Remotely Sensed Variables Predict Grassland Diversity Better at Scales Below 1,000 km as Opposed to Abiotic Variables That Predict It Better at Larger Scales

Global spatial patterns of vascular plant diversity have been mapped at coarse grain based on climate-dominated environment–diversity relationships and, where possible, at finer grain using remote sensing. However, for grasslands with their small plant sizes, the limited availability of vegetation plot data has caused large uncertainties in fine-grained mapping of species diversity. Here we used vegetation survey data from 1,609 field sites (>4,000 plots of 1 m²), remotely sensed data (ecosystem productivity and phenology, habitat heterogeneity, functional traits and spectral diversity), and abiotic data (water- and energy-related, characterizing climate-dominated environment) together with machine learning and spatial autoregressive models to predict and map grassland species richness per 100 m² across the Mongolian Plateau at 500 m resolution. Combining all variables yielded a predictive accuracy of 69% compared with 64% using remotely sensed variables or 65% using abiotic variables alone. Among remotely sensed variables, functional traits showed the highest predictive power (55%) in species richness estimation, followed by productivity and phenology (48%), spectral diversity (48%) and habitat heterogeneity (48%). When considering spatial autocorrelation, remotely sensed variables explained 52% and abiotic variables explained 41%. Moreover, Remotely sensed variables provided better prediction at smaller grain size (<∼1,000 km), while water- and energy-dominated macro-environment variables were the most important drivers and dominated the effects of remotely sensed variables on diversity patterns at macro-scale (>∼1,000 km). These findings indicate that while remotely sensed vegetation characteristics and climate-dominated macro-environment provide similar predictions for mapping grassland plant species richness, they offer complementary explanations across broad spatial scales.