Key points are not available for this paper at this time.
Soil organic matter (SOM) is a key indicator of cropland quality and carbon cycling. Accurate SOM mapping is essential for sustainable soil management and carbon sink assessment. This study investigated the effects of interannual climatic variability on SOM prediction using remote sensing and machine learning. Youyi Farm in the Sanjiang Plain, Heilongjiang Province, was selected as the study area, covering three representative years: 2019 (flood), 2020 (normal), and 2021 (drought). Based on multi-temporal Sentinel-2 imagery and environmental covariates, Random Forest models were used to evaluate single- and dual-period combinations. Results showed that combining bare-soil and crop-season images consistently improved accuracy, with optimal combinations varying by year (R2 = 0.544–0.609). Incorporating temperature, precipitation, and elevation enhanced model performance, particularly temperature, which contributed most to prediction accuracy. Feature selection further improved model stability and generalization. Spatially, SOM showed a pattern of higher values in the northeast and lower in the central region, shaped by topography and cultivation. This study innovatively integrates interannual climatic variability with remote sensing temporal combination and feature selection, constructing a climate-adaptive SOM mapping framework and providing new insights for accurate inversion of cropland SOM under extreme climates, highlights the importance of multi-temporal imagery, environmental factors, and feature selection for robust SOM mapping under different climatic conditions, providing technical support for long-term cropland quality monitoring.
Dou et al. (Tue,) studied this question.