Estimation of Pastureland Biomass in the Steppe and Gobi Regions of Mongolia Using Remote Sensing

The aim of this study is to evaluate the performance of random forest (RF) and support vector machine (SVM) models for estimating pastureland biomass in the steppe and Gobi regions of Mongolia using remote sensing (RS)-based spectral variables. Correlation analysis identified several vegetation indices, such as SR, ARVI, NDVI, VARI, and GNDVI as strong predictors of biomass, owing to their sensitivity to vegetation greenness and chlorophyll content. Additional indices, including SAVI, NDWI, NDMI, MSAVI, and EVI, also showed moderate to strong relationships, highlighting the importance of soil background and vegetation moisture in semi-arid environments. In contrast, NDWIM and NDBI exhibited negative correlations, reflecting their association with non-vegetated or low-productivity areas. The results demonstrated that the RF algorithm substantially outperformed SVM, achieving higher accuracy and lower error, while SVM showed comparatively weaker performance. Overall, the findings indicate that integrating RS data with machine learning techniques, particularly RF, provides a reliable and effective approach for estimating and mapping pasture biomass across Mongolia’s diverse landscapes.