Remote Sensing of Soil Organic Carbon at Regional Scale Based on Deep Learning: A Case Study of Agro-Pastoral Ecotone in Northern China

The North China agro–pastoral zone is a large, ecologically fragile in the arid and semi-arid regions. Quantitative remote sensing inversion of soil organic carbon (SOC) this region can facilitate understanding current status degraded land restoration provide data support for cycling research region. Deep learning (DNN) SOC has been W.a hot topic over past decade, but there have few studies at regional scale zones. In study, DNN model with five hidden layers skip connections was established using 644 spatially distributed samples Landsat 8 OLI imagery. compared random forest algorithm terms generalization ability. main conclusions were as follows: 1. establish high-precision (R2 = 0.52, RMSE 0.7), 90% errors concentrated range −2.5 to 2.5 kg·C/m2; 2. Boruta variable-screening effectively improve accuracy algorithm, due DNN’s better ability mine information data, improvement effect on limited; 3. areas are highly positively skewed, significant impact modeling DNN, conversion required obtain ability; 4. regions, weak correlation vegetation indices stronger temperature, elevation, aridity. This study reliable deep density large region, providing reference framework establishment models other