Arid inland river basins are highly sensitive to land--use change, yet the spatially heterogeneous predictors of ecosystem carbon storage remain insufficiently understood. In this study, a locally calibrated InVEST (Integrated Valuation of Ecosystem Services and Trade-offs) model was used to estimate carbon storage in the Hei River Basin for 2010, 2015, 2020, and 2025. A CA–Markov (Cellular Automata-Markov chain Model) model was then used to project the 2030 land-use pattern and associated carbon storage under the assumption that the current policy framework remains unchanged. Random Forest models interpreted with GeoShapley were applied to quantify the nonlinear and spatially varying contributions of seven environmental and anthropogenic predictors. Total modeled carbon storage increased slightly from 5.29181 × 108 t in 2010 to 5.33214 × 108 t in 2020, before declining marginally to 5.32310 × 108 t in 2025. The decline was mainly associated with reduced modeled grassland carbon storage. Carbon storage was generally higher in the southern and eastern parts of the basin, particularly in the Qilian Mountains, and lower in the northern and western desert regions. The whole-basin simulation projected a 4.34% increase by 2030, mainly associated with projected expansion of forestland and grassland. Predictor importance varied spatially: elevation ranked first in the upstream region, NDVI in the midstream oasis, and population density in the downstream desert. NDVI was the most important basin-wide predictor, although its importance partly reflected spatial covariance with land-cover classes. These findings demonstrate the value of integrating carbon-storage modeling, land-use simulation, and spatially explicit model interpretation for differentiated ecosystem management in arid inland basins.
Wang et al. (Sat,) studied this question.