in lateral cranial, ventral cranial, and lateral mandibular views

Optimizing landmark sets for morphologically conservative or damaged specimens requires a nuanced approach that balances statistical rigor with practical constraints. The protocols and frameworks presented herein provide a roadmap for making informed decisions about specimen inclusion and landmark configuration. Key principles emerge: (1) damaged specimens can valuably bolster sample sizes and enhance detection of dominant biological signals when combined with intact specimens; (2) modular and hierarchical landmarking strategies maximize data retention from imperfect specimens; and (3) systematic validation should precede full-scale analysis when working with mixed-quality specimens. By adopting these best practices, researchers can enhance the robustness, reproducibility, and biological insight of taxonomic studies using geometric morphometrics, ultimately advancing our understanding of morphological diversity and evolution.

In taxonomic research, the accurate identification of true species-specific shape differences is paramount. This process is complicated by the presence of asymmetry, a common feature in biological structures that, if unaccounted for, can obscure genuine taxonomic signals. Asymmetry represents the deviation from perfect symmetry, which is a fundamental feature of the body plans of most organisms and many of their parts [49]. For taxonomists utilizing geometric morphometrics (GMM), a sophisticated approach to quantifying and analyzing morphological variation, distinguishing between different types of asymmetry is not merely a methodological refinement but a necessity for robust classification.

The challenge lies in the fact that observed morphological variation comprises both directional biological signals and various forms of asymmetry-induced noise. Fluctuating asymmetry (FA), defined as random, non-directional deviations from perfect bilateral symmetry, is generally ascribed to developmental accidents or noise and serves as an indicator of developmental instability [50]. In contrast, directional asymmetry (DA) represents consistent differences between sides across a population, such as the arrangement of internal organs where traits are consistently developed differently on the right and left sides [49]. A third type, antisymmetry, describes patterns where deviations are consistently directed but randomly toward either the left or right side. The core objective for taxonomists is to isolate true shape differences from these confounding asymmetric variations, thereby ensuring that taxonomic decisions reflect evolutionary relationships rather than developmental noise or consistent asymmetric patterns.

This technical guide provides a comprehensive framework for addressing asymmetry within GMM workflows for taxonomy. By integrating both theoretical concepts and practical protocols, we establish best practices for separating fluctuating and directional asymmetry from true shape differences, thereby enhancing the reliability of taxonomic inferences derived from morphological data.

Theoretical Foundations of Symmetry and Asymmetry

Biological and Mathematical Concepts of Symmetry

Symmetry in biological structures can be defined as the repetition of parts in different positions and orientations to each other [49]. The most familiar type is bilateral symmetry, where left and right sides are approximate mirror images, characterized mathematically by a reflection about the median plane. However, biological systems also exhibit complex symmetries, including disymmetry (biradial symmetry), rotational symmetry, translational symmetry (serial homology), and spiral symmetries, each defined by different arrangements of repeated parts [49].

Mathematically, symmetry is formalized using group theory, where the set of all transformations that leave an object unchanged (e.g., reflection, rotation, translation) constitutes its symmetry group [49]. For bilateral symmetry, this group contains the reflection about the median plane and the identity transformation. Understanding these formal concepts is crucial because asymmetry is fundamentally defined as deviation from the expected symmetry pattern, and the analytical approach must be tailored to the underlying symmetry of the structure.

Typology of Asymmetry in Biological Structures

Biological asymmetry manifests in three primary forms, each with distinct characteristics and interpretations:

• Fluctuating Asymmetry (FA): Random, non-directional deviations from perfect bilateral symmetry that are normally distributed around a mean of zero [50]. FA is generally non-heritable and reflects developmental instability, serving as an indicator of how well an organism buffers its development against genetic and environmental stressors [50]. The degree of FA in a population is inversely related to developmental homeostasis.

• Directional Asymmetry (DA): Consistent, directional differences between sides across a population, where one side is consistently larger or differently shaped than the other [49]. While studies of size measurements have found DA only sporadically, directional asymmetry for shape appears to be nearly ubiquitous in all animals that have been examined in sufficiently large studies [49].

• Antisymmetry: Consistent directional deviations, but with the direction (left or right) varying randomly among individuals, resulting in a bimodal distribution of left-right differences [49].

Table 1: Characteristics of Primary Asymmetry Types in Biological Structures

Asymmetry Type	Population-Level Pattern	Biological Interpretation	Taxonomic Implications
Fluctuating Asymmetry	Random deviations with mean zero	Developmental instability	Confounding noise to be partitioned out
Directional Asymmetry	Consistent bias to one side	Heritable, adaptive asymmetry	Must be accounted for before species comparisons
Antisymmetry	Bimodal distribution of side differences	Specialized adaptation	Can be misinterpreted as discrete types

Geometric Morphometrics Framework for Asymmetry Analysis

Core Concepts of Geometric Morphometrics

Geometric morphometrics is a mathematical and statistical approach that quantitatively assesses shape variation while preserving the geometric properties of morphological structures throughout analysis [45] [13]. Unlike traditional morphometrics, which relies on linear measurements, distances, or ratios, GMM captures the geometry of morphological structures using landmarks—discrete, homologous points that can be precisely located across specimens [45].

The GMM workflow typically involves: (1) digitizing landmarks (and often semi-landmarks for curves and surfaces) from specimens; (2) performing Generalized Procrustes Analysis (GPA) to remove variation due to position, orientation, and scale; (3) statistical analysis of the resulting Procrustes shape coordinates; and (4) visualization of results back in the original morphology space [45] [13]. This framework is particularly powerful for asymmetry studies because it preserves the geometric relationships among landmarks throughout analysis, allowing for meaningful biological interpretation of results.

The Procrustes ANOVA Framework for Asymmetry

The most robust analytical approach for separating asymmetry components in taxonomic studies is the Procrustes ANOVA, which extends the conventional two-way ANOVA customary for analyses of fluctuating asymmetry to shape data [51]. This method partitions total shape variation into components attributable to individual effects, side effects, and individual-side interaction effects, providing a comprehensive assessment of different asymmetry types.

The fundamental model decomposes the total shape variation of a structure into:

• Individual Effect: Variation among individuals, representing true biological differences, including taxonomic signals
• Side Effect: Consistent differences between left and right sides, indicating directional asymmetry
• Individual × Side Interaction: Non-directional side-specific variation, representing fluctuating asymmetry
• Measurement Error: Variation due to methodological imprecision, estimated through replicate measurements

Table 2: Variance Components in Procrustes ANOVA for Asymmetry Analysis

Variance Component	Biological Interpretation	Statistical Test	Taxonomic Significance
Individual	True shape differences among specimens	F-test: Individuals MS / Interaction MS	Contains species differences
Side	Directional Asymmetry	F-test: Side MS / Interaction MS	Consistent bias; must be accounted for
Individual × Side	Fluctuating Asymmetry	F-test: Interaction MS / Error MS	Developmental noise; should be minimized
Measurement Error	Methodological imprecision	-	Should be minimized through protocol optimization

This partitioning is crucial for taxonomy because it allows researchers to isolate the individual-level variation that contains species differences from the asymmetry components that represent confounding noise or consistent directional patterns.

Experimental Protocols for Asymmetry Analysis

Specimen Preparation and Data Collection

The foundation of reliable asymmetry analysis lies in meticulous data collection. For 2D analyses, standardized imaging protocols are essential, ensuring consistent orientation, scale, and lighting across all specimens [13]. For 3D data, which is increasingly accessible through CT scanning and surface laser scanning, the same principles of standardization apply [45]. The choice between 2D and 3D approaches involves trade-offs: 2D methods are more accessible and efficient but may miss important aspects of morphological variation, while 3D approaches capture complete geometry but require more resources [13].

Landmark selection should include both fixed anatomical landmarks and sliding semi-landmarks to capture curves and surfaces [45]. For the nasal cavity study cited, researchers used 10 fixed landmarks and 200 sliding semi-landmarks to adequately capture the morphology of the region of interest [45]. This combination provides comprehensive coverage while maintaining homology across specimens. To ensure reliability, intra- and inter-operator repeatability should be assessed using metrics such as Lin's Concordance Correlation Coefficient (CCC) [45].

Landmarking and Data Preprocessing

The landmarking process begins with the identification of fixed anatomical landmarks present in all specimens. Subsequently, semi-landmarks are distributed across the morphological surface or curve of a template specimen and then projected onto each specimen in the dataset using Thin Plate Spline (TPS) warping, which minimizes bending energy [45]. These semi-landmarks are then allowed to slide tangentially along the surface to minimize artificial variance and ensure optimal homology across specimens [45].

Generalized Procrustes Analysis (GPA) is then performed to align all specimens into a common coordinate system by removing differences in position, orientation, and scale [45] [13]. This step is crucial as it isolates pure shape variation, which is the focus of subsequent asymmetry analyses. The aligned Procrustes coordinates serve as the input for the Procrustes ANOVA and other statistical analyses.

Diagram 1: Asymmetry Analysis Workflow (Total Width: 760px)

Statistical Analysis and Hypothesis Testing

The core analysis employs Procrustes ANOVA to test specific hypotheses about asymmetry patterns [51]. The following statistical tests are performed sequentially:

• Directional Asymmetry Test: The null hypothesis of no consistent side difference is tested using an F-ratio of side mean squares to individual × side interaction mean squares. A significant result indicates directional asymmetry that must be accounted for in subsequent taxonomic comparisons.

• Fluctuating Asymmetry Test: The null hypothesis of no individual-specific side differences is tested using an F-ratio of individual × side interaction mean squares to measurement error mean squares. A significant result indicates the presence of fluctuating asymmetry.

• Individual Differences Test: The null hypothesis of no consistent differences among individuals is tested using an F-ratio of individual mean squares to individual × side interaction mean squares. A significant result indicates genuine shape variation that may contain taxonomically informative signals.

For taxonomic applications, it is crucial to estimate effect sizes alongside statistical significance, as large sample sizes may yield statistically significant results with minimal biological importance [13]. Confidence intervals for shape differences can be generated through bootstrapping or permutation procedures.

Essential Research Reagents and Tools

Table 3: Essential Research Reagents and Computational Tools for Asymmetry Analysis

Tool Category	Specific Examples	Function in Analysis	Implementation Considerations
Imaging Equipment	CT scanners, digital cameras, laser surface scanners	Generate 2D/3D morphological data	Resolution, precision, and standardization critical
Landmarking Software	Viewbox 4.0 [45], tpsDig2, MorphoJ	Digitize landmarks and semi-landmarks	Supports both fixed and sliding semi-landmarks
Statistical Environment	R with geomorph package [45] [13]	Procrustes ANOVA and shape analysis	Comprehensive GMM analysis capabilities
Specialized GMM Software	Momocs [13], EVAN Toolbox	Outline analysis and shape visualization	Handles both landmark and outline data
Visualization Tools	MeshLab, ParaView, R visualization packages	3D shape visualization and rendering	Critical for interpreting results in morphological context

The R package geomorph is particularly valuable as it provides integrated functions for the entire workflow, from GPA through Procrustes ANOVA to visualization [45] [13]. For researchers new to GMM, user-friendly software with graphical interfaces may lower initial barriers, but programming-based approaches offer greater analytical flexibility and reproducibility [13].

Practical Application in Taxonomic Research

Case Study: Marmot Mandibles in Taxonomic Delineation

A comprehensive study of North American marmot mandibles illustrates the application of asymmetry analysis in taxonomic research [13]. This research employed Procrustes GMM to assess population differences while controlling for asymmetric variation. The protocol included:

• Preliminary Analyses: Assessment of measurement error, identification of outliers, and evaluation of statistical power [13]
• Asymmetry Partitioning: Application of Procrustes ANOVA to separate directional asymmetry, fluctuating asymmetry, and true individual variation
• Taxonomic Comparison: Analysis of size and shape differences among groups after accounting for asymmetry components
• Allometric Analysis: Examination of size-related shape changes within and between putative taxonomic groups

This approach revealed that failing to account for asymmetry components would have inflated estimates of among-group differences and potentially led to erroneous taxonomic conclusions. The study demonstrated that a significant portion of the total shape variation was attributable to asymmetry rather than genuine taxonomic signals.

Interpretation of Results in Taxonomic Context

In taxonomic applications, the variance component attributable to individual differences (after accounting for asymmetry) contains the signal of interest for species delimitation. The effect size of individual differences relative to asymmetry components provides an indication of how much morphological distinction exists beyond developmental noise and consistent asymmetric patterns.

When individual variation (taxonomic signal) substantially exceeds asymmetry components, researchers can have greater confidence in the taxonomic distinctions based on morphology. Conversely, when asymmetry components constitute a large proportion of total variance, taxonomic inferences based on morphology alone should be made cautiously, and integration with molecular, ecological, or behavioral data becomes particularly important [13].

Visualization of shape differences associated with taxonomic groups should focus on the individual component of variation after asymmetry has been partitioned out. This provides a clearer picture of genuine species-specific morphology without the confounding effects of developmental noise or population-level asymmetric biases.

The separation of fluctuating and directional asymmetry from true shape differences represents a critical methodological refinement in geometric morphometric approaches to taxonomy. By implementing the Procrustes ANOVA framework and associated protocols outlined in this guide, researchers can significantly enhance the reliability of taxonomic inferences derived from morphological data. The integration of careful experimental design, appropriate statistical partitioning of variance components, and thoughtful interpretation of results in a taxonomic context provides a robust foundation for identifying evolutionarily significant units and advancing our understanding of biodiversity.

Sample Size Considerations and Power Analysis for Robust Statistical Inference

Geometric morphometrics (GM) has revolutionized the quantitative analysis of biological shape, providing taxonomists with powerful tools for discriminating closely related taxa and understanding morphological evolution [48]. However, the statistical robustness of these analyses is critically dependent on appropriate sample size and rigorous power analysis. In taxonomic research, where morphological differences can be subtle and specimens are often limited, understanding these considerations becomes paramount for producing valid, reproducible scientific conclusions [13]. This guide examines the core principles of sample size determination and power analysis within the context of geometric morphometrics, establishing best practices for taxonomic applications.

The Impact of Sample Size on Geometric Morphometric Analyses

Empirical Evidence of Sample Size Effects

The influence of sample size on geometric morphometric results is well-documented. Systematic investigations using large intraspecific sample sizes (n > 70) for bat species have demonstrated that reducing sample size directly impacts estimates of mean shape and increases shape variance [48]. These findings underscore a critical challenge in taxonomic studies: small samples may fail to capture the true morphological variation within populations, potentially leading to erroneous taxonomic conclusions.

Similarly, sampling experiments investigating estimates of mean shape have revealed that inaccuracies can vary substantially depending on the geometric morphometric method employed [52]. The generalized Procrustes analysis (GPA) method has been shown to produce estimates with the least error and no pattern of bias, while other methods may exhibit both larger errors and systematic bias, particularly when sample sizes are inadequate [52].

Consequences of Inadequate Sampling

• Increased sampling error: Small samples yield less precise estimates of population parameters [13]
• Reduced statistical power: Diminished ability to detect true morphological differences between taxa [13]
• Inflated type I and type II error rates: Increased likelihood of both false positives and false negatives [52]
• Biased shape estimates: Systematic distortion of mean shape configuration [48]
• Compromised discriminability: Reduced ability to distinguish closely related species [48]

Table 1: Impact of Sample Size Reduction on Shape Estimates Based on Empirical Studies

Sample Size Reduction	Impact on Mean Shape	Impact on Shape Variance	Taxonomic Implications
Moderate reduction (n=30-50)	measurable distortion	noticeable increase	potential misclassification of marginal specimens
Substantial reduction (n=15-25)	significant bias	substantial inflation	compromised species discrimination
Severe reduction (n<15)	severe inaccuracy	extreme values	unreliable taxonomic conclusions

Methodological Framework for Sample Size Determination

Power Analysis in Geometric Morphometrics

Power analysis provides a principled approach to determining adequate sample sizes before conducting morphometric studies. The relationship between effect size, sample size, significance level, and statistical power follows established statistical principles, though with special considerations for shape data.

For taxonomic studies using geometric morphometrics, key considerations include:

• Effect size expectations: Based on preliminary data or published studies of related taxa
• Data dimensionality: Higher-dimensional data (e.g., more landmarks) typically require larger samples
• Study design: More complex designs (e.g., multiple groups, covariates) increase sample demands
• Statistical method: Different analytical approaches have varying power characteristics [52]

Practical Protocols for Sample Size Planning

Figure 1: Workflow for sample size determination in taxonomic morphometric studies

A systematic protocol for determining adequate sample sizes should include:

• Preliminary data collection: Obtain data from a small but representative sample (typically 10-15 specimens per group) [13]
• Effect size calculation: Quantify the expected morphological differences between groups using appropriate shape metrics
• Power analysis: Use statistical software to calculate sample requirements for desired power (typically 80% or higher) at standard alpha levels (e.g., 0.05)
• Practical adjustment: Balance statistical ideals with specimen availability, considering potential strategies to maximize power within constraints

Quantitative Guidelines for Taxonomic Studies

Table 2: Recommended Minimum Sample Sizes for Different Taxonomic Questions Based on Empirical Evidence

Taxonomic Application	Minimum Sample per Group	Recommended Sample per Group	Key Considerations
Intraspecific variation	15-20	30+	Sexual dimorphism, geographic variation must be accounted for
Interspecific discrimination (closely related)	20-25	40+	Effect sizes typically small; requires greater power
Cryptic species detection	25-30	50+	Minimal morphological differences demand large samples
Ontogenetic shape analysis	15-20 per stage	25+ per stage	Developmental stages may have different variance patterns
Geographic variation	15-20 per population	25+ per population	Hierarchical structure may require mixed models

Empirical studies provide concrete evidence for these recommendations. Research on lasiurid bats demonstrated that species discrimination between Lasiurus borealis and L. seminolus was statistically significant across all views and elements when adequate samples were employed [48]. Similarly, studies of macrostylid isopods successfully discriminated between species using geometric morphometrics with sample sizes ranging from 5-15 per species, though larger samples would strengthen such analyses [38].

Addressing Sample Size Challenges in Practical Taxonomy

Strategies for Limited Specimen Availability

Taxonomic research frequently faces practical constraints on specimen availability. Several strategies can enhance robustness when ideal samples are unattainable:

• Pooling data across collections: Combine specimens from multiple museums or sources, though this requires careful assessment of inter-observer and inter-collection bias [53]
• Data augmentation techniques: Employ statistical methods such as generative adversarial networks (GANs) to create synthetic specimens, though with appropriate validation [54]
• Focus on high-information landmarks: Prioritize landmarks that capture maximal morphological variation to improve power with limited specimens [55]
• Bayesian approaches: Incorporate prior information where available to strengthen inferences from small samples

Error Assessment and Protocol Validation

Regardless of sample size, rigorous assessment of measurement error is essential. Protocol should include:

• Repeated measurements: Assess intra-observer error through replicate digitizations [13]
• Multiple operators: Evaluate inter-observer error when collaborative datasets are used [53]
• Landmark validation: Confirm that landmarks can be placed consistently across specimens
• Statistical control: Ensure measurement error is substantially smaller than biological effect sizes

Table 3: Research Reagent Solutions for Geometric Morphometrics in Taxonomy

Tool/Category	Specific Examples	Function in Morphometric Research
Imaging Equipment	DSLR cameras (Canon EOS 70D), stereomicroscopes (Leica M165C), structured-light scanners (Artec Eva)	Generate 2D or 3D digital representations of specimens for landmark digitization
Digitization Software	tpsDig2, MorphoJ, Viewbox 4, geomorph R package	Collect landmark coordinates, perform Procrustes superimposition, and statistical shape analysis
Statistical Frameworks	Generalized Procrustes Analysis (GPA), Principal Component Analysis (PCA), Canonical Variate Analysis (CVA)	Extract shape variables, reduce dimensionality, and test group differences
Error Assessment Tools	Intraclass correlation coefficients, Procrustes ANOVA, measurement error modules in morphometric software	Quantify and control for sources of variation beyond biological signal
Data Augmentation Algorithms	Generative Adversarial Networks (GANs), bootstrap resampling methods	Address small sample size limitations through synthetic data generation

Figure 2: Strategies and considerations for addressing sample size limitations

Robust statistical inference in taxonomic geometric morphometrics requires careful attention to sample size considerations throughout the research process. Evidence consistently demonstrates that inadequate samples can distort estimates of mean shape, inflate variance, and compromise species discrimination. By incorporating power analysis during study design, implementing rigorous error assessment protocols, and employing appropriate strategies for limited specimens, taxonomists can strengthen the validity and reproducibility of their morphological conclusions. As geometric morphometrics continues to evolve as a tool in systematics, maintaining methodological rigor in sample size determination remains fundamental to advancing taxonomic knowledge.

Validating GM Results and Comparative Analysis with Other Methods

Geometric morphometrics (GM) has emerged as a powerful quantitative tool for capturing and analyzing biological shape, offering significant advantages over traditional morphometric approaches. In taxonomic research, accurately delineating species boundaries is fundamental, yet it is often complicated by phenotypic plasticity, morphological stasis, and homoplasy [56]. This guide benchmarks GM against molecular data and traditional morphometrics, framing the comparison within best practices for taxonomy. The objective is to provide researchers with a structured framework for evaluating when and how to integrate GM into species identification and delineation protocols, especially in contexts where molecular methods may be impractical or cost-prohibitive.

Core Methodologies in Comparison

Geometric Morphometrics (GM)

GM is a landmark-based analytical tool that enables the complete quantification of shape by analyzing the geometric coordinates of defined points on an organism [56]. The core methodology involves:

• Landmark Data Collection: Cartesian coordinates of Type I, II, and III landmarks are digitized from 2D images or 3D models. Type I landmarks represent biologically defined point correspondences (e.g., vein intersections), Type II represent maxima of curvature, and Type III are defined by geometric constructions [57].
• Generalized Procrustes Analysis (GPA): This procedure removes the non-shape effects of size, position, and orientation by superimposing landmark configurations. It minimizes the sum of squared distances between corresponding landmarks across specimens to extract pure shape information [57] [58].
• Shape and Size Variables: The output of GPA is a set of Procrustes shape coordinates. Centroid size (CS), calculated as the square root of the sum of squared distances of all landmarks from their centroid, is used as a size metric that is statistically independent of shape [57] [58].
• Statistical Analysis: Principal Component Analysis (PCA) is routinely performed on the Procrustes coordinates to visualize the major patterns of shape variation in a reduced-dimensional space [57].

Traditional Morphometrics

Traditional morphometrics typically relies on linear measurements, angles, or ratios between defined points.

• Data Collection: Measurements are taken using calipers or from images, often including lengths, widths, and circumferences [53].
• Analysis: Data are analyzed using multivariate statistics like PCA or discriminant analysis. However, these analyses are based on inter-landmark distances rather than the geometric configuration of landmarks, potentially losing information about the relative spatial arrangement of structures [53].

Molecular Systematics

Molecular techniques use genetic data to infer evolutionary relationships and delimit species.

• Common Markers: For species identification and phylogeny, commonly used markers include the mitochondrial Cytochrome c Oxidase Subunit I (COI) gene, Internal Transcribed Spacer (ITS) regions, and other nuclear genes like Tyrosine Hydroxylase (TH) [58].
• Phylogenetic Analysis: DNA sequences are aligned, and phylogenetic trees are constructed using methods like maximum likelihood or Bayesian inference to visualize evolutionary relationships and assess species monophyly [58].

Comparative Analysis: Strengths and Limitations

Table 1: Benchmarking Geometric Morphometrics against Traditional Morphometrics and Molecular Data.

Aspect	Geometric Morphometrics	Traditional Morphometrics	Molecular Data
Data Type	Geometric coordinates of landmarks and semilandmarks [57].	Linear distances, angles, ratios [53].	DNA or RNA nucleotide sequences [58].
Primary Output	Procrustes shape coordinates; visualization of shape change [56].	Covariance matrices of measurements; size-adjusted values.	Phylogenetic trees; genetic distance matrices.
Key Advantage	Visually intuitive; retains full geometric information; powerful for subtle shape differences [56] [53].	Methodologically simple; low technical barrier; fast data collection.	Direct insight into evolutionary history and gene flow; high resolution for cryptic species [58].
Key Limitation	Susceptible to digitization error and operator bias [53].	Loss of geometric shape information; limited to predefined measurements.	Does not directly address phenotypic disparity; can be costly and time-consuming [58].
Typical Application	Quantifying symmetric and asymmetric shape variation; identifying cryptic species based on shape [56] [57].	Distinguishing groups based on gross size differences [53].	Determining evolutionary relationships and species boundaries [58].
Cost & Time	Moderate (requires specific software and training).	Low.	High (requires lab facilities and reagents).
Error Sources	Landmark mis-placement; intra- and inter-operator bias [53].	Measurement inaccuracy; orientation bias.	Sequencing errors; homoplasy; incomplete lineage sorting [58].

Table 2: Empirical Performance Comparison from Case Studies.

Study System	GM Performance	Molecular Performance	Key Finding	Reference
Carex spp. (Sedges)	Utricle shape variation supported the exclusion of C. herteri from the C. phalaroides group and showed affinities to sect. Abditispicae.	Not available for the studied specimens, necessitating the use of GM.	GM provided systematic insights where molecular data was unavailable, confirming its utility in taxonomic resolution [56].	[56]
Anopheles spp. (Mosquitoes)	Cross-validation accuracy of 74.8% for identifying 8 species; effective but not definitive.	COI region could not clearly distinguish some species; ITS2 and TH were more useful.	GM alone was not sufficient for definitive identification of all species; an integrative approach was recommended [58].	[58]
Wild vs. Domestic Pigs	Effectively discriminated taxa based on molar shape with various landmark/semi-landmark protocols.	Not applied in the cited study.	Highlighted the importance of selecting a morphometric protocol with low measurement error for successful discrimination [53].	[53]

Best Practices and Experimental Protocols

A Protocol for Taxonomic Studies Using GM

The following workflow is recommended for benchmarking GM in a taxonomic context:

• Sample Selection and Imaging: Select specimens representing the taxonomic groups of interest. Standardize imaging protocols (e.g., camera settings, orientation, scale) rigorously to minimize non-biological variance [53].
• Landmarking Protocol:
  • Define a landmark scheme that includes biologically homologous points (Types I and II). For complex curves, semilandmarks can be applied [57].
  • The same operator should digitize all specimens, preferably in a randomized order. If multiple operators are involved, a formal assessment of inter-operator bias must be conducted [53].
• Data Analysis:
  • Perform GPA to obtain Procrustes shape coordinates and centroid size.
  • Use PCA to explore the major patterns of shape variation.
  • Perform discriminant analysis if the goal is classification. Use cross-validation to estimate misclassification rates, as was done in the Anopheles study [58].
• Validation and Benchmarking:
  • Against Traditional Morphometrics: Collect linear measurements from the same specimens. Compare the group separation and classification accuracy achieved by both GM and traditional methods using the same statistical tests.
  • Against Molecular Data: For a subset of specimens, generate molecular data (e.g., COI, ITS2) to build a reference phylogeny. Compare the group assignments from GM (based on shape) with the clades identified in the molecular phylogeny [58].

Diagram 1: Workflow for benchmarking Geometric Morphometrics in taxonomy.

Managing Measurement Error and Data Pooling

A critical best practice is the quantification of measurement error (ME), especially when pooling datasets from multiple operators or studies.

• Intra-operator Error: A single operator should digitize a subset of specimens multiple times (e.g., 3-5x) in a randomized order. The variance among these replicates quantifies intra-operator ME [53].
• Inter-operator Error: Multiple operators should digitize the same set of specimens. The variance introduced by different operators must be compared to the biological variance of interest. If the inter-operator error is significant and approaches the magnitude of the biological signal, datasets should not be pooled, or statistical models must account for this bias [53].
• Protocol Optimization: Test different morphometric protocols (e.g., landmark-only vs. landmark-and-semilandmark) on a small subset of data. Select the protocol that provides the best discrimination power with the lowest associated ME for the specific research question [53].

The Researcher's Toolkit

Table 3: Essential Research Reagents and Solutions for Geometric Morphometrics.

Item / Solution	Function / Application	Technical Notes
Imaging Setup	High-resolution capture of specimen morphology for landmark digitization.	Use a standardized setup with a DSLR/microscope, fixed focal length lens, and scale bar. Ensure consistent lighting [53].
Digitizing Software	Software used to collect landmark coordinates from digital images.	Examples include TPSDig2 [58] and ImageJ [57]. Essential for creating TPS files.
Statistical Software with GM Packages	Platforms for performing Procrustes superimposition and subsequent statistical analyses.	The R environment with packages like geomorph [58] is the standard for comprehensive GM analysis.
Canada Balsam / Mounting Medium	For preparing and mounting delicate structures (e.g., insect wings) on microscope slides.	Prevents movement and deformation during imaging, as used in the Anopheles wing study [58].
Voucher Specimens	Authoritatively identified reference specimens stored in a collection.	Crucial for validating taxonomic identity and providing a permanent reference for morphological studies.

Benchmarking studies consistently demonstrate that geometric morphometrics is a powerful tool for taxonomy, particularly for resolving complexes where morphological differences are subtle or confounded by homoplasy [56]. However, its highest utility is realized not in isolation, but as part of an integrative taxonomic framework. GM often outperforms traditional morphometrics by capturing more complex shape data and providing intuitive visualizations, but it may not achieve the definitive resolution of molecular methods, especially for recently diverged or cryptic species [58]. The optimal approach for modern taxonomy is one that strategically combines the phenotypic insights from GM with the evolutionary context provided by molecular data, all while adhering to rigorous protocols that minimize and quantify measurement error.

Geometric morphometrics (GM) is an indispensable tool in modern taxonomy and evolutionary biology, providing a statistically rigorous framework for analyzing biological shape. By utilizing coordinate-based data from anatomical landmarks, GM allows researchers to quantify subtle morphological variations that are often crucial for discriminating between closely related species or understanding intraspecific diversity. The power of GM, however, is fully realized only when appropriate statistical tests are applied to evaluate the significance of observed shape differences. Within taxonomic research, establishing whether shape variations represent statistically significant differences is fundamental to making reliable inferences about species boundaries, phylogenetic relationships, and adaptive evolution.

The statistical landscape of GM is built upon specialized implementations of multivariate analysis of variance, primarily Procrustes ANOVA and MANOVA, which are designed to handle the unique properties of shape data. These methods test hypotheses about group differences while accounting for the complex covariance structure of landmark coordinates. Subsequent post-hoc tests then pinpoint specific group contrasts that drive significant overall effects. For taxonomists, this analytical progression provides an objective methodology for evaluating morphological distinctness, thereby offering critical evidence for taxonomic decisions. This guide details the theoretical foundations, practical application, and interpretation of these core statistical tests within the context of taxonomic research, emphasizing best practices to ensure robust and reproducible conclusions.

Theoretical Foundations of Shape Statistics

The Nature of Shape Data

Shape data in geometric morphometrics are represented as Procrustes shape coordinates, which are derived from raw landmark coordinates through Generalized Procrustes Analysis (GPA). GPA removes the non-shape variations of size, position, and orientation by optimally translating, scaling, and rotating landmark configurations [30]. The resulting Procrustes coordinates exist in a curved, non-Euclidean space known as Kendall's shape space. For practical statistical analysis, these coordinates are projected onto a linear tangent space where standard multivariate statistical techniques can be applied. This projection is valid when shapes are sufficiently similar, a condition typically met in intra-familial or intra-generic taxonomic studies.

The statistical analysis of shape coordinates must account for their inherent dimensionality and constraints. For a configuration of (k) landmarks in (m) dimensions, the resulting Procrustes coordinates have (km - m(m+1)/2 - 1) dimensions after removing the effects of position, orientation, and size. This reduced dimensionality, along with the complex correlations among landmark coordinates, necessitates specialized statistical approaches. Furthermore, the Procrustes distance between two shapes—defined as the square root of the sum of squared differences between corresponding landmarks—serves as the fundamental metric for quantifying shape differences in all subsequent statistical tests.

The Rationale for Procrustes-Based Tests

Traditional statistical tests assume that variables are independent and measured without error, assumptions that are violated by the highly correlated nature of landmark coordinates. Procrustes-based statistical methods are specifically designed to accommodate the unique properties of shape data. They operate directly on the Procrustes coordinates or the distances between them, thereby respecting the geometry of shape space. This approach provides several critical advantages for taxonomic research:

• Biological Relevance: Tests based on Procrustes distance quantify actual morphological disparity as perceived in biological terms, rather than relying on linear measurements that may capture only partial aspects of shape variation.
• Statistical Power: By utilizing the complete shape information embedded in landmark configurations, these tests can detect subtle but taxonomically significant shape differences that might be missed by traditional morphometric approaches.
• Error Assessment: Procrustes ANOVA provides a natural framework for partitioning and quantifying measurement error, which is crucial for evaluating whether observed differences reflect true biological variation or methodological artifacts.

Key Statistical Tests: Principles and Applications

Procrustes ANOVA

Procrustes ANOVA (also known as Goodall's F-test) is a fundamental statistical procedure in geometric morphometrics used to assess the significance of shape variation attributable to one or more factors. Unlike traditional ANOVA, which analyzes univariate measurements, Procrustes ANOVA operates on the multivariate shape configuration as a whole.

Table 1: Components of a Typical Procrustes ANOVA for Taxonomic Research

Variation Source	Degrees of Freedom	Sums of Squares	Mean Square	F-value	p-value
Group (Species)	(g-1)	SSG	MSG	F = MSG/MSR	p-value
Residual (Within Group)	(n-g)	SSR	MSR
Total	(n-1)	SST

The mathematical foundation of Procrustes ANOVA involves decomposing the total sum of squared Procrustes distances from the mean shape into components attributable to the factor of interest (e.g., species designation) and residual variation. The test statistic is calculated as:

[ F = \frac{\text{MS}{\text{group}}}{\text{MS}{\text{residual}}} = \frac{\text{SS}{\text{group}} / (g-1)}{\text{SS}{\text{residual}} / (n-g)} ]

where (g) represents the number of groups and (n) the total sample size. The significance of the F-statistic is typically assessed via a permutation test (with 10,000 iterations recommended), which provides a robust non-parametric alternative that does not rely on strict distributional assumptions [59]. In taxonomic applications, a significant Procrustes ANOVA indicates that at least one group mean shape differs from the others, warranting further investigation through post-hoc tests to identify specific group differences.

MANOVA and Its Role in Morphometrics

While Procrustes ANOVA tests for overall group differences based on Procrustes distances, MANOVA (Multivariate Analysis of Variance) operates directly on the tangent space coordinates and tests for differences in the multivariate mean vectors among groups. In taxonomic studies, MANOVA is particularly useful when researchers want to model multiple categorical predictors simultaneously (e.g., species + sex + their interaction) or when the focus is on the multivariate mean vectors themselves.

The MANOVA test statistic for group differences in shape can be formulated as:

[ \Lambda = \frac{|\mathbf{W}|}{|\mathbf{T}|} = \frac{|\mathbf{W}|}{|\mathbf{B} + \mathbf{W}|} ]

where (\mathbf{W}) is the within-group sum of squares and cross-products matrix, (\mathbf{B}) is the between-group sum of squares and cross-products matrix, and (\mathbf{T} = \mathbf{B} + \mathbf{W}) is the total sum of squares and cross-products matrix. Several test statistics are available for MANOVA, including Pillai's trace, Wilks' lambda, Hotelling-Lawley trace, and Roy's largest root. For morphometric applications, Pillai's trace is generally recommended as it is the most robust to violations of assumptions, particularly when sample sizes are unequal or the data deviate from multivariate normality [59].

Table 2: Comparison of Procrustes ANOVA and MANOVA for Shape Analysis

Feature	Procrustes ANOVA	MANOVA
Data Type	Procrustes distances	Tangent space coordinates
Null Hypothesis	No group differences in Procrustes distance	No group differences in multivariate means
Test Statistic	F-statistic	Pillai's trace, Wilks' lambda, etc.
Key Assumption	Isotropic variation (may be relaxed via permutation)	Homogeneity of covariance matrices
Taxonomic Application	Overall test of morphological disparity	Modeling complex group effects and interactions

Post-hoc Testing in Shape Space

When a significant overall group effect is detected by either Procrustes ANOVA or MANOVA, post-hoc tests are necessary to determine which specific group pairs differ significantly. In geometric morphometrics, two primary distance-based metrics are used for pairwise comparisons: Procrustes distance and Mahalanobis distance.

The Procrustes distance between two mean shapes provides a measure of raw morphological disparity, while the Mahalanobis distance incorporates information about within-group variation and covariance structure. The Mahalanobis distance between groups (i) and (j) is calculated as:

[ D^2 = (\bar{\mathbf{x}}i - \bar{\mathbf{x}}j)^\top \mathbf{S}^{-1} (\bar{\mathbf{x}}i - \bar{\mathbf{x}}j) ]

where (\bar{\mathbf{x}}i) and (\bar{\mathbf{x}}j) are the mean shape vectors for groups (i) and (j), and (\mathbf{S}) is the pooled within-group covariance matrix. For both distance measures, statistical significance is typically assessed via permutation tests (with 10,000 permutations recommended) that maintain the family-wise error rate through appropriate correction methods such as Bonferroni or false discovery rate [60].

In taxonomic practice, Procrustes distance is most appropriate when the primary question concerns the absolute magnitude of shape difference between taxa, while Mahalanobis distance is more powerful for discrimination and classification as it accounts for the patterns of covariation within groups. The results of post-hoc tests provide quantitative evidence for taxonomic decisions by identifying which species pairs exhibit statistically significant morphological differentiation.

Experimental Protocols and Workflows

Standardized Data Collection Protocol

Robust statistical inference in geometric morphometrics begins with meticulous data collection. The following protocol outlines key steps for generating high-quality landmark data for taxonomic studies:

• Specimen Selection: Carefully select specimens that represent the taxonomic and geographic range of interest. Sample sizes should be maximized wherever possible, as recent research indicates that reduced sample sizes can substantially impact estimates of mean shape and increase shape variance [48]. Specimens should be adults whenever possible to avoid confounding taxonomic differences with ontogenetic variation.

• Imaging and Landmark Digitization: Capture high-resolution images (2D or 3D) using standardized equipment and positioning. Digitize landmarks and semilandmarks following a consistent protocol. All digitization should ideally be completed by a single researcher in a concentrated time period to minimize the "visiting scientist effect," where time lags between digitization sessions can introduce systematic bias [61].

• Landmark Configuration Validation: Prior to analysis, ensure all landmark configurations are properly aligned and that semilandmarks have been appropriately slid. Check for outliers and potential digitization errors using Procrustes distance plots and other diagnostic tools.

Statistical Analysis Workflow

The following workflow diagram illustrates the sequential process for statistical testing of shape differences in taxonomic research:

This workflow begins with quality-checked, Procrustes-aligned coordinates. Exploratory Principal Component Analysis (PCA) provides an initial visualization of group separation and major patterns of shape variation. The core statistical testing phase employs either Procrustes ANOVA or MANOVA to test the global hypothesis of no group differences. If this overall test is significant, post-hoc pairwise tests identify which specific taxon pairs differ significantly. The results inform biological interpretation and taxonomic decisions.

Validation and Error Assessment Protocol

Accurate statistical inference requires careful assessment of measurement error and validation of results:

• Measurement Error Assessment: Conduct replicate digitizations of a subset of specimens (recommended ≥10% of sample) to quantify measurement error. Use Procrustes ANOVA to partition variance components between biological variation and measurement error. A significant measurement error relative to biological variation indicates problematic landmarking consistency that must be addressed before proceeding with taxonomic comparisons [61].

• Effect Size Evaluation: For significant results, calculate effect sizes (e.g., partial η² for Procrustes ANOVA) to distinguish statistical significance from biological meaningfulness. In taxonomic contexts, even small effect sizes may be important if they represent consistent differences between putative species.

• Classification Validation: Apply linear discriminant analysis or Canonical Variate Analysis (CVA) to assess the predictive power of shape differences. Use cross-validation (leave-one-out or k-fold) to obtain unbiased estimates of classification accuracy, which provides complementary evidence for taxonomic distinctness [59].

The Scientist's Toolkit: Essential Research Reagents

Table 3: Essential Software Tools for Geometric Morphometric Analysis

Software Tool	Primary Function	Application in Taxonomic Research
tpsDig2	Landmark digitization	Collecting 2D landmark coordinates from specimen images
MorphoJ	Integrated morphometric analysis	Performing Procrustes ANOVA, CVA, and discriminant analysis with user-friendly interface
R geomorph package	Comprehensive shape analysis	Conducting Procrustes ANOVA, MANOVA, and other advanced statistical tests in a programmable environment
Imaging Equipment	Specimen documentation	Generating high-resolution 2D/3D images of specimens for landmark digitization

Table 4: Statistical Approaches for Taxonomic Hypothesis Testing

Method	Implementation	Taxonomic Application
Permutation Tests	10,000 permutations recommended	Assessing statistical significance without distributional assumptions
Effect Size Metrics	Partial η², Procrustes distance	Quantifying magnitude of group differences beyond statistical significance
Cross-Validation	Leave-one-out or k-fold	Validating discriminatory power of shape characters for classification
Measurement Error Analysis	Procrustes ANOVA of replicates	Quantifying and controlling for digitization error

Statistical significance testing through Procrustes ANOVA, MANOVA, and appropriate post-hoc tests provides the analytical foundation for rigorous taxonomic research using geometric morphometrics. These methods enable taxonomists to move beyond qualitative descriptions of morphological difference to quantitative, statistically grounded assessments of distinctness. The integration of proper experimental design, careful attention to measurement error, and appropriate interpretation of statistical results ensures that geometric morphometrics fulfills its potential as a powerful tool for taxonomic inquiry. As with all taxonomic characters, shape differences should be interpreted in conjunction with other lines of evidence—including genetic, ecological, and behavioral data—to build robust, integrative taxonomic hypotheses.

In the field of taxonomic research, accurately classifying specimens is fundamental to understanding biodiversity, evolutionary relationships, and ecological dynamics. Geometric morphometrics (GM), which involves the quantitative analysis of biological form using landmark coordinates, has emerged as a powerful tool for discriminating between closely related species, particularly in cases where traditional morphological characters are insufficient. The power of any classification method, however, hinges on the robust evaluation of its success. This guide details the core performance metrics—specifically Procrustes distance and Mahalanobis distance—that are central to evaluating classification success in geometric morphometrics. Framed within best practices for taxonomy, we provide a technical overview of these metrics, their computational methodologies, and their interpretation, supported by practical examples from contemporary research.

Procrustes distance provides a global measure of shape dissimilarity by quantifying the difference between two landmark configurations after superimposition, which removes differences in location, rotation, and scale [62]. In contrast, Mahalanobis distance is a multivariate statistical measure that accounts for the covariance structure within groups, making it particularly sensitive to group differences in shape space and a powerful tool for classification and discriminant analysis [4]. Together, these metrics form the backbone of statistical shape analysis in taxonomic studies, from distinguishing invasive insect species [4] to resolving cryptic complexes in rodents [63] and beetles [64].

Theoretical Foundations of the Core Metrics

Procrustes Distance: A Measure of Pure Shape Difference

Procrustes distance is derived from Procrustes analysis, a superimposition method that removes non-shape variations from landmark data. The process involves three key steps:

• Translation: Centering each landmark configuration to a common origin (usually the centroid).
• Scaling: Scaling configurations to a common unit size (typically Centroid Size).
• Rotation: Rotating configurations to minimize the sum of squared distances between corresponding landmarks.

After this Procrustes superimposition, the shape of an object is represented by its Procrustes coordinates, which reside in a non-Euclidean shape space [62]. The Procrustes distance (PD) between two specimens is then calculated as the square root of the sum of squared differences between their corresponding Procrustes-aligned landmark coordinates.

Statistical significance of shape differences between groups is often tested using a Goodall's F-test, a type of permutation test (e.g., with 10,000 iterations) applied to Procrustes distances, which evaluates whether observed group differences are greater than those expected by chance [4] [62]. In taxonomic studies, a larger Procrustes distance between the mean shapes of two groups indicates greater morphological disparity, which can be interpreted as evidence for species delimitation.

Mahalanobis Distance: A Multivariate Classification Metric

Mahalanobis distance (MD) is a multivariate statistic that measures the distance between a point and a group distribution, or between the centroids of two groups, while accounting for the variance-covariance structure of the data. In geometric morphometrics, MD is computed in the tangent space, a linear approximation of the curved shape space, which allows for the application of standard multivariate statistics [62].

The power of MD in classification stems from its ability to incorporate the inherent correlation between shape variables. Unlike Euclidean distance, which treats all variables as independent and equally variable, MD accounts for the fact that certain directions in shape space are more variable than others. This makes it particularly effective for:

• Classifying unknown specimens into pre-defined groups.
• Assessing the distinctness of groups in a way that is sensitive to within-group variation.
• Highlighting shape features that most contribute to group separation.

In practice, the significance of Mahalanobis distances is typically assessed using permutation tests or by relating the squared MD to an F-distribution, providing a p-value for the hypothesis that the two groups have the same mean shape [4]. Its application is widespread in taxonomy, as seen in studies of thrips [4] and marmots [60].

Quantitative Comparison of Distance Metrics

The table below summarizes the core characteristics, strengths, and limitations of Procrustes and Mahalanobis distances as applied in taxonomic morphometrics.

Table 1: Comparative Overview of Procrustes and Mahalanobis Distances in Taxonomic Morphometrics

Feature	Procrustes Distance	Mahalanobis Distance
Definition	Geometric distance between two landmark configurations after Procrustes superimposition [62].	Multivariate distance accounting for group covariance structure [4].
Sensitivity to Variation	Measures pure shape difference; insensitive to within-group variance.	Explicitly incorporates within-group variance and correlations.
Primary Use Case	Quantifying absolute magnitude of shape difference; visualizing shape divergence.	Classification, discriminant analysis, and hypothesis testing of group differences.
Output Interpretation	Larger PD = greater dissimilarity in mean shape.	Larger MD = greater distinctness between groups relative to within-group variation.
Statistical Testing	Goodall's F-test, Permutation tests on Procrustes coordinates [62].	Permutation tests, F-statistic approximation [4].
Dimensionality	Inherently accounts for the non-Euclidean nature of shape space.	Requires a full-rank covariance matrix; can be unstable with high-dimension, low-sample-size data.
Example from Literature	Used to show significant head shape differences among Thrips species (Permutation test, p<0.0001) [4].	Used to distinguish Thrips species pairs (e.g., T. hawaiiensis vs T. palmi, MD: 4.21, p<0.05) [4].

Experimental Protocols for Metric Evaluation

A standard workflow for evaluating taxonomic classification success using these metrics involves a series of methodical steps, from data collection to statistical inference.

Diagram 1: A workflow for evaluating classification success in geometric morphometrics, highlighting the role of the core performance metrics.

Data Acquisition and Preprocessing

The initial phase involves collecting high-quality morphological data.

• Landmarking: Anatomically homologous landmarks (Type I, II, or III) are digitized on 2D images or 3D models of the structures of interest (e.g., insect heads, mammal mandibles) using specialized software like TPS Dig2 [4] [64].
• Data Cleaning: The raw coordinate data is checked for outliers and missing data. Specimens with low precision landmarks or severe damage should be removed to ensure a "clean" dataset, a critical step for robust results [60].
• Software Tools: Common software packages for these steps include TPS series (e.g., TPS Dig2), MorphoJ, and the geomorph package in R [4] [63].

Procrustes Superimposition and Initial Analysis

This core step prepares the data for shape analysis.

• Procrustes Fit: The landmark configurations are subjected to a Generalized Procrustes Analysis (GPA), which standardizes all specimens by translating them to a common origin, scaling them to unit Centroid Size, and rotating them to minimize the Procrustes distances among them [4] [62].
• Visualizing Shape Space: The resulting Procrustes coordinates are often analyzed via Principal Component Analysis (PCA) to visualize the distribution of specimens in a low-dimensional morphospace. This helps identify major trends of shape variation and potential natural groupings [4] [63].

Calculation of Distance Metrics and Statistical Testing

With the aligned shape data, the core performance metrics can be computed and their significance evaluated.

• Calculating Procrustes Distance: The PD between group mean shapes is calculated directly from their Procrustes coordinates. To test if this observed difference is statistically significant, a permutation test for Procrustes distance (with 10,000 permutations) is typically performed [4]. A significant p-value (e.g., p < 0.05) supports the rejection of the null hypothesis that the groups share the same mean shape.
• Calculating Mahalanobis Distance: MD is typically computed in the space of principal components (PCs) that capture the majority of shape variance, which helps mitigate dimensionality issues. The significance of the MD between groups is likewise assessed using a permutation test [4]. Furthermore, Canonical Variate Analysis (CVA), which maximizes between-group relative to within-group variation, is often used in conjunction with MD to produce a low-dimensional projection ideal for visual classification and for calculating posterior probabilities of group membership [62].

Validation and Reporting

Robust taxonomic studies validate their findings to ensure reliability.

• Cross-Validation: The classification performance based on these distances should be evaluated using a leave-one-out cross-validation (LOOCV) procedure. This involves iteratively classifying each specimen based on a model built from all other specimens, providing an unbiased estimate of misclassification error [60].
• Effect Size Reporting: Best practices dictate reporting not just statistical significance (p-values) but also effect sizes. In this context, the Procrustes and Mahalanobis distances themselves are measures of effect size, quantifying the magnitude of the morphological difference [60].

The Scientist's Toolkit: Essential Reagents & Software

Successful implementation of the described protocols relies on a suite of specialized software and analytical tools.

Table 2: Essential Research Reagents & Software for Geometric Morphometrics

Tool Name	Type/Function	Key Utility in Analysis
TPS Dig2 [4]	Software application	Primary digitization of landmark coordinates from 2D image files.
MorphoJ [4] [62]	Integrated software platform	Conducts Procrustes superimposition, PCA, CVA, and calculation of Mahalanobis/Procrustes distances with permutation tests. User-friendly GUI.
R Statistical Environment (with geomorph [4] & shapes packages)	Programming environment and packages	Provides a comprehensive, flexible, and reproducible pipeline for all steps of GM, from GPA to advanced statistical modeling and validation.
Image Editing Software (e.g., Adobe Photoshop) [4] [64]	Image processing tool	Prepares and enhances specimen images (cropping, contrast adjustment) prior to landmarking to ensure clarity and consistency.
Permutation Tests [4] [62]	Statistical resampling method	Provides a distribution-free method for assessing the statistical significance of Procrustes and Mahalanobis distances.

Procrustes and Mahalanobis distances are complementary pillars for evaluating classification success in taxonomic geometric morphometrics. Procrustes distance offers an intuitive, geometric measure of overall shape dissimilarity, while Mahalanobis distance provides a powerful, variance-sensitive metric for discrimination and classification. The rigorous application of the experimental protocols outlined herein—including proper landmarking, Procrustes alignment, statistical testing via permutation, and cross-validation—ensures that conclusions about species boundaries and morphological distinctness are both statistically sound and biologically meaningful. As morphometrics continues to integrate with novel computational approaches like machine learning [65], these foundational metrics will remain essential for quantifying and interpreting the complex patterns of biological form.

In taxonomic research, accurately quantifying phenotypic variation is fundamental for discriminating species, understanding evolutionary relationships, and identifying evolutionarily significant units. For decades, scientists relied on traditional morphometrics (TM), which uses linear dimensions, angles, and ratios. The emergence of geometric morphometrics (GM) has revolutionized the field by providing powerful methods to capture, analyze, and visualize complex shape geometry. This whitepaper details the conceptual and practical superiority of GM over TM, framing the discussion within best practices for taxonomy. We provide a technical guide with direct comparisons, experimental case studies, and detailed protocols to help researchers adopt these robust methodologies.

Traditional morphometrics has been a cornerstone of biological classification, relying on multivariate statistical analyses of measured distances and ratios [66]. However, a fundamental limitation of TM is that linear distances are not always defined by the same landmarks, making comparative studies challenging. More critically, TM does not capture the complete variation of shape in space; for instance, an oval and a teardrop shape with identical length and width measurements would be deemed morphologically identical [66].

Geometric morphometrics overcomes these limitations by analyzing the geometric configuration of Cartesian landmark and semilandmark coordinates. GM uses Procrustes-based methods to separate shape variation from differences in size, position, and orientation of specimens, preserving the full geometric information throughout the analysis [66] [14]. This allows for sophisticated statistical analyses and, crucially, the visualization of shape changes through deformation grids, making it an indispensable tool for modern taxonomic research [13] [14].

Conceptual and Methodological Comparison

The table below summarizes the core differences between the two approaches.

Table 1: A comparison of Traditional Morphometrics and Geometric Morphometrics.

Feature	Traditional Morphometrics (TM)	Geometric Morphometrics (GM)
Data Type	Linear distances, angles, ratios	Cartesian coordinates of landmarks and semilandmarks [66]
Shape Capture	Incomplete; cannot distinguish between different shapes with identical measurements [66]	Comprehensive; preserves full geometry of the structure [66] [14]
Landmark Homology	Not always required or enforced for measurements [66]	Fundamental; analyses are based on homologous points [66] [27]
Size Correction	Problematic; various methods can yield conflicting results [66]	Standardized via Procrustes superimposition (scaling to unit Centroid Size) [66]
Statistical Power	Good, but limited by data type	Increased statistical power for shape analysis [14]
Visualization of Results	Limited to charts and graphs	Powerful visualization of shape change via deformation grids and wireframes [66] [67]
Primary Software	General statistical packages (e.g., PAST)	Specialized software (e.g., MorphoJ, TPS series, R packages like geomorph and Momocs) [13] [27]

Case Study Evidence: GM Reveals What TM Misses

Empirical studies directly comparing both methods consistently demonstrate the superior capability of GM in detecting biologically meaningful shape differences.

3.1 Lizard Head Dimorphism A seminal study on the Argentine black and white tegu lizard (Tupinambis merianae) investigated sexual dimorphism in head shape. While both linear and geometric methods showed differences in the mandible, only geometric morphometrics detected subtle but functionally significant shape differences in the cranium, specifically in areas related to jaw musculature insertion. These local shape changes, which have direct consequences for bite force, were completely missed by the analysis of linear dimensions [68].

3.2 Fish Body Shape and Sexual Dimorphism Research on Colossoma macropomum used an integrated approach. Geometric morphometrics quantified overall body shape differences between males and females, revealing that males exhibit a longer and broader morphology with distinct positioning of the pectoral and anal fins. Linear morphometrics complemented these findings by confirming significant variations in the head region and anterior body width [67]. This study highlights how GM provides a holistic view of shape changes that can be further contextualized with specific linear measurements.

Table 2: Key findings from comparative morphometric studies.

Study Organism	Traditional Morphometrics Findings	Geometric Morphometrics Findings	Advantage of GM
Lizard (Tupinambis merianae) [68]	Detected intersexual differences in mandible dimensions.	Revealed cranial shape differences in muscle insertion areas; provided insights into functional morphology.	Captured local, functionally relevant shape changes invisible to TM.
Fish (Colossoma macropomum) [67]	Confirmed sex-specific variations in head and anterior body width.	Identified overall body shape dimorphism: shorter, narrower females vs. longer, broader males; visualized fin positioning.	Provided a comprehensive, visualized quantification of overall body form.
Ovenbird (Seiurus aurocapilla) [69]	Used tip angle and width for age classification.	Outline-based methods (semi-landmarks, EFA) enabled high classification rates of age based on subtle tail feather shape.	Enabled accurate classification based on complex outline curves, not just simple metrics.

A Best-Practice Workflow for Geometric Morphometrics in Taxonomy

A robust GM study follows a structured pipeline. The following diagram and subsequent breakdown outline the essential steps from hypothesis to biological interpretation.

Step-by-Step Experimental Protocol

• Study Design: Clearly define the taxonomic hypothesis and the morphological structures (e.g., mandible, leaf, skull) relevant to testing it [66] [14].
• Data Collection:
  • Image Acquisition: Obtain high-resolution 2D or 3D images under standardized conditions [27] [14].
  • Landmarking: Digitize landmarks using software like tpsDig2. Landmarks are classified as:
    • Type I: Anatomical points of clear biological homology (e.g., junction of bones, tip of nose) [27].
    • Type II: Mathematically defined points (e.g., point of maximum curvature) [27].
    • Type III: Constructed points (e.g., midpoints between landmarks) or semilandmarks for curves and outlines [27].
• Data Standardization (GPA): Use software like MorphoJ or R to perform Generalized Procrustes Analysis. This step superimposes all landmark configurations by:
  • Translating them to a common center.
  • Scaling them to a unitless size (Centroid Size).
  • Rotating them to minimize the Procrustes distance between corresponding landmarks [66].
• Statistical Analysis:
  • Principal Component Analysis (PCA): To explore the major, unsupervised patterns of shape variation within the entire sample [66].
  • Canonical Variate Analysis (CVA): To maximize the separation between pre-defined taxonomic groups and assess classification accuracy [69] [67].
  • Multiple Regression: To test for allometry (the effect of size on shape) by regressing Procrustes shape coordinates against Centroid Size [66].
• Interpretation & Visualization: Interpret statistical outputs in a biological context. Use Thin-Plate Spline (TPS) deformation grids to visualize the shape changes associated with statistical axes (e.g., along a CV or PC). This shows a graphical transformation from the mean shape to the target shape, highlighting regions of expansion, contraction, or bending [66] [67].

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Essential software and materials for a geometric morphometrics study.

Item Name	Category	Function/Brief Explanation
tpsDig2 [27]	Software	A standard program for digitizing landmarks from 2D image files.
MorphoJ [27] [67]	Software	User-friendly software for performing Procrustes superimposition, PCA, CVA, and other multivariate analyses.
R packages (geomorph, Momocs) [13] [27]	Software	Powerful, flexible open-source platforms for comprehensive GM analysis, from GPA to advanced statistical modeling.
Generalized Procrustes Analysis (GPA) [66]	Analytical Method	The core mathematical procedure that standardizes landmark configurations for shape comparison.
Thin-Plate Spline (TPS) [66]	Visualization Tool	A mathematical function used to visualize shape differences as a smooth deformation of a grid.
Micro-CT Scanner [70]	Hardware	For non-destructively obtaining high-resolution 3D models of internal and external structures (e.g., skulls).
Flatbed Scanner / DSLR Camera [14]	Hardware	For acquiring high-quality 2D images of flat structures (e.g., leaves, fish) or specimens.
Centroid Size [66] [37]	Metric	A computed, isometric size measure derived from landmarks, used for allometric studies and as a proxy for biological size.

The power of geometric morphometrics lies in its ability to quantify what has traditionally been described qualitatively. For taxonomy, this translates into a more rigorous, reproducible, and insightful framework for delimiting taxa and understanding phenotypic evolution. While traditional morphometrics still has its place, particularly for rapid assessments of specific traits, GM provides a comprehensive and geometrically faithful representation of form. By adopting the detailed workflows and best practices outlined in this whitepaper—from careful landmarking to robust statistical analysis and intuitive visualization—researchers can fully leverage the power of shape data to uncover the subtle yet taxonomically significant morphological patterns that define biodiversity.

Geometric Morphometrics (GM) has undergone a revolutionary transformation through integration with artificial intelligence (AI) and machine learning (ML), creating a powerful confluence that is redefining taxonomic research and beyond. This integration addresses fundamental challenges in traditional morphometrics by enhancing the ability to detect subtle morphological patterns, classify specimens with unprecedented accuracy, and analyze shape variations in high-dimensional spaces. Where conventional GM methods like Procrustes analysis once provided the foundation for quantifying shape by removing the effects of position, scale, and rotation [71], the incorporation of ML algorithms now enables researchers to navigate complex morphological spaces with enhanced predictive power and analytical sophistication. This technical guide examines the core methodologies, implementations, and applications of this integrated approach within the context of taxonomic best practices, providing researchers with a comprehensive framework for leveraging these advanced analytical techniques.

The synergy between GM and ML represents more than merely applying new statistical tools; it constitutes a fundamental shift in how morphological data are analyzed and interpreted. By combining the precise shape quantification of GM with the pattern recognition capabilities of ML, researchers can now tackle previously intractable problems in taxonomy, including the identification of cryptic species, analysis of complex allometric relationships, and understanding of morphological responses to environmental pressures [5] [72]. This whitepaper explores the theoretical foundations, practical implementations, and cutting-edge applications of this integrated approach, providing taxonomists with the knowledge needed to leverage these powerful tools in their research.

Methodological Foundations: From Landmarks to Classifications

Core Geometric Morphometrics Workflow

The GM workflow begins with the acquisition of landmark data, which consists of discrete, homologous points located consistently across all specimens in a study. These landmarks capture the essential geometry of biological structures, whether from crania [5], wings [72], or cut marks [73]. The subsequent Procrustes superposition aligns these landmark configurations through translation, rotation, and scaling to isolate pure shape information by removing positional, orientational, and size differences [71] [13]. This process generates Procrustes shape coordinates that reside in a curved, non-Euclidean shape space.

For statistical analysis, these aligned shapes are typically projected into a linear tangent space, allowing the application of conventional multivariate statistics [71]. Principal Component Analysis (PCA) is then commonly employed to reduce dimensionality and visualize the major axes of shape variation within the dataset. While this traditional GM pipeline effectively quantifies and visualizes shape, its discriminatory power can be limited for closely related taxa with subtle morphological differences, creating the need for more sophisticated analytical approaches.

Machine Learning Integration Framework

The integration of machine learning with GM addresses limitations of traditional multivariate statistics by introducing algorithms capable of learning complex patterns directly from shape data. This framework typically follows one of two approaches: utilizing Procrustes coordinates as direct input features for ML classifiers [72], or employing functional data analysis (FDA) to transform discrete landmarks into continuous curves before analysis [5].

Multiple ML algorithms have demonstrated efficacy in morphometric classification tasks. Support Vector Machines (SVM) with radial basis functions have shown particular success, correctly classifying 83% of Anopheles maculipennis s.s. and 79% of An. daciae specimens in mosquito wing studies [72]. Random Forests offer the advantage of feature importance evaluation, identifying which landmarks contribute most to classification accuracy. Artificial Neural Networks (ANNs) can model complex nonlinear relationships in shape data, while naïve Bayes classifiers provide probabilistic classification based on shape feature distributions [5].

The functional data approach to GM (FDGM) represents a significant methodological advancement, converting 2D landmark data into continuous curves through interpolation and basis function expansion [5]. This approach better captures shape information between landmarks and has demonstrated superior classification performance compared to traditional GM in shrew craniodental studies, particularly when combined with machine learning classifiers [5].

Table 1: Machine Learning Algorithms for Morphometric Classification

Algorithm	Key Features	Taxonomic Application	Performance
Support Vector Machine (SVM)	Effective in high-dimensional spaces, versatile kernels	Mosquito species identification [72]	83% correct classification for An. maculipennis [72]
Random Forest	Ensemble method, feature importance evaluation	Shrew species classification [5]	Superior to PCA in cross-validation [5]
Artificial Neural Network (ANN)	Nonlinear pattern recognition, complex relationships	Multi-species mosquito classification [72]	Higher accuracy than traditional methods [72]
Naïve Bayes	Probabilistic classification, computational efficiency	Shrew craniodental morphology [5]	Effective with functional data approach [5]

Integrated Workflow: From Data Acquisition to Classification

The synergistic workflow combining GM and ML follows a structured pipeline from specimen preparation to final classification, with each stage building upon the previous to maximize analytical precision. The following diagram illustrates this integrated approach:

This workflow begins with rigorous data acquisition, where consistent imaging protocols are critical for reliable results. For 2D analyses, standardized photography with scale references and controlled lighting conditions ensures comparability across specimens [15]. For 3D structures, structured-light scanners [73] or micro-CT scanners generate high-resolution models for landmark placement. The landmark digitization phase requires careful identification of homologous points, with consideration for anatomical consistency across taxa.

Following Procrustes alignment and shape space projection, the ML processing phase begins with feature selection, where algorithms identify the most informative landmarks or shape components for classification [72]. Model training employs cross-validation techniques to optimize parameters and prevent overfitting, with performance metrics including ROC-AUC analysis providing quantitative measures of classification accuracy [72]. The final validation phase tests the model on independent datasets to assess real-world performance and generalizability.

Experimental Protocols and Case Studies

Cut Mark Analysis in Archaeological Contexts

A compelling application of integrated GM-ML methodology appears in taphonomic studies, where researchers have employed these techniques to identify tool types from cut marks on bone surfaces. In a study of Iron Age cut marks from the Ulaca oppidum in Spain, researchers combined 3D scanning, GM, and ML to determine whether marks were produced by metal or flint tools [73]. The experimental protocol involved:

• Sample Preparation: Modern bovine (Bos taurus) long bones with intact meat coverage were used as experimental substrates [73].
• Tool Marks Generation: Controlled cut marks were created using both flint flakes and metal knives (Molybdenum Vanadium C 0.5 CR 14 MO 0.5 VA 0.25) with consistent cutting angles and force application [73].
• 3D Digitalization: A DAVID structured-light scanner SLS-2 captured high-resolution 3D models of each cut mark [73].
• Profile Extraction: Cross-sectional profiles were extracted at 30-70% of mark length using Global Mapper software, with seven landmarks defining profile extremes, depth, and curvature [73].
• ML Classification: The archaeological cut marks were compared against the experimental database using machine learning classifiers to identify tool type.

This approach yielded significant insights, revealing that despite the Iron Age context, most cut marks at Ulaca were produced with flint rather than metal tools, challenging assumptions about technological adoption in daily activities [73].

Insect Taxonomy Using Wing Venation

The integration of GM and ML has proven particularly valuable in entomological taxonomy, where distinguishing cryptic species presents significant challenges. Research on the Anopheles Maculipennis complex demonstrates this application:

• Wing Preparation: Right wings were removed and mounted on slides with cover slips [72].
• Image Acquisition: Digital images were captured with standardized magnification and lighting [72].
• Landmark Configuration: 18 landmarks were placed at wing vein junctions to capture venation patterns [72].
• Validation: Molecular identification via DNA barcoding provided ground truth for training and validation [72].
• Classifier Comparison: Multiple ML algorithms (SVM, Random Forest, ANN, Ensemble) were evaluated against traditional multivariate methods [72].

This study demonstrated the superiority of ML approaches, with SVM achieving 83% classification accuracy for An. maculipennis s.s. and 79% for An. daciae - significantly outperforming traditional discriminant analysis [72]. ROC-AUC analysis further identified landmarks 11, 16, and 15 as most important for discriminating between these sibling species [72].

Table 2: Taxonomic Classification Performance Across Study Organisms

Study Organism	Traditional GM Accuracy	ML-Enhanced Accuracy	Most Effective Algorithm
Mosquitoes (Anopheles Maculipennis complex) [72]	Limited discrimination of sibling species	83% for An. maculipennis, 79% for An. daciae [72]	Support Vector Machine (SVM) [72]
Shrews (Craniodental morphology) [5]	Not specified	Superior classification with functional data [5]	Random Forest [5]
Beetles (Tetropium pronotum shape) [15]	Effective for species discrimination	Not assessed in study	Not applicable
Seeds (Archaeobotanical classification) [74]	Lower performance compared to DL	Outperformed by Deep Learning (CNN) [74]	Convolutional Neural Network [74]

Mammalian Craniodental Morphometrics

The classification of shrew species using craniodental morphology illustrates the application of Functional Data Geometric Morphometrics (FDGM) combined with machine learning [5]. The experimental protocol included:

• Specimen Preparation: 89 crania from three shrew species (S. murinus, C. monticola, and C. malayana) were cleaned and prepared for imaging [5].
• Multiview Imaging: Digital images were captured from dorsal, jaw, and lateral views to comprehensively capture craniodental morphology [5].
• Landmark Digitation: 2D landmarks were placed on consistent anatomical locations across all specimens [5].
• Functional Transformation: Landmark data was converted to continuous curves using basis function expansions [5].
• Comparative Analysis: Traditional GM and FDGM approaches were compared using multiple ML classifiers [5].

This study demonstrated that FDGM outperformed traditional GM in classification accuracy, with the dorsal view providing the best discriminatory power for distinguishing the three shrew species [5]. The integration of functional data analysis with machine learning proved particularly effective for capturing subtle shape variations between closely related taxa.

Implementing an integrated GM-ML pipeline requires specific computational tools and software resources. The following table details essential components of the modern morphometrician's toolkit:

Table 3: Essential Computational Tools for Integrated GM-ML Research

Tool/Software	Function	Application Context
Morphops [71]	Python library for GM operations	Performing Procrustes alignment, thin-plate spline warping [71]
R Momocs Package [74]	Outline and landmark analysis	Traditional GM analyses, particularly for 2D outlines [74]
geomorph R Package [13]	Collection and analysis of GM shape data	Multivariable shape analysis and visualization [13]
DAVID SLS-2 Scanner [73]	Structured-light 3D scanning	High-resolution 3D model acquisition for cut mark analysis [73]
Global Mapper [73]	Spatial data analysis	Cross-sectional profile extraction from 3D models [73]
Python Scikit-learn [5]	Machine learning algorithms	Implementing SVM, Random Forest, and other classifiers [5]
ImageID Database [15]	Standardized image repository	Consistent imaging protocols for taxonomic comparisons [15]

Future Directions and Emerging Methodologies

The confluence of morphometrics and machine learning continues to evolve, with several emerging methodologies showing particular promise for taxonomic applications. Deep Learning approaches, especially Convolutional Neural Networks (CNNs), have begun to demonstrate superior performance in some classification tasks compared to traditional GM methods [74]. In archaeobotanical seed classification, CNNs significantly outperformed outline-based morphometrics, suggesting that automated feature extraction may surpass even expert-defined landmarks for certain applications [74].

Functional Data Analysis represents another frontier, with FDGM providing enhanced sensitivity to subtle shape variations by treating landmark data as continuous functions rather than discrete points [5]. This approach has shown particular utility for analyzing complex biological structures where important shape information may reside between traditional landmarks.

The integration of 3D Geometric Deep Learning with traditional GM pipelines presents an exciting direction for future research. While current studies have predominantly utilized 2D data, the increasing accessibility of 3D scanning technologies promises to enable more sophisticated analyses of complex morphological structures in their native spatial contexts [73].

As these methodologies continue to develop, the taxonomic community must establish standardized protocols for data sharing, algorithm validation, and performance reporting to ensure reproducibility and comparability across studies. The creation of comprehensive public databases of morphological data, encompassing broad geographic and ecological diversity, will further enhance the power of these integrated approaches to resolve complex taxonomic questions [15].

The integration of Geometric Morphometrics with artificial intelligence and machine learning represents a paradigm shift in taxonomic research, enabling unprecedented precision in species identification, morphological analysis, and evolutionary inference. By combining the rigorous shape quantification of GM with the powerful pattern recognition capabilities of ML, researchers can now address taxonomic challenges that were previously intractable using traditional approaches alone.

The methodologies and case studies presented in this technical guide provide a framework for taxonomists to implement these integrated approaches in their research programs. As the field continues to evolve, the ongoing development of computational tools, algorithmic refinements, and expanded morphological databases will further enhance our ability to extract meaningful biological insights from shape data, ultimately advancing our understanding of biodiversity and evolutionary processes.

Conclusion

Geometric morphometrics has firmly established itself as an indispensable, statistically robust tool in the taxonomist's toolkit, capable of quantifying subtle shape variations that are often invisible to traditional methods. The foundational principles of landmark-based analysis and Procrustes superimposition provide a rigorous framework for comparing biological forms. When combined with a meticulous methodological workflow and thorough validation protocols, GM delivers powerful and reproducible results for species discrimination. The future of GM in biomedical research is particularly promising, with direct applications ranging from identifying disease vectors to informing personalized medicine strategies, such as optimizing intranasal drug delivery based on anatomical variability. As the field progresses, the integration of GM with cutting-edge AI and geometric deep learning promises to further accelerate discovery, enabling the analysis of more complex shapes and the uncovering of deeper biological insights at the intersection of form and function.

References

• 1. Geometric morphometrics on fossil shark teeth [https://palaeo-electronica.org/content/2025/5603-geometric-morphometrics-on-fossil-shark-teeth]
• 2. Noctuidae) to support pest identification in invasive species ... [https://www.frontiersin.org/journals/insect-science/articles/10.3389/finsc.2025.1542467/full]
• 3. When Shape Defines: Geometric Morphometrics Applied to ... [https://www.mdpi.com/1424-2818/17/10/680]
• 4. geometric morphometrics of head and thorax shapes in ... [https://www.frontiersin.org/journals/insect-science/articles/10.3389/finsc.2025.1558242/full]
• 5. Functional data geometric morphometrics with machine ... [https://www.nature.com/articles/s41598-024-66246-z]
• 6. Geometric Morphometrics of Astragalus and Shape Variation ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12023762/]
• 7. Exploring desert-adapted beetles with 3D geometric ... [https://www.nature.com/articles/s41598-025-05967-1]
• 8. Assessing the application of landmark-free morphometrics to ... [https://bmcecolevol.biomedcentral.com/articles/10.1186/s12862-025-02377-9]
• 9. Visualizations in geometric morphometrics: How to read and ... [https://research.manchester.ac.uk/en/publications/visualizations-in-geometric-morphometrics-how-to-read-and-how-to-/]
• 10. Geometric Morphometrics as an Analytical Tool in Plant Biology [https://pmc.ncbi.nlm.nih.gov/articles/PMC11901851/]
• 11. Landmark detection in 2D bioimages for geometric ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5765108/]
• 12. Geometric morphometrics approach for classifying ... [https://www.nature.com/articles/s41598-025-85718-4]
• 13. A practical, step-by-step, guide to taxonomic comparisons ... [https://europeanjournaloftaxonomy.eu/index.php/ejt/article/view/2527]
• 14. Leaf Morphology, Taxonomy and Geometric Morphometrics [https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0025630]
• 15. Shape as a Key to Taxonomy: Morphometric Analysis of ... [https://www.mdpi.com/2075-4450/16/4/386]
• 16. Comparing Spatial Maps of Human Population-Genetic ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC2861313/]
• 17. Procrustes Analysis for High-Dimensional Data | YoungStatS [https://youngstats.github.io/post/2023/03/02/procrustes-analysis-for-high-dimensional-data/]
• 18. Local Superimpositions Facilitate Morphometric Analysis of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8699094/]
• 19. Building and Visualising Ordinations of Shape Data in R ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12178081/]
• 20. Procrustes Analysis [https://john-quensen.com/tutorials/procrustes-analysis/]
• 21. Sizing the Shape: Understanding Morphometrics - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC4347171/]
• 22. Evaluating statistical models for establishing morphometric ... [https://www.sciencedirect.com/science/article/abs/pii/S0305440322000681]
• 23. ACSII II Guide to Geometric Morphometrics in MorphoJ [http://webhome.auburn.edu/~armbrjw/gmguide/Geometric_Morphometrics_Guide/PCA.html]
• 24. Geometric morphometrics and ecological niche modelling for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10064238/]
• 25. A novel approach to geometric morphometrics [https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1009061]
• 26. Geometric morphometrics for the study of facial ... [https://www.nature.com/articles/s41598-019-46330-5]
• 27. Quantitative Analysis of Fish Morphology Through Landmark ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11540051/]
• 28. Geometric morphometrics and ecological niche modelling ... [https://www.sciencedirect.com/science/article/pii/S2667114X23000079]
• 29. Comparison of geometric morphometric outline methods in the ... [https://frontiersinzoology.biomedcentral.com/articles/10.1186/1742-9994-3-15]
• 30. A Step-by-Step Guide for Geometric Morphometrics of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6191499/]
• 31. Geometric Morphometrics - The Polly Lab [https://www.pollylab.org/courses/geometric-morphometrics]
• 32. Multivariate and Geometric Morphometrics Reveal ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12467238/]
• 33. Leveraging machine learning tools and algorithms for ... [https://www.nature.com/articles/s41598-022-11258-w]
• 34. Morphometrics/Statistics/Programming - The Aguirre Lab [https://www.waguirrelab.com/morphometricsstatisticsprogramming.html]
• 35. MorphoJ [https://morphometrics.uk/MorphoJ_page.html]
• 36. 3D software [https://www.sbmorphometrics.org/soft-3d.html]
• 37. Exploring the utility of Geometric Morphometrics to analyse ... [https://www.nature.com/articles/s41598-024-56889-3]
• 38. Shape matters: investigating the utility of geometric ... [https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2024.1380594/full]
• 39. Assessing the application of landmark-free morphometrics to ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12034209/]
• 40. Missing data in multivariate morphometrics [https://jessicaarbour.com/missing-data/]
• 41. The Silent Extinction of Species and Taxonomists—An ... [https://www.mdpi.com/1424-2818/15/10/1053]
• 42. Measurement error in geometric morphometrics [https://link.springer.com/article/10.1007/s00427-016-0537-4]
• 43. a case study of Patagonian lizards of the genus Liolaemus ... [https://bicyt.conicet.gov.ar/fichas/produccion/12004467]
• 44. An example of inter-operator bias in 3D landmarks and its ... [https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0197675]
• 45. A Geometric Morphometrics Approach for Predicting Olfactory ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12565324/]
• 46. Testing inter-observer error under a collaborative research ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9525927/]
• 47. Bolstering geometric morphometrics sample sizes with ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8128762/]
• 48. Sample Size and Geometric Morphometrics Methodology ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10833145/]
• 49. Analyzing Fluctuating Asymmetry with Geometric ... [https://www.mdpi.com/2073-8994/7/2/843]
• 50. Fluctuating Asymmetry - an overview | ScienceDirect Topics [https://www.sciencedirect.com/topics/medicine-and-dentistry/fluctuating-asymmetry]
• 51. ANALYZING PATTERNS OF FLUCTUATING ASYMMETRY ... [https://pubmed.ncbi.nlm.nih.gov/28565401/]
• 52. Bias and error in estimates of mean shape in geometric ... [https://www.sciencedirect.com/science/article/abs/pii/S0047248403000472]
• 53. Optimizing digitalization effort in morphometrics - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC7723759/]
• 54. Geometric Morphometric Data Augmentation Using ... [https://www.mdpi.com/2076-3417/10/24/9133]
• 55. Defining, Reconstructing and Investigating Shape Through a ... [https://journal.caa-international.org/articles/10.5334/jcaa.161]
• 56. Geometric Morphometrics sheds light on the systematics ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10540071/]
• 57. A Step-by-Step Guide for Geometric Morphometrics of ... [https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2018.01433/full]
• 58. Analysis of geometric morphometrics and molecular ... [https://www.nature.com/articles/s41598-023-49536-w]
• 59. Canonical Variate Analysis - MorphoJ User's Guide [https://morphometrics.uk/MorphoJ_guide/comparison/cva.htm]
• 60. A practical, step-by-step, guide to taxonomic comparisons ... [https://europeanjournaloftaxonomy.eu/index.php/ejt/article/view/2529]
• 61. assessing the impact of 2D landmark digitization error in ... [https://link.springer.com/article/10.1007/s00435-025-00717-3]
• 62. A Practical Introduction to Landmark-Based Geometric ... [https://www.cambridge.org/core/journals/the-paleontological-society-papers/article/practical-introduction-to-landmarkbased-geometric-morphometrics/571DF87EC538B4281BB9EDF513247BAE]
• 63. Arvicolinae (Cricetidae, Rodentia, Mammalia) - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC11410883/]
• 64. Morphometric Analysis of Tetropium Species (Coleoptera ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12028269/]
• 65. A deep learning approach for morphological feature ... [https://www.nature.com/articles/s41540-023-00293-6]
• 66. Geometric morphometrics in anthropology [https://en.wikipedia.org/wiki/Geometric_morphometrics_in_anthropology]
• 67. Geometric morphometric analysis of body shape and ... [https://www.sciencedirect.com/science/article/abs/pii/S0940960225002869]
• 68. Linear versus geometric morphometric approaches for the ... [https://pubmed.ncbi.nlm.nih.gov/24740578/]
• 69. Comparison of geometric morphometric outline methods in the ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC1592095/]
• 70. Morphometric analysis of the size-adjusted linear ... [https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-023-09162-2]
• 71. vaipatel/morphops: Geometric Morphometrics in Python [https://github.com/vaipatel/morphops]
• 72. Geometric morphometrics and machine learning as tools ... [https://www.sciencedirect.com/science/article/abs/pii/S1567134821003324]
• 73. Geometric Morphometrics and Machine Learning Models ... [https://www.mdpi.com/2076-3417/13/6/3967]
• 74. Deep Learning Outperforms Geometric Morphometrics for ... [https://archaeo.peercommunityin.org/PCIArchaeology/articles/rec?id=502]