Abhatoo - Centre Nationale de Documentation

Data availability problems for hydrological and soil studies are undoubtedly a critical constraint for all scientists around the world. This was also our challenge in this thesis, where the study area is poorly documented and devoid of any hydrological study. The first part of this thesis report was devoted to the execution and applicability of the Soil and Water Assessment Tool (SWAT) model to predict runoff and to assess soil erosion rate in three watersheds belonged to Settat - Ben Ahmed region, namely Tamedroust (642.42 km2), Mazer (179.2 km2) and El Himer (177.7 km2). A semi-arid climate and irregular rainfall also characterize this zone. SWAT model inputs were collected and extracted from different sources and simulations were carried out over eight years (January 1995 –December 2002). For soil data, seventy-seven samples were sampled from 0-40 cm depth and analyzed to obtain different soil parameters such as texture, organic matter (OM), soil aggregate stability, pH and electrical conductivity (EC). This soil database (TAMED-SOIL) was compared with the Harmonized World Soil Database (HWSD) to analyze the effects of soil data quality on the SWAT model performance and hydrologic process Tamedroust watershed. Before calibration, results showed a considerable variability and a significant effect of the soil characteristics on the different components of the hydrological cycle. After the calibration period, both soil databases improved the model performance in terms of streamflow, with values of R2 and NSE (Nash-Sutcliffe Efficiency) between 0.64 and 0.65. Model validation was acceptable and similar for both databases with R2 and NSE values of 0.76 and 0.57, respectively. The results also show that all sub-watersheds of Tamedroust present a weak soil erosion rate for both soil databases.
Using the regionalization method between Mazer (gauged watershed) and El Himer (ungauged watershed), SWAT model results showed a good correlation between observed and simulated streamflow with an NSE of 0.65 and 0.89, and with R2 of 0.75 and 0.95 for calibration and validation, respectively. The fitted values for the most sensitive parameters obtained at the Mazer watershed are transferred to the El Himer watershed to estimate streamflow and erosion. The results showed that all studied sub-watersheds present a weak rate of soil erosion.
The last part of this thesis report focuses on the comparison of the capabilities of Multiple Linear Regression (MLR) and a machine learning technique (Random Forest (RF)) to predict soil aggregate stability (SAS) index from Pedotransfer Functions (PTFs) using different sets of input variables (soil properties and remote sensing parameters).
The results obtained were satisfactory for both models. However, the sample size must be increased to ensure more excellent uniformity to predict the SAS index better. Thus, the PTFs developed in this study can be used worldwide as a basis for predicting the soil aggregate stability in another area with the same climatic and edaphic characteristics.