Predictive Modeling of Soil Erosion Using GIS and Artificial Intelligence. Tafna watershed case.

Abstract

Water erosion still presents a significant challenge that leads to the negative change of productivity of arable land located sources of watersheds through loss of soil, fertilizers, nutrients, and organic matter, which imultaneously reduces the longevity of hydraulicstructures downstream, such as dams, due to quickly ccumulating sediments in their reservoirs. Water erosion is triggered by factors that include a combination of physicoclimatic and anthropogenic agents, such as rainfall intensities, terrain morphology, land use, vegetative cover, and soil characteristics, and in the Tafna watershed in northwestern Algeria, all of these factors are particularly effective in the triggering and encouragement of water erosion. The aim of this research is to estimate soil loss by water erosion and to assess areas susceptible to this change in the Tafna watershed. The Revised Universal Soil Loss Equation (RUSLE) was used to estimate soil loss and areas affected, within a Geographic Information System (GIS). The application of the RUSLE model considers five main factors, which are Rainfall Erosivity; Topographical factor (Slope, length, slope steepness); Soil Erodibility; Cover agement/Land cover/ Vegetation; and Anti-Erosion Practices. Their assessment and spatial integration in GIS allowed the simulation of potential annual soil loss and identification of areas with priority issues for anagement. The results indicate that soil loss in the Tafna watershed is as low as 2.92 t/ha/year and as high as 5.87 t/ha/year based upon slope, land use, and rainfall erosivity, indicating a variability of erosion potential. While, also the vast majority of the watershed continues to be below an acceptable erosion threshold (less than 7.0 t/ha), local areas experience considerably higher soil loss; many areas experience soil loss rates over 45 t/ha/year in local hotspots in steep poorly vegetated sub-basins in the watershed catchment. These unresolved critical zones contribute an inordinate share of the sediment yield in the watershed and thus, are indeed priority areas for soil conservation and management. Also, modeling of sediment transport was undertaken using artificial neural networks (ANN) - specifically the Multi Layer Perceptron (MLP) versions in MATLAB - based upon measured suspended sediment concentrations and flow rates. The performance of the models to predict sediment yield were deemed satisfactory, as validated through determination of correlation constant and Nash–Sutcliffe Efficiency (NSE),The correlation coefficient(R) highlighting their utility in forecasting sediment yield