Spatio-temporal modelling of relationship between Organic Carbon Content and Land Use using Deep Learning approach and several covariables: application to the soils of the Beni Mellal in Morocco

Description

In recent decades, population growth has led to rapid urbanisation associated with a land degradation process that threatens soil organic carbon stocks (SOCS). This paper aims to model the interrelationships between SOCS and land use/land cover (LULC). The approach was based on the use of environmental covariates derived from Landsat-5 TM/8 OLI images, forty soil samples, Kriging spatial interpolation method and a Multi-layer Perceptron (MLP) model for the geo-spatialisation of SOCS. The analysis shows a high positive autocorrelation (R 2 >0.75) between vegetation indices and SOCS, particularly higher for SOCS derived from spatial modelling with MLP. On the other hand, the relationship between LULC and SOCS from the three approaches is very variable depending on the dynamics of LULC. The autocorrelations between SOCS and LULC units are very weak in 1985 and 2000 but significant for the year 2018. This suggests that the land use dynamics in the area are favourable to SOCS. In general, the results show that SOCS increased in the tree crop, unused land and forest areas but decreased in the cropland. The SOCS varied in the following order: forest cover>unused land>cropland>urban area>tree crops. This indicates that LULC, topography and vegetation types had an impact on SOCS distribution characteristics.

Abstract

International audience

Additional details