Assessing and modelling soil organic carbon dynamics in a mining spill restoration area

Soil organic carbon (SOC) is a key indicator of soil quality and an essential component in climate change mitigation. Its monitoring faces limitations when based solely on field data, which drives the search for complementary methodologies such as remote sensing and simulation models. The aim of this study was to assess the potential of integrating remote sensing–derived information into the estimation and modeling of SOC dynamics in the Guadiamar Green Corridor (Seville, Spain), an area undergoing restoration following the 1998 mining spill. Two methodological approaches were employed at landscape and sublandscape level: (i) the spatial prediction of SOC and clay content using a Random Forest (RF) model trained with Sentinel-2 spectral variables, and (ii) the simulation of SOC dynamics with the RothC model under seven boundary conditions (BC0–BC6), in which field-measured variables were progressively replaced by proxies obtained from remote sensing.

The Random Forest model showed moderate performance (R² ≈ 0.47 in training and validation), displaying spatial coherence between areas with higher clay content and higher SOC levels. In the case of RothC, except for BC1, all simulations reproduced a decreasing trend in SOC but did not reach the magnitude of loss observed in the field. Scenario BC2, which simulated with clay percentage data obtained through RF, showed the greatest similarity to the reference scenario (BC0), while BC5, based on remote sensing–derived potential evapotranspiration data, generated a marked underestimation of final SOC, highlighting the model’s sensitivity to this parameter. The results suggest that remote sensing is a valuable tool to complement field measurements in SOC modeling, especially in contexts with limited data availability. However, accuracy depends on the variable being substituted and on model calibration to the specific conditions of ecological restoration.