Optimizing seasonal discharge predictions: a hybridized approach with AI and non-linear models

Discharge prediction plays a crucial role in hydrology and to reduce the flash flood hazards during monsoon season, it is necessary to build an accurate flow prediction system. Therefore, to forecast the discharge seasonally, data is divided into four seasons as Early Winter Season (EWS), Peak Winter Season (PWS), Monsoon Season (MS) and Summer Spring Season (SSS). For this purpose, a discharge data of 35 years of Beas River at Pandoh is pre-processed by converting the 13,148 daily datasets into seasonal datasets. Also, the time series models are combined with deep learning models to create hybrid models to enhance the predictive accuracy. The predictive accuracy of the proposed models was evaluated using several statistical parameters. The results of the study reveals that the proposed hybrid model SARIMA-ANN perform better as compared to other models for all the four seasons with R² value of 0.9638 (EWS), 0.9061 (MS), 0.9444 (PWS) and 0.9583 (SSS). Considering the accuracy of the proposed SARIMA-ANN model, signifies to be an alternate solution to estimate the discharge in different phases of civil engineering projects.