Improving ANN based streamfow estimation models for the Upper Indus Basin using satellite derived snow cover area

The mountainous catchments often witness contrasting regimes and the limited available meteorological network creates uncertainty in both the hydrological data and developed models. To overcome this problem, remotely sensed data could be used in addition to on-ground observations for hydrological forecasting. The fusion of these two types of data gives a better picture and helps to generate adequate hydrological forecasting models. The study aims at the improvement of ANN-based streamfow estimation models by using an integrated data-set containing, the satellite-derived snow cover area (SCA) with on-ground fow observations. For this purpose, SCA of three sub catchments of Upper Indus Basin, namely Gilgit, Astore and Bunji coupled with their respective gauge discharges is used as model inputs. The weekly stream-fow models are developed for infows at Besham Qila located just upstream of Tarbela dam. The data-set for modeling is prepared through normal izing all variables by scaling between 0 and 1. A mathematical tool, Gamma test is applied to fuse the inputs, and a best input combination is selected on the basis of minimum gamma value. A feed forward neural network trained via two layer Broyden Fletcher Goldfarb Shanno algorithm is used for model development. The models are evaluated on the basis of set of performance indicators, namely, Nash–Sutclife Efciency, Root Mean Square Error, Variance and BIAS. A comparative assessment has also been made using these indicators for models developed, through data-set containing gauge discharges, only and the data-set fused with satellite-derived SCA. In particular, the current study concluded that the efciency of ANN-based streamfow estimation models developed for mountainous catchments could be improved by integrating the SCA with the gauge discharges.