Representing South Indian water tanks in a hydrologic model using remote sensing data

Water tanks in South India have undeniable impacts on the natural hydrological regime of rivers by storing water during the monsoon season and releasing it for irrigation purposes during the dry season. As data on water tanks is limited, they are often not considered in hydrological modeling, which could reduce model performance. Therefore, this study aims at representing water tanks in the hydrologic model SWAT+ and evaluating their impacts on the model performance for a catchment model of the upper Adyar River catchment in South India. To obtain data on the spatio-temporal variations in water storage of the tanks for the years 2015-2018 a random forest classification of water areas is carried out using Sentinel-2 satellite data. Two model setups are evaluated, one with and another one without water tanks. A multi-metric approach including the Kling–Gupta efficiency (KGE), the Nash-Sutcliffe efficiency (NSE), and the ratio of the root mean square error to the standard deviation (RSR) was applied to calibrate and validate the hydrologic model for the time periods 2012-2018 and 2004-2011 respectively. The water tanks are considered as reservoirs in the hydrologic model and the required data such as the location, the surface area, and the volume of reservoirs are extracted from the satellite data. Our results show that implementing water tanks in the SWAT+ model leads to a better representation of the monthly streamflow by having an effect on the peak flows of the wet season. A higher goodness of fit is achieved for the validation period with KGE = 0.67, NSE = 0.76, and RSR = 0.62 in comparison to the calibration period where KGE and NSE are 0.56 and 0.61, respectively. The agreement between simulated and observed streamflow is the highest for the period 2015-2018 (KGE = 0.76, NSE = 0.81 and RSR = 0.43). Therefore, it can be concluded that implementing water tanks in a hydrologic model enhances the performance of the model.