Prediction of soil organic carbon stock using digital mapping approach in humid India

There is a need of information on soil organic carbon data for global environmental management and food security. However, the difficulty lies in obtaining soil carbon data, especially in fragile mountainous regions like Northeast India which is complex in nature and difficult to access. The present study aims to model the distribution of soil carbon stock using digital mapping approach, to predict and generate continuous spatially explicit soil carbon map in Northeast India. Firstly, negative exponential depth function has been used to fit the vertical distribution of soil carbon data, and then Random Forest model has been trained and tuned to predict the parameters of the exponential function using climate data and satellite images. The obtained parameters were finally interpolated using ordinary Kriging method and spatial distribution map across the study area has been generated. Results indicate that the negative exponential function fits the data accurately with 94% of data having R2 > 0.7. Land use and topographic factors particularly elevation was found to have the most influence on SOC distribution in Northeast India. The findings from this study indicate good results for the application of this technique to predict and monitor soil carbon of the study area as a function of topographic factors and changes in land use and climate variables. The obtained results can also be connected to global carbon models to improve the understanding of carbon dynamics.