The record 2017 flood in South Asia: State of prediction and performance of a data-driven requisitely simple forecast model
Document Type
Article
Date of Original Version
10-1-2020
Abstract
In August 2017, South Asia was affected by devastating floods that caused widespread death and destruction in Nepal, India, and Bangladesh. The Brahmaputra River experienced a record flow in Bangladesh. However, the prediction of a 200-year flood from an international forecasting agency did not materialize. In fact, the evolution and dynamics of 2017 flood was very different from previous floods in the region—localized in nature and short-lived in duration. We examine the hydrometeorology of the Ganges, Brahmaputra, and Meghna River basins to show how the 2017 August flood emerged differently from past major floods in the region. Assessment of available rainfall datasets reveals the usefulness of satellite sources, such as TMPA, GPM, ERA-Interim, CPC Global Precipitation, and WRF to capture the extreme events and highlights the poor spatial coverage of in-situ measurements. We then present a data-driven Requisitely Simple (ReqSim) flood forecast model. Findings from the real-time application of ReqSim during the 2017 flood provide encouraging results with accurate predictions of peak flood timing and magnitude for several locations. For example, the ReqSim provides 10-day forecasts with Nash Sutcliffe efficiency (NSE) values 0.89–0.97 and Mean Absolute Errors (MAE) 0.33–0.54 m for the Ganges River system inside Bangladesh. For the Brahmaputra River system, NSE value ranges 0.49–0.68 and MAE 0.32–0.48 m for 10-day forecasts and NSE 0.62–0.78 and MAE 0.36–0.68 m for 7-day forecasts. Comparison of 2017 flood with past major regional floods suggests that—it is not the intensity of rainfall or severity of flood in either the Ganges or the Brahmaputra basin—but the lack of peak flow synchronization of these two rivers may help explain why 2017 flood did not become as severe as the floods of 1998 and 2007 in Bangladesh. Ease of operationalization and reliable forecasting accuracy of ReqSim for 2017 flood is of particular relevance for large rivers, where access to upstream gauge-measured rainfall and flow data are limited, and detailed modeling approaches are operationally prohibitive and functionally ineffective. Plain language summary: Vast areas of South Asia—across the border regions of Nepal, India, and Bangladesh—were flooded by intense rainfall and overflowing rivers in August 2017. This record flood event killed over a thousand people and affected over 40 million people. However, forecasts of a 200-year flood event, issued by an international agency and covered widely in local media outlets, did not materialize. It was found that the evolution and dynamics of 2017 flood was very different from previous major floods in the region—localized in nature and short-lived in duration. We examine the rainfall and flood generating processes before and during the floods and the performance of a simple flood forecast model in detail to identify the mechanisms that led to these apparent false forecasts including the nature and timing of the peak rainfall events and subsequent floods.
Publication Title, e.g., Journal
Journal of Hydrology
Volume
589
Citation/Publisher Attribution
Palash, Wahid, Ali S. Akanda, and Shafiqul Islam. "The record 2017 flood in South Asia: State of prediction and performance of a data-driven requisitely simple forecast model." Journal of Hydrology 589, (2020). doi: 10.1016/j.jhydrol.2020.125190.