Hydrol. Earth Syst. Sci., 23, 1409–1429, 2019 https://doi.org/10.5194/hess-23-1409-2019 © Author(s) 2019. This work is distributed under the Creative Commons Attribution 4.0 License. Attributing the 2017 Bangladesh floods from meteorological and hydrological perspectives Sjoukje Philip1, Sarah Sparrow2, Sarah F. Kew1, Karin van der Wiel1, Niko Wanders3,4, Roop Singh5, Ahmadul Hassan5, Khaled Mohammed2, Hammad Javid2,6, Karsten Haustein6, Friederike E. L. Otto6, Feyera Hirpa7, Ruksana H. Rimi6, A. K. M. Saiful Islam8, David C. H. Wallom2, and Geert Jan van Oldenborgh1 1Royal Netherlands Meteorological Institute (KNMI), De Bilt, the Netherlands 2Oxford e-Research Centre, Department of Engineering Science, University of Oxford, Oxford, UK 3Department of Physical Geography, Utrecht University, Utrecht, the Netherlands 4Department of Civil and Environmental Engineering, Princeton University, Princeton, NJ, USA 5Red Cross Red Crescent Climate Centre, The Hague, the Netherlands 6Environmental Change Institute, Oxford University Centre for the Environment, Oxford, UK 7School of Geography and the Environment, University of Oxford, Oxford, UK 8Institute of Water and Flood Management, Bangladesh University of Engineering and Technology, Dhaka, Bangladesh Correspondence: Sjoukje Philip ([email protected]) and Geert Jan van Oldenborgh ([email protected]) Received: 10 July 2018 – Discussion started: 23 July 2018 Revised: 14 February 2019 – Accepted: 14 February 2019 – Published: 13 March 2019 Abstract. In August 2017 Bangladesh faced one of its worst change in discharge towards higher values is somewhat less river flooding events in recent history. This paper presents, uncertain than in precipitation, but the 95 % confidence inter- for the first time, an attribution of this precipitation-induced vals still encompass no change in risk. Extending the analy- flooding to anthropogenic climate change from a combined sis to the future, all models project an increase in probability meteorological and hydrological perspective. Experiments of extreme events at 2 ◦C global heating since pre-industrial were conducted with three observational datasets and two times, becoming more than 1.7 times more likely for high 10- climate models to estimate changes in the extreme 10-day day precipitation and being more likely by a factor of about precipitation event frequency over the Brahmaputra basin up 1.5 for discharge. Our best estimate on the trend in flooding to the present and, additionally, an outlook to 2 ◦C warming events similar to the Brahmaputra event of August 2017 is since pre-industrial times. The precipitation fields were then derived by synthesizing the observational and model results: used as meteorological input for four different hydrological we find the change in risk to be greater than 1 and of a similar models to estimate the corresponding changes in river dis- order of magnitude (between 1 and 2) for both the meteoro- charge, allowing for comparison between approaches and for logical and hydrological approach. This study shows that, for the robustness of the attribution results to be assessed. precipitation-induced flooding events, investigating changes In all three observational precipitation datasets the climate in precipitation is useful, either as an alternative when hydro- change trends for extreme precipitation similar to that ob- logical models are not available or as an additional measure served in August 2017 are not significant, however in two out to confirm qualitative conclusions. Besides this, it highlights of three series, the sign of this insignificant trend is positive. the importance of using multiple models in attribution stud- One climate model ensemble shows a significant positive in- ies, particularly where the climate change signal is not strong fluence of anthropogenic climate change, whereas the other relative to natural variability or is confounded by other fac- large ensemble model simulates a cancellation between the tors such as aerosols. increase due to greenhouse gases (GHGs) and a decrease due to sulfate aerosols. Considering discharge rather than precip- itation, the hydrological models show that attribution of the Published by Copernicus Publications on behalf of the European Geosciences Union. 1410 S. Philip et al.: Attributing the 2017 Bangladesh floods 1 Introduction 30 years to an event occurring once in 100 years, depend- ing on the data source: water level and discharge data at In August 2017 Bangladesh faced one of the worst river Bahadurabad (the main station for discharge representing flooding events in recent history, with record high water lev- the Brahmaputra in Bangladesh) and the flooding forecast els, and the Ministry of Disaster Management and Relief re- system GloFAS. These estimates, however, were implicitly ported that the floods were the worst in at least 40 years. based on the assumption of a stationary climate and did not Due to heavy local rainfall, as well as water flow from account for the possibility that the frequency of such flooding the upstream hills in India, the various rivers in northern events may be changing. Bangladesh burst their banks. This led to the inundation of Extreme rainfall events that subsequently lead to river basin areas in the northern parts of Bangladesh, starting widespread flooding, such as the 2017 event in Bangladesh, on 12 August and affecting over 30 districts. The National are one of the main types of extreme weather events that we Disaster Response Coordination Centre (NDRCC) reported are expecting to see more of in a warming climate. But with that around 6.9 million people were affected, with 114 peo- rainfall not only being driven by thermodynamic processes ple reported dead and at least 297 250 people displaced. Ap- but also being affected by changing atmospheric processes, proximately 593 250 houses were destroyed, leaving families it is not clear a priori if such events at a particular location displaced in temporary shelters. will increase in likelihood or if the dynamic changes will Bangladesh is a highly flood-prone country, with flat to- mean that the overall chance of extreme rainfall decreases pography and many rivers that regularly flood and are used to there (Otto et al., 2016). Furthermore, in the current climate, irrigate crops and for fishing. The August 2017 floods were drivers other than greenhouse gases (GHGs) often play a particularly impactful as they followed two earlier flooding role that is currently difficult to quantify but likely to mask episodes in late March and July that year, increasing the vul- or exacerbate the effect of greenhouse-gas emissions so far nerability of people. Nearly 85 % of the rural population in on the occurrence likelihood of extreme rainfall events (e.g. Bangladesh works directly or indirectly with agriculture, and aerosols, van Oldenborgh et al., 2016). Hence regional attri- rice is the main staple food, contributing to 95 % of total bution studies are necessary for identifying whether and to food production. As is typical after such flooding, farmers what extent extreme rainfall events are changing and for pro- started to plant aman, the monsoon rice that is almost en- viding insight into which drivers have been contributing to tirely rain dependent. However, the August flood was worse those changes and whether the trend is likely to continue into than that of July, and areas such as Dinajpur and Rangpur that the future. Attribution studies require both observational data normally do not flood were also flooded (see Fig.1). These and models to fully estimate the impact of changes in the cli- are areas that contain significant rice production. As a result, mate system. The reported advances in model development 650 000 ha of croplands were severely damaged during the for the Brahmaputra region and their success in forecasting August monsoon flooding in the year. Aman rice is histori- gives good confidence in the models’ ability to accurately cally the most variable, and yields tend to drop dramatically represent the region. during major flood years (Yu et al., 2010). The flood-induced Hydrological models are increasingly used for studies on crop losses in 2017 resulted in the record price of rice, nega- flooding in Bangladesh. As upstream flow data are absent for tively affecting livelihood and food security. Beyond impacts Bangladesh, a lot of effort has been made to develop flood to agriculture, the floods destroyed transport infrastructure forecasting systems based on satellite data and weather pre- such as railways lines, bridges and roads, leaving some areas dictions. Webster et al.(2010), for instance, developed a sys- inaccessible to disaster relief efforts. The rise in water and tem that forecasts the Ganges and Brahmaputra discharge strong current breached roads and embankments and swept into Bangladesh in real time on 1-day to 10-day time hori- away livestock, houses and assets that may have otherwise zons. In a recent study Priya et al.(2017) show that, by using been protected. At least 2292 schools were damaged, affect- a new long lead flood forecasting scheme for the Ganges– ing education for weeks, and 13 035 cases of waterborne ill- Brahmaputra–Meghna basin, skillful forecasts are provided nesses were reported in the aftermath of the floods. that inherently not only express a prediction of future water The 2017 flood was markedly different from previous ma- levels but also supply information on the levels of confidence jor flood events in 1988 and 1998, when both the Ganges and with each forecast. Hirpa et al.(2016) used reforecasts to im- Brahmaputra flooded simultaneously (Webster et al., 2010). prove the flood detection skill of forecasts. Based on forecasts it was feared that a similar event would Previous scientific studies generally show an increasing occur in 2017, but in this case, the swelling of the Brahmapu- trend in climate projections of extreme rainfall and high dis- tra; its tributary, the Atrai; and the Meghna caused flooding. charge in the region. For example, Gain et al.(2011) use the The worst impacts were along the main reach of the Brahma- PCR-GLOBWB model with input from 12 global circula- putra River (Fig.1b).