Deep Learning and ANN for Forecasting of Flood in Satluj River at Kasol, Himachal Pradesh, (India)

This is the authors' abstract. We don't add key points for this paper.

In the last few decades, a significant increase in of high intensity rainfalls in India, is witnessed, leading to floods in the affecting catchment areas. This has been a motivation for researchers to develop early and accurate flood forecasting techniques to prevent the losses to the human lives, agriculture, communication networks and infrastructure etc. In the present work, one such technique has been developed for prediction of flood using the supervised statistical modeling of Rainfall-Runoff records of Satluj River, India. Statistical modeling for flood forecasting, has been proven to be an effective technique in multiple literatures, as with its use, typical hydrological relationships among the variables for any catchment area is not required to be known. Deep Learning or Deep Neural Networks (DNN) is much suitable technique for analyzing big data of high variety so suits well for flood forecasting, which makes use of continuous large data received from multiple sensors. Auto Correlation (ACF), Partial Correlation (P ACF) and Cross Correlation (CCF) of the rainfall data have been calculated for selection of the Input Vector (IV). The lead-times of 1 to 7 days are taken for performance evaluation along with the statistical indices-correlation coefficient, percentage error in peak flow estimation and root mean squared Error (RMSE). The results of deep learning (DNN) forecasting models depict quite significant improvement in the performance in terms of accuracy and precision as compared to some other conventional and simple techniques used in hydrological researches. The DNN model performance observes deterioration with increased lead time.