Streamflow Prediction Upstream of a Dam Using SWAT and Assessment of the Impact of Land Use Spatial Resolution on Model Performance

This paper aims to examine the sensitivity of streamflow prediction of SWAT model to the spatial resolution of land use data. The land use quality impact was studied with a low-resolution global dataset (ESA-CCI with a 300-m resolution) and a refined dataset classified based on Sentinel-2 satellite images (20-m resolution). Two models were successfully built under ArcSWAT, and SWAT-CUP software, which were calibrated and validated after sensitivity and uncertainty analyses. The measured precipitation in the Bine El Ouidane dam watershed was used for model calibration and validation. Snowmelt parameters were first calibrated before analyzing the sensitivity of flow parameters and calibrating the most sensitive ones. The performance of the calibration and validation was very satisfactory for both models based on statistical coefficients and graphical evaluation in addition to parameter uncertainty of the 95 PPU band. The results revealed slightly higher Nash-Sutcliffe efficiency coefficient, root mean square error and R-2 values for the SWAT-CCI model than those for the SWAT-Sentinel-2 model (in calibration, NSE was approximately 0.85 for the first model and 0.76 for the second model, whereas in validation, NSE was 0.70 and 0.61, respectively). This study demonstrated that the streamflow assessment accuracy obtained with the SWAT model was not enhanced when applying finer-spatial resolution land use data and that the accuracy was not directly related to the increase in the number of delineated hydrological response units (HRUs).