Futuristic Streamflow Prediction Based on Cmip6 Scenarios Using Machine Learning Models

Abstract

Accurate streamflow estimation is vital for effective water resources management, including flood mitigation, drought warning, and reservoir operation. This research assesses the predictive performance of popular machine learning algorithms (LSTM, Regression Tree, AdaBoost, and Gradient Boosting) for daily streamflow forecasting in the Swat River basin. Three key predictor variables (maximum temperature, minimum temperature, and precipitation) are utilized. The study evaluates and compares the effectiveness of ML models (Gradient Boosting, Regression Tree, AdaBoost, and LSTM) during two timeframes (2021–2050 and 2051–2100). Bias-corrected data from ten general circulation models, considering greenhouse gas trajectories (SSP245 and SSP585), are used. Statistical metrics like Coefficient of determination (R2), Mean square Error (MSE), Mean Absolute error (MAE), and Root Mean Square Error (RMSE) are employed for evaluation. Regression Tree exhibits exceptional performance (R2: 0.88 during training, 0.78 during testing). Ensembling Regression Tree, AdaBoost, and Gradient Boosting, future daily streamflow projections are made for SSP245 and SSP585 scenarios. Bias correction enhances reliability, with the ensemble mean indicating an increase in mean annual streamflow between the 2050s and 2080s (3.26–7.52% for SSP245, 3.77–13.55% for SSP585).