DEVELOPMENT OF STOCHASTIC MODELS FOR THREE SUB-CATCHMENTS OF TEHRI DAM

Abstract

Streamflow forecasting is a crucial step in many of the activities related to planning, management and operation components of water resources systems. Streamflow forecasting is important to the water resources system managers for making proper allocations of water to hydropower generation, irrigation, domestic and other uses on day to day basis. In recent times, due to the effect of changing the climate, the job of water managers has become more important and risky. In a country like in India, where the rainfall occurs mainly during the south-west monsoon months (June to September), the storage and proper utilization of water is a basic need. The development of a proper inflow forecasting system can be very useful for suitable utilization of storage waters. The forecasting of streamflow could be done for short-term as well as for long term basis. In this research, the short term duration of one day has been used for the development of forecasting models. The main aim of the present study is to develop the stochastic models for three sub-catchments of the Tehri dam. Tehri dam was constructed on the confluence point of Bhagirathi and Bhilangana river, which are one of the sources of great Ganges river of India. The dam is built for multipurpose use. It is the main source of water supply for the Ganga canal and millions of people are dependent on the water supply from the Tehri reservoir. Therefore, the proper utilization of the storage water from the dam is very important for the people living in the command area of the canals which are receiving water from the dam. To fulfil the objective, at first, the rating curves have been developed for two sub-basins, namely Bhilangana and Balganga of Tehri catchment using method of least squares and ANN technique. Following this, the stochastic models have been developed for three main sub-catchments of Tehri dam. The results of the stochastic models have been compared with the results of HEC-HMS. For developing the stage-discharge relationships, the data set of 1st June 2016 to 30th November 2018 from two gauging stations, namely Ghansali in Bhilangana river and Sarasgaon in Balganga river have been used. The performance of both the methods have been evaluated using Nash Sutcliffe Efficiency (NSE) and the coefficient of determination (R2). The results of the analysis show the good performance of both methods. For the method of least squares, the NSE was more than 95% and the coefficient of determination was more than 0.9. However, the efficiency of the ANN method was slightly better than the method of least squares. The RMSE was far less in the case of ANN. VIII Stochastic models have been developed for three main sub-catchments of Tehri dam, namely Bhagirathi at MBII, Bhilangana at Ghanshali, and Balganaga at Sarasgaon. In the present study four stochastic models namely Autoregressive (AR) model, Autoregressive models with exogenous inputs (ARX), Autoregressive moving average (ARMA) model, and Autoregressive moving average model with exogenous inputs (ARMAX) have been developed and used for daily streamflow forecasting purpose for monsoon and non-monsoon seasons. The rainfall and discharge data from June 2016 to May 15, 2019, for the three sub-basins, namely Bhagirathi at MB II, Bhilangana at Ghansali and Balganga at Sarasgaon were collected from Real-time inflow forecasting system website of Tehri dam. All the developed models were calibrated and validated by dividing the data into two parts. The performance of all the developed stochastic models has been checked using 6 indices namely NSE, RMSE, PBIAS%, R2, MAE and AIC. The comparison of the results of stochastic with and HEC-HMS model results shows that the performance of selected stochastic models is far better than the HEC-HMS model for the three sites of the Tehri catchment during calibration and validation period. The programs have also been prepared using R-studio version 3.4.3 for the simulation of daily streamflow by stochastic models. The recommendations made on the basis of the study and scope for future work are listed below:  The stage-discharge relationship was drawn only using the data from 2016 to 2018, which may not cover the higher flood records and therefore, during the floods, the developed relationship may give lesser value than actual. For this, the relationship could be redrawn in future by using more dataset and a new relationship can be drawn only for flood situation i.e. for higher values of the flood stages.  In case of the stochastic model, only AR model was developed for non-monsoon season. In future, development of other stochastic models considering the rainfall and temperature are expected to give better results.  More efforts are required to be put in for increasing the efficiency of the HEC-HMS model with extended data bases. With extended data base, the efficiency of HEC-HMS is expected to improve further.  The updating of parameters of stochastic models on a daily basis is recommended in future work.