An Integrated Statistical-Machine Learning Approach for Runoff Prediction

Nowadays, great attention has been attributed to the study of runoff and its fluctuation over space time. There is a crucial need for good soil water management system overcome challenges scarcity other natural adverse events like floods landslides, among others. Rainfall–runoff (R-R) modeling an appropriate approach prediction, making it possible take preventive measures avoid damage caused by hazards such as floods. In present study, several data-driven models, namely, multiple linear regression (MLR), adaptive splines (MARS), support vector machine (SVM), random forest (RF), were used rainfall–runoff prediction Gola watershed, located in south-eastern part Uttarakhand. The model analysis was conducted using daily rainfall data 12 years (2009 2020) watershed. first 80% complete train model, remaining 20% testing period. performance models evaluated based on coefficient determination (R2), root mean square error (RMSE), Nash–Sutcliffe efficiency (NSE), percent bias (PBAIS) indices. addition numerical comparison, evaluated. Their performances graphical plotting, i.e., time-series line diagram, scatter plot, violin relative Taylor diagram (TD). comparison results revealed that four heuristic methods gave higher accuracy than MLR model. Among learning RF (RMSE (m3/s), R2, NSE, PBIAS (%) = 6.31, 0.96, 0.94, ?0.20 during training period, respectively, 5.53, 0.95, 0.92, respectively) surpassed MARS, SVM, forecasting all cases studied. outperformed models’ periods. It can be summarized best-in-class delivers strong potential