Grassland is a crucial component of the global ecosystem, essential for maintaining biodiversity, regulating climate, and stabilizing soil. The Mongolian Plateau, with its extensive grasslands, serves as a key region influencing the global carbon cycle. Therefore, accurately estimating above- and below-ground biomass carbon is critical. This study selects the Ordos Grassland on the Mongolian Plateau as a case study to evaluate the performance of five machine-learning models, including random forest (RF), support vector machine (SVM), decision tree (DT), k -nearest neighbors (KNN), and backpropagation neural network (BPNN), in predicting biomass carbon using key remote sensing variables as predictors. RF model exhibited excellent capability and stability in predicting both above-ground biomass carbon (AGBC) and below-ground biomass carbon (BGBC), outperforming SVM, DT, KNN, and BPNN models. The total AGBC and BGBC of Ordos Grassland were approximately 16.40×10 5 and 83.88×10 5 Mg C, respectively, with mean carbon densities of 116.94 and 575.44 g C/m 2 , respectively. Natural grasslands contributed 89.00% of the total carbon storage. Spatial analysis showed significant heterogeneity, with higher carbon storage values in the eastern region decreasing westward. Temporal analysis from 2019 to 2023 indicated that 36.00% of pixels showed increasing AGBC trends and 39.00% showed increasing BGBC trends. Climate correlation analysis revealed that precipitation was the primary controlling factor for AGBC through positive correlation, while temperature was the most significant factor affecting BGBC through negative correlation, with distinct spatial variations across different climatic conditions. These findings not only provide a scientific basis for the sustainable management of grassland ecosystems on the Mongolian Plateau, but also offer critical data support for regional carbon neutrality strategies and global climate change mitigation. Furthermore, the optimized machine-learning framework established in this study can be extended to large-scale grassland biomass carbon estimation in other similar ecosystems worldwide.
Fan et al. (Fri,) studied this question.