Pankaj KumarSingh, Abhinav Kumar2021-12-242021-12-242021-11https://krishikosh.egranth.ac.in/handle/1/5810179808The prediction of runoff has a significant role in water resource planning and management. There is a great need for good soil and water management system to overcome challenges of water scarcity and other natural adverse events like- floods, landslides, etc. Rainfall-runoff modelling is an appropriate approach for runoff prediction, which makes it possible to take preventive measures to avoid damage caused by natural hazards. In the present study, machine learning techniques namely: Multiple linear regression (MLR), Multiple adaptive regression splines (MARS), Support vector machine (SVM), and Random forest (RF) were used for runoff prediction of the Gola watershed, located in the south-eastern part of the Uttarakhand. The rainfall data for 12 years (2009-2020) of three rain-gauge stations (Nainital, Bhimtal & Kathgodam) and runoff data at the outlet of watershed i.e. Kathgodam were obtained from their respective irrigation departments for the analysis. Thiessen polygon method was used for the calculation of mean areal rainfall of the watershed. Gamma test was conducted to obtain the best inputs for the models. The complete dataset has been divided into training and testing datasets, where 80% of data was used in training and rest 20% was used for the testing period. The goodness of fit for the models was evaluated by root mean square error (RMSE), coefficient of determination (R2), Nash- Sutcliffe coefficient of efficiency (NSE), and percent bias (PBIAS). For runoff prediction, the overall performance-wise rankings of models were RF, MARS, SVM, and MLR. Among all four models, the RF model outperformed in training and testing periods. It can be summarized that the RF model is best-in-class and delivers a strong potential for runoff prediction of the Gola watershed.EnglishThesis