Prediction of regional cropland soil organic carbon content and distribution using deep learning: a case study of the...

Soil organic carbon (SOC) is a critical component of soil fertility and plays a significant role in global carbon sequestration. The decline in SOC content across global croplands poses significant challenges to both agricultural productivity and environmental sustainability. Traditional methods for identifying the spatial distribution and driving factors of SOC at medium to large scales are often inefficient due to high costs and limited accessibility of data. This study focuses on the Northeast Plain of China, a region with critical agricultural and ecological importance. By leveraging extensive field data, remote sensing (RS), meteorological, and terrain variables, we developed an ensemble model combining ResNet and Deep Forest algorithms to predict SOC spatial distribution and identify the key drivers of SOC spatial variability. The model was validated using 1000 field-measured samples. The results show the following: 1) the ensemble model (ResNet-Deep Forest) outperformed individual models, with a test-set (20% hold-out) MSE of 0.18 and R² of 0.56. 2) SOC content exhibited a gradual increase from southwest to northeast in the study area, with coastal regions showing lower SOC likely due to salinity and waterlogging issues. 3) Meteorological variables were identified as the most significant drivers of SOC spatial distribution. Lower temperatures and higher precipitation favored SOC accumulation, followed by RS and terrain factors. The study highlights the potential of integrating deep learning with machine learning for large-scale SOC mapping and provides valuable insights into SOC management. The findings offer robust scientific support for improving agricultural soil quality and enhancing carbon sequestration in croplands.