Inverse modeling and coupled analysis of soil moisture content and organic carbon in oasis areas of arid regions

Soil moisture content (SMC) and soil organic carbon (SOC) are important environmental factors influencing ecosystem stability and vegetation growth in arid regions. However, the synergistic evolution mechanisms of soil water and carbon, as well as the inversion of deep soil properties, remain insufficiently understood. This study selected the Weigan-Kuqa River Delta Oasis as the study area and integrated multi-source remote sensing data from Sentinel-1 and Landsat with field-measured samples, while combining the Bootstrap-Optimized Shrinkage (BOSS) feature selection algorithm to optimize variable combinations, Multi-layer soil water-carbon inversion models, including convolutional neural networks (CNN), long short-term memory (LSTM) networks, and CNN-LSTM, were constructed, and geographically weighted regression (GWR), bivariate Moran’s I, and coupling coordination degree models were used to analyze the spatial relationships between soil water and carbon.The results indicate that: (1) The CNN-LSTM model achieves the best performance in multi-depth inversion, with the highest accuracy at the 0–10 cm layer (SMC: R 2 =0.72; SOC: R 2 =0.74), and the accuracy gradually decreases with increasing depth; (2) Both SMC and SOC exhibit significant spatial heterogeneity, with SMC increasing and SOC decreasing with soil depth. The spatial distribution of both variables showed a pattern of “higher values in the oasis center and lower values at the edges.” (3) The soil water-carbon relationship exhibited clear vertical differences, with a significant correlation in the surface layer that gradually weakened with depth. The bivariate Moran’s I reached a maximum of 0.83, indicating strong spatial clustering between the two variables; however, the overall coupling coordination level still has room for improvement. This study provides methodological support for multi-depth remote sensing inversion of soil properties in arid regions and enhances the understanding of water-carbon interaction mechanisms in oasis ecosystems.