Machine learning algorithms for streamflow forecasting of Lower Godavari Basin

Abstract

The present study applies three Machine Learning Algorithms, namely, Bi-directional Long Short-Term Memory (Bi-LSTM), Wavelet Neural Network (WNN), and eXtreme Gradient Boosting (XGBoost), to assess their suitability for streamflow projections of the Lower Godavari Basin. Historical data for 39 years of daily rainfall, evapotranspiration, and discharge are used, of which 80% were for the model training and 20% for validation. A Random Search method is used for hyperparameter tuning. XGBoost performs better than WNN, and Bi-LSTM with an R2, RMSE, NSE, and PBIAS of 0.88, 1.48, 0.86, and 29.3% during training, with corresponding values of 0.86, 1.63, 0.85, and 28.5%, respectively, during validation indicate consistency. Therefore, it is used further for projecting streamflow from a climate change perspective. Global Climate Model, Ec-Earth3 is used because of its potentiality, as observed from previous studies. Four Shared Socioeconomic Pathways (SSPs) are considered. Downscaling of future climate variables is based on Empirical Quantile Mapping. Eight decadal streamflow projections are computed – D1 to D8 (2021–2030 to 2091–2099) – exhibiting more pronounced changes within the warming range. They are compared with three historic time horizons of H1 (1982–1994), H2 (1995–2007), and H3 (2008–2020). The highest daily streamflow is observed in D1, D3, D4, D5, and D8 in SSP245; these are D6 and D7 in SSP585 as per XGBoost analysis.