Spatiotemporal Modelling of Soil Organic Carbon Stocks in a Semi-Arid Region Using a Multilayer Perceptron Algorithm

Spatial modelling of soil organic carbon stock (SOCS) and its future dynamics are essential for the sustainable management of terrestrial ecosystems and the planning of carbon sequestration measures. In this study, a spatial modelling approach of the dynamics of the SOCS distribution between 1985 and 2050 and its relationship with land use/land cover (LULC) change in the Béni-Mellal region was accomplished by performing a spatial regression using a multilayer perceptron (MLP) driven by 10 predictors and SOCS data from 40 soil samples. Predictors were extracted from Landsat 5 TM/8 OLI and Sentinel-2 MSI multispectral images and CA-Markov was used for geo-simulations predicting future dynamics. This result shows that the spatial distribution of SOCS and its temporal dynamics in terms of positive and negative variations are strongly linked to spatiotemporal changes in LULC. Over the period 1985–2018, the results showed both progressive variations in the soils of tree crops, unused land and soils in urban areas, slight variations in forest soils and significantly regressive variations in the soils of cropland (− 606 kg.106). The future dynamics from 2018 to 2050 suggest a very significant positive evolution of the SOCS in forest soils with a rate of change of 35.6 kg.106, while the regressive evolution of the SOCS in cropland should continue at − 73.1 kg.106. Furthermore, the spatial autocorrelation results suggest that the spatial distribution of LULC units, topography and vegetation indices are the main factors influencing the quantitative distribution of SOCS in the study area, with correlations ranging from 0.8 to 0.94.