What is Morphometrics?#
Morphometrics provides quantitative representation of morphological traits by extracting shape information as geometric invariants. It offers mathematical tools for modeling morphological properties, bridging the gap between raw digitized data and biologically meaningful phenotypic values.
Form and Shape#
In geometric morphometrics, form and shape have precise mathematical definitions:
Form: Geometric properties invariant to translation and rotation
Shape: Geometric properties invariant to translation, rotation, and scaling
This distinction is fundamental in geometric morphometrics: shape analysis removes size information, allowing comparison of organisms regardless of their absolute dimensions.
Approaches in Morphometrics#
ktch implements two major approaches to morphometric analysis:
Landmark-based Morphometrics#
Landmark-based methods model morphological properties as sets of corresponding points among specimens. Shape is extracted via Generalized Procrustes Analysis (GPA), which removes position, size, and orientation.
Key characteristics:
Requires identification of corresponding points across specimens
Suitable for structures with clearly identifiable anatomical features
Captures local shape information at specific points
Applications:
Grass phytoliths
Leaf shape analysis
Flower morphology
Skeletal morphology
See also
Landmark-based Morphometrics for details on Procrustes methods
Harmonic-based Morphometrics#
Harmonic-based methods describe shape using mathematical functions that capture the outline or surface geometry without requiring point-to-point correspondence between specimens.
Elliptic Fourier Analysis (EFA): Models closed outlines by approximating x and y coordinates as Fourier series, quantifying shapes through normalized Fourier coefficients.
Spherical Harmonic Analysis: Models closed 3D surfaces using spherical harmonic functions.
Applications:
Seed morphology
Leaf outlines
Petal shapes
Fruit shapes
See also
Harmonic-based Morphometrics for details on harmonic methods
Choosing an Approach#
Consideration |
Landmark-based |
Harmonic-based |
|---|---|---|
Data type |
Discrete corresponding points |
Continuous outlines/surfaces |
Homology |
Requires explicit point correspondence |
No explicit point correspondence required |
Automation |
Partially manual |
Highly automatic |
Use landmark-based methods when:
Clear, identifiable anatomical points exist
Biological homology between points is established
Use harmonic-based methods when:
No clear landmarks are available
The outline or surface itself is the feature of interest
The scikit-learn Compatible Workflow#
ktch follows the scikit-learn API design:
from ktch.landmark import GeneralizedProcrustesAnalysis
from ktch.harmonic import EllipticFourierAnalysis
from sklearn.decomposition import PCA
# Landmark workflow
gpa = GeneralizedProcrustesAnalysis()
shapes = gpa.fit_transform(landmarks)
# Harmonic workflow
efa = EllipticFourierAnalysis(n_harmonics=20)
coefficients = efa.fit_transform(outlines)
# Combine with PCA
pca = PCA(n_components=3)
scores = pca.fit_transform(shapes) # or coefficients
See also
Use with scikit-learn Pipeline for practical examples
References#
Noshita, K. (2022). Model-based phenotyping for plant morphometrics. Breeding Science, 72(1), 3-13.