Species Combinations and Sample Size: Optimizing Spectral Reflectance Models for Aquatic Plant Leaf Trait Prediction

Plant functional traits are critical indicators of ecosystem health, yet predicting aquatic leaf traits via spectral reflectance remains challenging due to limited sample sizes and the underrepresentation of rare species. We hypothesized that dominant species’ spectral models could infer rare species’ traits even with constrained data. To test this, we measured leaf reflectance spectra and eleven functional traits across diverse freshwater macrophyte species, developing Partial Least Squares Regression (PLSR) models under varying species combinations (All-families, Dominant-families, Non-Cyperaceae, etc.) and sample sizes (40–240). Results demonstrated that species composition exerted greater influence than sample size on validation accuracy for most traits when samples ranged from 120 to 240. A minimum threshold of 160 samples was identified for robust trait prediction, though model performance diverged significantly between All-families and dominant-family combinations, suggesting dominant taxa alone inadequately represent quadrat-level trait diversity. These findings challenge assumptions that dominant species compensate for rare species’ scarcity in spectral modeling. We advocate prioritizing rare species sampling to enhance model generalizability in wetland ecosystems. This work establishes actionable guidelines for scaling spectral trait prediction in marshes, advancing ecological monitoring and restoration efforts.