Fig. 2
Spatial prediction of chlorophyll content and model comparison using hyperspectral imaging and machine learning. a, Schematic of the workflow for applying machine learning models to predict spatial distributions of leaf traits and microbiome diversity indices, and for conducting trait–microbiome cross-correlation analysis. b-d, Predicted chlorophyll (Chl) content [µg/cm2 ] across a sunlit Quercus robur leaf collected in late August 2024, based on models trained on September sunlit spectra (PLSR: R2 = 0.74, Stacking: R2 = 0.76) (b) Spatial predictions from the PLSR model (c) Spatial predictions from the stacking model (d) Difference map between PLSR and stacking predictions (PLSR minus stacking). Red regions indicate higher Chl values predicted by the PLSR model, blue regions indicate higher predictions by the stacking model, and white regions indicate minimal difference. e, RGB image of the same leaf, providing a visual reference for structural and pigmentation features. f, g, Observed vs. predicted Chl content for (f) the PLSR model and (g) the stacking model. Data points are grouped by chlorophyll concentration: low (< 30 µg/cm2 ; red dashed fit), medium ( 30–37.5 µg/cm2; green dashed fit), and high (> 37.5 µg/cm2; blue dashed fit). Orange dashed line indicates the ideal 1:1 relationship.
