A Simple Streamflow Forecasting Scheme For The Ganges Basin.
Jiang, Yudan.
2017-04-24T15:10:20.272Z
-
Abstract: This thesis develops two statistical models to predict Ganges river discharge at Hardinge Bridge in Bangladesh near India/Bangladesh border using only the previous local flow and upstream rainfall. Current operational flood forecasting in Bangladesh provides no more than 3-day lead-time forecast, matching the time it takes water in the Ganges to flow through Bangladesh, mainly due to the... read moreunreleased river flow information from India. The upstream rainfall used in this thesis is derived from freely available gridded precipitation data estimated from observed daily precipitation from the rain-gauge-observation network. The Ganges is a highly seasonal river with high discharge during July to October, and for other time the monthly mean discharge is smaller than one eighth of the critical flood level at Hardinge Bridge. Historical discharge data shows that almost all floods at Hardinge Bridge happened in August and September during the peak flow season, which is one month lagged in respect to peak season of upstream rainfall. As a result, the selected forecasting period is August-September. The first model, termed Q--Q, utilizes the persistence of previous stream flow (Q) to make predictions of future discharge. The model is able to provide promising forecasting performance within a 3-day horizon. The second model, termed Q+P--Q, combines upstream rainfall (P) with previous values of discharge to predict river flow. Results show that the Q--Q model has comparable forecast ability to Q+P--Q model within 5-day horizon, but Q+P--Q model begins to show advantage beyond 5-day horizon. For 10-day lead-time forecast, Q+P--Q model can explain 20% more variance in river discharge than Q--Q model when no rainfall forecast is included. The utility of rainfall forecast in flow predictions is also examined, model performances of 10-day lead-time flow forecast are compared when different lead-time rainfall forecasts are incorporated, and results suggest that 8-day rainfall forecast yields the best predicting capability while 6-day rainfall forecast is able to generate comparable model result. This thesis also assesses the impact of basin scale on flow forecasting capability. Two more rivers of different drainage area are studied, Mississippi River at Clinton and Des Moines River at Keosauqua. Experiments indicate that flow forecasting capability deteriorates with the decreasing catchment size. In addition, this thesis compares hybrid model, which was proposed by Nash and Barsi (1983) and particularly developed for highly seasonal rivers, with the proposed two models in their performances. Results show that for 10-day lead-time flow forecast, Q+P--Q model outperforms hybrid model during August and September, but hybrid model provides better forecast ability than Q--Q model for whole-year forecasting.
Thesis (M.S.)--Tufts University, 2013.
Submitted to the Dept. of Civil Engineering.
Advisor: Shafiqul Islam.
Committee: Shafiqul Islam, David Small, and Levine Stephen.
Keywords: Hydrologic sciences, Water resources management, and Statistics.read less - ID:
- 2r36v850w
- To Cite:
- DCA Citation Guide EndNote
