Spatiotemporal patterns and driving factors of net carbon intensity in arid cropland systems: An explainable machine learning study in the Hexi corridor

Arid and semi-arid cropland systems play an important role in food production and terrestrial carbon cycling, but their high dependence on agricultural inputs may intensify trade-offs between productivity and carbon mitigation. This study developed a net carbon intensity (NCI) assessment framework for cropland systems in the Hexi Corridor, a typical arid region of northwestern China, and analyzed the spatiotemporal dynamics and driving factors of county-level NCI from 2001 to 2022. By integrating carbon emissions from agricultural inputs and management activities with crop carbon sequestration, we quantified cropland NCI using multi-source statistical data from 20 counties (cities and districts). Spatial analysis and an explainable machine learning framework (XGBoost–SHAP) were further applied to identify key drivers and their nonlinear effects. Results showed that cropland systems in the Hexi Corridor maintained a stable net carbon sink during the study period, while NCI exhibited an overall fluctuating upward trend (4.28–5.57 t C·ha −1 ). Spatial disparities among counties gradually decreased, accompanied by an expansion of high-NCI areas in oasis agricultural regions. Diesel input intensity (DII), cropland labor density (CLD), and the multiple cropping index (MCI) were identified as the dominant drivers of NCI. SHAP analysis revealed pronounced threshold effects and nonlinear interactions among these factors, suggesting that cropland carbon performance is governed by the coordinated effects of agricultural inputs, production systems, and environmental constraints rather than by individual drivers in isolation. These findings provide scientific support for region-specific low-carbon agricultural management and agricultural decarbonization in arid regions.