Soft Computing–Based Workable Flood Forecasting Model for Ayeyarwady River Basin of Myanmar

Abstract

It is a challenging task for working hydrologists of Myanmar to get information from all gauge and discharge sites during a flood to model the forecast properly. In such a case, the concept of this work is very useful for real-time flood forecasting, particularly when data of all the gauge sites are not available regularly or timely. In that context, one has to rely on some accessible sites to get a workable forecast. Additionally, the best combination of the available data can be selected for making the flood forecast. The study is done for the establishment of a flood forecasting model with maximum efficiency using very little information. Three upstream sites named as Sagaing, Monywa, and Chauk of the Ayeyarwady river are selected as the base station and the downstream Pyay as the forecasting station in this study. The artificial neural network (ANN) multilayered feed forward (MLFF) network along with the Takagi-Sugeno (TS) fuzzy inference model are applied in this paper. The developed model is used to forecast the stage from 1 to 4 days in advance. The values of three performance evaluation criteria, namely the efficiency, the root-mean-square error (RMSE), and the coefficient of correlation, were found to be very good and consistent. The results of ANN and fuzzy models remain at par, but the fuzzy model remains somewhat better than the ANN model. It is determined that for stage forecasting at Pyay, preferably the stage at Sagaing-Monywa-Chauk, Sagaing-Monywa, or Sagaing-Chauk is necessary on a priority basis. Regarding the influence of base stations on forecasting, Chauk remains the best, followed by Sagaing and Monywa. The fuzzy model performs better than the ANN model when the case of peak modeling comes. The study provides a best combination of available data for workable flood forecasting with sufficient lead time for planning and operating relief measures.