The Use of Artificial Neural Network and Advanced Statistics to Model Sediment Yield on a Large Scale: Example of Morocco

Abstract

Morocco ranks among countries with the greatest achievements in the field of dams in Africa but is affected by the sedimentation phenomenon due to soil erosion in upstreams. The assessment of Sediment Yield (SY) and Suspended Sediment Yield (SSY) remains a challenging global issue, especially in Morocco, characterized by a great diversity of morphological, climatic, and vegetation cover. The main objective of this paper was to perform advanced statistics and artificial neural networks (ANN) in order to understand the spatial distribution of sediment yield and the factors most controlling it, including factors of the RUSLE model (Revised Universal Soil Loss Equation). In order to produce a model able to assess SY, we collected and analyzed extensive data of most variables that can be affecting SY using 42 catchments of the biggest and important dams of Morocco. Statistical analysis of the studied watersheds shows that SY is mainly related to the watershed area and the length of the drainage network.  On the other hand, the SSY is higher in watersheds where gully erosion is abundant and lower in areas with no soil horizon. The SSY is mainly related to the altitude, aridity index, sand fraction, and drainage network length. In front of the complexity of preserving this phenomenon, the ANN was applied and gave very good satisfactory results in predicting the SSY (NSE=0.93, R2=0.93).