Modelling soil organic carbon stock distribution across different land-uses in South Africa: A remote sensing and deep learning approach

Abstract

Soil organic carbon (SOC) is a critical measure for ecosystem health and offers opportunities to understand carbon fluxes and associated implications. However, SOC can be significantly influenced by anthropogenic land use change, with intensive and extensive disturbances resulting in considerable SOC loss. Consequently, understanding the spatial distribution of SOC across different land uses, particularly at national level characterised by different biomes, is vital for integrated land-use planning and climate change mitigation. Remote sensing and deep learning (DL) offer a reliable largescale mapping of SOC by leveraging on their big data provision and powerful analytical prowess, respectively. This study modelled SOC stocks across South Africa’s major land uses using Deep Neural Networks (DNN) and Sentinel-3 satellite data. Based on 1936 soil samples and 31 spectral predictors, results show a relatively high accuracy with an R2 and RMSE value of 0.685 and 10.15 t/h (26% of the mean), respectively. From the seven land uses evaluated, grasslands (31.36%) contributed the most to the overall SOC stocks while urban vegetation (0.04%) contributed the least. Moreover, although SOC stock was found to be relatively proportional to land coverage, commercial (46.06 t/h) and natural (44.34 t/h) forests showed a higher carbon sequestration capacity. These findings provide an important guideline to managing SOC stocks in South Africa, useful in climate change mitigation through sustainable land-use practices. Whereas landscape restoration, and other relevant interventions are encouraged to improve SOC storage, care must be taken within land-use decision making to maintain an appropriate balance between carbon sequestration, biodiversity, and general ecosystem functions.