Caught between dimensions: <scp>2D</scp> versus <scp>3D</scp> geometric morphometrics in biodiversity assessment

Geometric morphometrics is widely used to quantify shape variance, yet debate persists over the extent to which two-dimensional (2D) data adequately estimate three-dimensional (3D) morphology. We evaluated the consequences of dimensionality reduction by comparing 2D and 3D landmark datasets derived from the same specimens, imaging methods, and landmark configurations. Using photogrammetric 3D models of museum specimens representing ~91% of the inland fishes of the Lower Mississippi River Basin, we analyzed whole-body shape across the assemblage and within a morphologically constrained clade (Catostomidae). We contrasted bilateral 3D landmark configurations, left-side-only 3D configurations, and 2D projections of the same landmarks. Principal Component Analyses, Mantel tests of Procrustes distances, and axis-by-axis correlations revealed that 2D and 3D datasets recover concordant broad-scale patterns of shape variance. The primary axes of variation (PC1-PC2) were strongly conserved, with specimens retaining similar relative positions in morphospace across dimensionalities. However, correspondence declined across lower-variance PCs, particularly for taxa with pronounced 3D features (e.g., body width), underestimating disparity in 2D analyses. These effects are modest in diverse assemblages but more pronounced within a single family, where fine-scale morphological differences are represented in the higher-variance axes. Our results demonstrate that 2D methods reliably capture dominant gradients of body shape but underrepresent subtle components of shape variation associated with 3D structure that may have functional or biological consequences (e.g., body width). Methodological choice should therefore be guided by research scale and objectives: 2D data are generally suitable for broad comparative surveys or analyses relying on major axes of variance, whereas 3D approaches remain essential for studies of fine-scale disparity, functional morphology, and width-related traits.