Uhe, P. F., Mitchell, D. M., Bates, P. D., Sampson, C. C., Smith, A.M., & Islam, A. S. (2019). Enhanced flood risk with 1.5 °c globalwarming in the Ganges-Brahmaputra-Meghna basin. EnvironmentalResearch Letters, 14(7), [074031]. https://doi.org/10.1088/1748-9326/ab10ee

Publisher's PDF, also known as Version of recordLicense (if available):CC BYLink to published version (if available):10.1088/1748-9326/ab10ee

Link to publication record in Explore Bristol ResearchPDF-document

This is the final published version of the article (version of record). It first appeared online via IOP Publishing athttps://iopscience.iop.org/article/10.1088/1748-9326/ab10ee. Please refer to any applicable terms of use of thepublisher.

University of Bristol - Explore Bristol ResearchGeneral rights

This document is made available in accordance with publisher policies. Please cite only thepublished version using the reference above. Full terms of use are available:http://www.bristol.ac.uk/pure/user-guides/explore-bristol-research/ebr-terms/

Environmental Research Letters

LETTER • OPEN ACCESS

Enhanced flood risk with 1.5 °C global warming in theGanges–Brahmaputra–Meghna basinTo cite this article: P F Uhe et al 2019 Environ. Res. Lett. 14 074031

 

View the article online for updates and enhancements.

This content was downloaded from IP address 137.222.190.205 on 15/08/2019 at 14:40

Environ. Res. Lett. 14 (2019) 074031 https://doi.org/10.1088/1748-9326/ab10ee

LETTER

Enhanced flood risk with 1.5°C global warming in theGanges–Brahmaputra–Meghna basin

PFUhe1 , DMMitchell1 , PDBates1 , CC Sampson2, AMSmith2 andAS Islam3

1 School of Geographical Sciences, University of Bristol, United Kingdom2 Fathom, Engine Shed, StationApproach, Bristol, United Kingdom3 Institute ofWater and FloodManagement, BangladeshUniversity of Engineering andTechnology, Dhaka, Bangladesh

E-mail: Peter.Uhe@bristol.ac.uk

Keywords:flooding, climate change, Ganges–Brahmaputra–Meghna, 1.5°Cglobal warming

Supplementarymaterial for this article is available online

AbstractFlood hazard is a global problem, but regions such as southAsia, where people’s livelihoods are highlydependent onwater resources, can be affected disproportionally. The 2017monsoon flooding intheGanges–Brahmaputra–Meghna (GBM) basin, with record river levels observed, resulted in∼1200 deaths, and dramatic loss of crops and infrastructure. The recent Paris Agreement calledfor research into impacts avoided by stabilizing climate at 1.5 °Cover 2 °Cglobal warming abovepre-industrial conditions. Climatemodel scenarios representing thesewarming levels were combinedwith a high-resolution flood hazardmodel over theGBMregion. The simulations of 1.5 °Cand 2 °Cwarming indicate an increase in extreme precipitation and corresponding flood hazard over theGBMbasin compared to the current climate. So, for example, evenwith global warming limited to 1.5 °C,for extreme precipitation events such as the southAsian crisis in 2017 there is a detectable increase inthe likelihood inflooding. The additional∼0.6 °Cwarming needed to take us from current climate to1.5 °Chighlights the changed flood risk evenwith low levels of warming.

1. Introduction

In many regions around the globe, climate change isincreasing the severity of damaging flooding events[1–4]. Flooding in large rivers such as the Ganges–Brahmaputra–Meghna (GBM) system can affect mil-lions of people through damage to property, crops andlivestock and risks to life. Globally, climate change isexpected to result in more rainfall, due to the ability ofa warmer atmosphere to hold more water [5].However, changes to local and regional rainfall are alsoimpacted by a number of factors such as topography,atmospheric composition (e.g. aerosols) [6, 7], land-use change [8], ocean currents and atmosphericcirculation. So when evaluating the risk of severestorms and flooding, it is critical to look at changes onregional scales.

The recent United Nations Framework Conven-tion of Climate Change agreement in Paris has com-mitted to restricting warming levels to well below 2 °Cand aiming for 1.5 °C above pre-industrial levels [9].

There has recently been a concerted effort to runclimate simulations designed to inform us of theimpacts of 1.5 °C and 2 °C warming. Two initiativeshave designed climate simulations to represent 1.5 °Cand 2 °C of global warming: (1) the Half a degreeAdditional warming, Prognosis and Projected Impacts(HAPPI) project [10], includes many atmospheric-only simulations using super-ensembles ( 100> ) andmultiple models to give a range of possible climateresponses. (2) The ‘CESMLowWarming’ project [11],uses a single coupled atmosphere-ocean model toachieve climates stabilized at 1.5 °C or 2 °C of globalwarming, thereby providing a more complete sampleof ocean variability than HAPPI, but at the expense ofsmaller sample sizes. These simulations give decisionmakers more targeted information than more generalinitiatives, about the benefits of restricting the level ofglobal warming [12].

This study investigates flooding in the GBM riversystem which covers a wide area, through Bangladesh,Bhutan, China, India and Nepal. Rainfall over the

OPEN ACCESS

RECEIVED

26March 2018

REVISED

13March 2019

ACCEPTED FOR PUBLICATION

19March 2019

PUBLISHED

16 July 2019

Original content from thisworkmay be used underthe terms of the CreativeCommonsAttribution 3.0licence.

Any further distribution ofthis workmustmaintainattribution to theauthor(s) and the title ofthework, journal citationandDOI.

© 2019TheAuthor(s). Published by IOPPublishing Ltd

GBM system is dominated by the monsoon season(June–September). Around 80% of Bangladesh isfloodplain, with floods affecting tens of millions ofpeople occurring every six years or so [13]. Any ampli-fication of flood hazard may have grave implicationsfor the vulnerable and exposed populations in theseregions. Previous studies for the GBM have predictedincreases in future peak river discharge [14–18] andflood extent [19]. Recent studies [3, 20–22], looked atflood risk at 1.5 °C and 2 °C of global warming, forchanges to peak discharge of the Brahmaputra or glo-bal flood risk including the Ganges–Brahmaputra.However, these used high emissions scenarios todetermine 1.5 °Cand 2 °Cof global warming.

The precipitation response for a given level ofwarming can differ between low and high emissionsscenarios [12] and between transient and stabilized cli-mates. It has been found that short duration extremerainfall is constrained by the amount of global warm-ing [23], however thismay vary due to aerosols in areaswith high levels of pollution [24]. Additionally, in theGBM, we consider longer duration extremes whichnot may be as directly constrained by temperature.The targeted low warming scenarios are designed torepresent a stabilized climate, and avoid complicationsintroduced by determining specific warming levelsfrom transient simulations rising to higher levels ofwarming. The low warming scenarios also have loweraerosol levels than present day [25], as projected forthe end of the 21st century. These aerosol levels willdiffer significantly from those in time slices at 1.5 °Cand 2 °C, from high emissions scenarios, as these willoccur earlier in the 21st century. So results from theseexperimental designs will differ where aerosols play animportant role.

Determining flood impacts requires a nonlineartransformation of river discharge using a hydro-dynamic model because the floodplain topographyand channel-floodplain hydraulic interactions inclu-ded in such schemes may either amplify or dampenthe flooding response to changing discharge. Wetherefore extend previous studies by analyzing floodinundation using a high-resolution hydraulic modelto represent the possible change in flood risk, based onthese lowwarming scenarios.

2.Methods and datasets

2.1.HAPPI atmospheric simulationsSimulations were used from the HAPPI project:atmosphere-only climate simulations, forced by sea-surface temperatures (SSTs), sea-ice concentration(SIC) and green-house gas concentrations. SSTs andSIC were from the OSTIA observational dataset [26]for current day (Hist) simulations (2006–2015), andSSTs from the Coupled Model IntercomparisonProject Phase 5 (CMIP5 [27]) output were used toestimate the future scenarios corresponding to 1.5 °C

and 2 °C global warming above pre-industrial condi-tions, at the end of the 21st century [10]. Largeensembles were produced by running simulationswith different initial condition perturbations. Sevenmodels were used for this analysis (table S2), and mostof themodels had around 100 simulations or more foreach of the scenarios (table S3).

2.2. CESM-CAM5 lowwarming simulationsSimulations using the CESM-CAM5 coupled climatemodel were designed using specific GHG concentra-tion pathways, to stabilize temperatures at 1.5 °C and2 °C global warming above pre-industrial conditionsby 2100 [11]. These simulations cover 2006–2100, andare continuation runs of 11 CESM-CAM5 historicalsimulations (1920–2005) [28] run as per the CMIP5design [27]. For current day climate, the 2006–2015decade from the 2 °C simulations was analyzed, tomatch the time period of the HAPPI current daysimulations. The 2090–2099 decade was analyzed forthe future scenarios.

2.3. Climate data analysisArea averages over the GBM river basin were calcu-lated from climate model outputs. The basin defini-tions were identified based on theHydroBasins dataset[29]. For each model, years from different ensemblemembers were pooled for analysis, resulting in a largenumber of years representing the historical, 1.5 °C and2 °C worlds in each model (10 years per simulationmultiplied by number of ensemble members as pertable S3). Values such as ensemble means werecalculated across the distributions of years and ensem-ble members. Observational datasets were used formodel evaluation (see table S4).

We primarily analyzed the yearly maximum ofmonthly rainfall (RXmonthly). This is because thecharacteristic duration of rainfall event resulting in thelargest flooding events in the GBM is on the order of amonth or longer, so this variable was chosen as a proxyfor the change in river discharge in the GBM (seesection 2.5). We also looked at yearly average pre-cipitation and yearly maximum of 5 day mean rainfall(RXx5day), which is a standard climate change indexdefined by the Expert TeamonClimate ChangeDetec-tion and Indices [30, 31]. RXx5day represents extremerainfall connected to flooding in small catchments ortributaries. We validated the seasonal cycle of pre-cipitation and monsoon winds in the region for eachof themodels.

The climate models have precipitation biases,(compare ‘OBS/Reanalysis’ with ‘Hist’ in figures 1(a)and (b)), with the majority of models over-predictingthe peak rainfall. Some of the models also tend tosimulate an early monsoon onset compared toobserved (figure 2). We additionally note that theobservational datasets have limitations. There are dif-ferences between precipitation observation datasets

2

Environ. Res. Lett. 14 (2019) 074031

(figures 1(a) and (b)), and it has also been suggestedthat high-altitude precipitation (such as over theHimalayas), may be significantly underestimated inobservations, due to poor coverage of stations andunderestimation of solid precipitation [32, 33].

Our projection of future changes in flooding isbased on precipitation changes in the climate models,rather than the absolute conditions in those models.The response of precipitation to climate change pre-dicted by themodels is based on physical mechanisms,

Figure 1.Distributions of precipitation averaged over theGBMbasin. (a) Shows boxplots of RXmonthly for each of themodels andalso observations. (b) as per (a), but showing RXx5day, (c) percentage changes in ensemblemeanRXmonthly, between the differentscenarios ‘1.5 °C—Hist’, ‘2 °C—Hist’, and ‘2 °C–1.5 °C’. Error bars show the 5%–95% range of sampling uncertainty in the ensemblemean change, based on randomly resampling each distribution 1000 times. Color indicates additionalmeasures of significance: Redsymbols indicate the distributions of simulated years are not distinguishable between the two scenarios compared, based on a twosidedKolmogorov–Smirnov test at p=0.05. For the other colors, there is a detectable difference between the distributions.Additionally, blue and yellow symbols give an idea of themagnitude of change relative to year-to-year variability. Blue or yellowsymbols indicate whether the proportion of ensemblemembers which change in the same direction as themean is greater or less than67%. The first panel of (c) shows themulti-model summarywhich does not used color to indicate significance. OBS/Reanalysisdatasets are defined in table S4.

Figure 2. Seasonal cycle of precipitation averaged over theGBMbasin. Eachmodel is shown in a separate panel. Two observationalproducts (CRU-TS andGPCC), defined in table S4, are shown in the left-most panels.

3

Environ. Res. Lett. 14 (2019) 074031

which may represent the relative increase or decreasein precipitation even in the presence of biases. Weconcentrate our analysis on the basin scale, to reducethe influence of small scale effects, which the modelshave difficulty representing. Bias-correction, can beapplied to remove biases in the mean (and variability),but should leave the relative change in precipitationunchanged. However, this is not always the case forchanges in extremes [34], so bias-correction methodsneed to be applied with caution. We analyze scalingfactors based on the relative change in precipitation,and do not apply bias correction for this study.

As the Asian monsoon drives the majority of pre-cipitation over the GBM region, it is important thatthe climate models represent the monsoon circula-tion. Separate analysis including most of the HAPPImodels used here has been made for the Asian mon-soon precipitation [35, 36] and specifically for mon-soon onset and length [37]. These studies determinedthat the HAPPI models capture the Asian monsooncirculation sufficiently to investigate future changes inprecipitation. Reference [35] showed an increase inboth intensity and frequency of extreme precipitationin the region, and increases in particularly damagingpersistent rainfall extremes in northern India. An eva-luation of the large scale 850hPawinds for eachmodel,and its change between scenarios, is shown in figure 3.The models have varying biases in the monsoonwinds, although MIROC5, with the highest precipita-tion, has a notable strengthening of the monsoonwinds relative to the ERA-Interim reanalysis. The pat-terns of change in the monsoon circulation varybetween between 1.5 °C—‘Hist’, and 2 °C–1.5 °C, andvary between models, which is consistent with [36],who additionally concludes that the precipitationchange is dominated by the thermodynamic responseand changes related to circulation aremore uncertain.

2.4. Flood hazardmodel simulationsFlood hazard was estimated by the use of the Bristolglobal flood model [38]. In this modeling framework,calculations of flood extent are performed with animplementation of the well-known LISFLOOD-FPflood inundation model [39]. LISFLOOD-FP is ahydraulic model, solving the 2-dimensional shallowwater equations. This configuration of the model uses arecently published bare-earth version of the ShuttleRadar Topography Mission (SRTM) global elevationdatabase (MERIT DEM [40]), and global river andcatchment hydrography from HydroSHEDS [29] todetermine catchment areas and channel locations. Itapplies a Regional Flood Frequency Analyses (FFA) [41]using global data for river discharge (Global RunoffDataCentre (GRDC) dataset) and rainfall. In this approach,river hydrographs for locations not included in theGRDC dataset were estimated based on distributionsfrom rivers with similar characteristics. For this study,the FFA based on global data was adjusted using gauge

data from the Ganges and Brahmaputra rivers torepresent the local dischargemore accurately.

A limitation of the MERIT DEM used by thehydraulic model, is that it does not include flooddefenses. For example the western banks of the Brah-maputra are protected by embankments from Chil-mari to Sirajganj, so flood extents simulated by themodel will differ with observations in these areas.

This model has previously undergone extensivevalidation for catchments in the UK and Canada [38],and in the USA [42]. They found that the model per-formance approaches the skill expected by modelsbuilt with high quality local data, and that the modelperforms better for wet regions and rivers with largercatchments [42]. The improved MERIT DEM datasetused in this study may also improve the model perfor-mance (compared to SRTM topography used in [38]and [42]). We evaluate the performance of the modelalong a stretch of the Brahmaputra river in section 3.2.

For this paper, new simulations were performedover the region 21–31N, 84–94E. The hydraulic modelsimulated flood inundation at 30 arc second (∼900 mat the equator) resolution, which was downscaled tothe MERIT DEM at 3 arc second resolution. Themodel simulates fluvial flooding in catchments above∼50 km2. For this study, flood hazard for return peri-ods of 1 in 5, 1 in 20 and 1 in 100 years were calculated.This covers a large portion of the GBM basin, includ-ing the whole of Bangladesh. Firstly a ‘baseline’ simu-lation was calculated using observed distributions ofcurrent river discharge and rainfall, and secondlysimulations were run with scaled river discharge torepresent changes due to global warming. Changes inflood area and flood depth, between the baseline andfuture scenarios, were analyzed over the region andadditionally over the sub-basins Ganges, Brahmaputraand Meghna (regions shown in figure S1 is availableonline at stacks.iop.org/ERL/14/074031/mmedia).

2.5. Scaling discharge from future scenariosTo determine scaling factors based on the globalwarming scenarios, the change in RXmonthly wasused as a proxy for the change in river discharge. Thiswas chosen as we are interested in the peak flows, andin the GBM system the highest flood waters due to themonsoon rains build up over a period of at least amonth. Thismetric represents the change in precipita-tion driving flooding events, however we note that thisdoes not take into account, evaporation or catchmenthydrology, which may cause discharge to scale differ-ently to the change in precipitation. This may also varyby return period, for example table 3 in [20], howeverthese numbers have large uncertainty ranges (e.g.figure 5 in [22]). Furthermore, a comparison ofprecipitation and runoff for models used in this study,show consistent changes between these two variables(figure S3). As discharge results from the accumulationof runoff, this gives us confidence in our proxy as an

4

Environ. Res. Lett. 14 (2019) 074031

approximation of future change. The advantage of thissimple approach is that it does not rely on potentiallyproblematic bias correction methods, or hydrologicalmodeling thatmay be under-constrained due to sparseobservations and hence introduce a greater level ofuncertainty.

We determined the percentage change inRXmonthly, between the current day climate and1.5 °C and 2 °C worlds, averaged over the GBM basin.The ensemble mean was calculated separately for each

of the 8 models considered (7 HAPPI models andCESM). A weighted average taking into account thesampling uncertainty and model spread was used toproduce a best estimate (referred to as the Multi-Model Summary, see text S3). The best estimates were7.0%±3.6% and 10.7%±4.7%. The best (medium)estimates along with the upper (high) and lowerboundwere used to scale the baseline/present day pre-cipitation and river discharge in the hydraulic modelfor the future scenarios.

Figure 3. Figures showing JJA average 850 hPawinds. (a)ERA-Interimwinds and differences between eachmodel and ERA-Interimfor 2006–2015 period. (b)Difference between 1.5 °Cand ‘Hist’ scenarios for eachmodel. (c)Difference between 2 °Cand 1.5 °C foreachmodel. All data was interpolated to a common 2° horizontal resolution grid before plotting.

5

Environ. Res. Lett. 14 (2019) 074031

3. The 2017flooding event in theGBMbasin

High rainfall in the monsoon season in South Asiacaused particularly severe flooding in 2017. Thisflooding was reported in the media to have killed over1200 people in India, Nepal and Bangladesh, andmillions were evacuated or otherwise affected [43, 44].In Bangladesh, reports indicated that over 6.1 millionpeople were affected with a death toll of at least 134[43, 45]. The high impact of this event makes it arelevant case study to evaluate the skill of the hydraulicmodel in simulating inundation extent from a recentevent. The following section describes the peak level ofthe Brahmaputra river in August 2017, and a compar-ison showing the skill of modeled inundation extentagainst satellite observations.

3.1. Peak river levelmeasured at BahadurabadOn the 16th of August, the Brahmaputra recorded arecord high level of 20.84 m at Bahadurabad. How-ever, an analysis of river discharge at this stationshows that despite being a record river level for thisparticular gauging site, the discharge for this event(78 500 m3 s−1, measured on the 17th of August) wasonly a 1 in 5 year return period flow (95% confidenceinterval 3–9 years).

Other than the uncertainty in discharge measure-ments, the discrepancy between the return periods forriver level and discharge at Bahadurabad is most likelyexplained by very local changes in the river channelsfrom sedimentation and construction of floodembankments. The relationship between dischargeand river height is a local effect around the gauge as aresult of erosion and deposition as mobile sedimentwaves move through the system. However at the reachscale that we consider here these local variations willcancel out and our overall estimates of the impact ofincreasing flows on inundation extent will be reason-able ones. The discharge return period for this event isused for the following comparison, as discharge is usedto drive the hydraulicmodel.

3.2. Representation of 2017 event in thefloodhazardmodelBecause ground based measurements of flood extentdo not exist at the resolution of the flood model, wecompared flood extent from the model with estimatesfrom two satellite products: (1) Copernicus Sentinel-1Synthetic Aperture Radar (SAR) data and (2) the JointResearch Centre Global Surface Water (GSW) dataset[46], which is produced from Landsat imagery. TheSentinel data at ∼10 m resolution was processed andwater bodies were detected based on the backscatteramplitude (supplementary text S1). SAR products canpenetrate clouds, so the Sentinel-1 data can give asnapshot of inundation extent. The GSW dataset

provides a flood recurrence product at ∼25 m, basedonmany images over the 1984–2015 period.

The 1 in 5 year modeled hazard was comparedover a region downstream of the Bahadurabad gauge(89–90E 24.4–25.2N, figure S2). We compared thisagainst a single Sentinel-1 image from 22 August 2017.This is not exactly like-for-like, as the floodmodel usesa 1 in 5 year discharge everywhere, whereas the actualflow in different river segments will be greater or lowerdepending on local conditions. We also comparedagainst flood recurrence greater than 20% (5 years) intheGSWdataset.

Detection of flood extent in satellite data is notexact, both false positives and missed detection arepossible. For example, in SAR data, smooth surfacessuch as roads or mud flats may in some cases be classi-fied as water, water roughened by wind may be classi-fied as land, and flooded areas covered by vegetationmay not be detected due to backscatter effects. Topo-graphical shadow effects may cause false positives aswell. Some of these effects, such as topographical sha-dows, can be corrected by the image processing, butthis is still not a perfect representation of the groundsituation. The GSW dataset will also not detect bodiesof water obscured by vegetation, and persistent cloudcover may cause short duration flooding events to bemissed.

Figures 4(a)–(b) compares model flooded regionswith satellite data, at the satellite’s resolution. Thefloodmodel captures a large proportion of the compli-cated braided river structure and flood plain.Figure 4(a) shows the modeled fluvial flood extentagainst the Sentinel data for 2017. As only ∼7% of thecatchment areas reside in Bangladesh, we expect thefluvial flooding to drive the majority of the flooding.On the flood plain to the east of the river, the modeledfluvial flooding under-predicts the Sentinel floodextent (figure 4(a)). Figure 4(b) shows the modeledfluvial flooding against the GSW flooding. The modelshows better agreement with the GSWdataset than theSentinel image. This may be either to do with the nat-ure of the different instrumentation and processingpicking up different types of flooding or because thethe GSWdata is a recurrence product whichmore clo-sely represents the 1 in 5 year maximum extent as themodel does, compared to the Sentinel data which is asnapshot offlooding on 22August 2017.

In the region to the west of the river, the model isover-predicting the observed event. This is probablybecause the western bank of the Brahmaputra river isprotected by an embankment that is not representedin theMERITDEM and therefore not in themodel. Inaddition, the hydraulic model simulates a 1 in 5 dis-charge in all river segments in the domain, so thiscomparison is less valid in other parts of our domainwhichmay have a different return period for 2017.

Metrics representing this comparison were calcu-lated over this region. Using the Critical Success Index(see [42], supplementary text S2) the model fluvial

6

Environ. Res. Lett. 14 (2019) 074031

flood extent has a score of 0.47 for GSW and 0.36 forSentinel. This shows lower performance than shownin [42] and [38]. As matching the higher resolutionsatellite data is a very difficult test we also calculatedthe absolute fractional error between data aggregatedto larger scales (figure 4(d)), with errors of 0.22 (Senti-nel) and 0.16 (GSW) at 100 m, reducing to 0.13 and0.15 respectively at 5 km.

Given this < 15% error in fractional flooded areaat 5 km, we conclude that the model is fit for the thesub-basin scale relative change analysis which we per-form here. The hydraulic model is a physically basedmodel, and ismass andmomentum conserving. So theevaluation of the 1 in 5 year event gives us confidencethat the relevant physics and the topography is wellrepresented by the model and a greater return period

Figure 4. Satellite data versusmodel: (a) comparison ofmodel 1 in 5 year fluvialflooding versus Sentinel flooding from22August2017. (b)Comparison of 1 in 5 year flood extent,model fluvialflooding versus GSW. (c)Absolute fractional error and bias betweenmodel and satellite flood areas aggregated to different scales.

7

Environ. Res. Lett. 14 (2019) 074031

of river flow will result in a realistic flood inundationfor the 1 in 20 and 1 in 100 year events.

4. Changes inflood risk at 1.5 °Cand 2 °Cwarming

4.1. Simulatedmean and extreme precipitation intheGBMChanges to GBM precipitation are shown forRXmonthly in figure 1(c). All except for one of themodels shows a wetter climate in the future scenarios,and in the 2 °C simulations, greater than two-thirds ofyears are wetter than the mean in the current climate,for all except two of the models. ECHAM6-3-LRshows a small drying to 1.5 °C, but an increase between1.5 °C and 2 °C. Similar but slightly higher changes areseen for RXx5day (figure S4) and lower changes foryearly precipitation (figure S5), showing a greaterwetting change for shorter duration events. Figure 1(c)gives estimates for the ensemble mean change, whichin generalmay differ from the changes of higher returnperiod events. However, in this study, changes fordifferent return periods are consistent within thesampling uncertainty and there is no systematic trendof higher or lower scaling of precipitation at higherreturn periods (figure S6). We additionally note thatusing the ensemble mean may result in underestimat-ing the uncertainty as the sampling uncertaintyincreases for higher return periods.

In all of themodels apart fromCAM4-2degree, the1.5 °C—‘Hist’ change is greater than the 2 °C–1.5 °Cchange. This is partly because the global warmingfrom historical to 1.5 °C is greater (∼0.6 °C) thanbetween 1.5 °C and 2 °C (0.5 °C). Comparing theRXmonthly change per degree of warming (figure S7),half of themodels still show a greater change for 1.5 °C—‘Hist’, however the other models show no-changeor a smaller change, so there are nonlinear changesbetween scenarios which vary between the models.These differences may be due to the removal of sup-pressive effect of rainfall between ‘Hist’ and 1.5 °C,and varying representations of aerosols in the model.The patterns of circulation changes also differ between1.5 °C—‘Hist’ and 2 °C–1.5 °C in all of the models(figure 3), so different mechanisms may be dominat-ing in the differentmodels.

4.2. Changes infloodingFlood hazardmaps were produced using the hydraulicmodel for the baseline period and low, medium andhigh estimates for the 1.5 °C and 2 °C scenarios. Thechanges in flooded area over the region are shown infigures 5(a) and (b) for the medium estimate of the1.5 °C scenario for 1 in 5 and 1 in 100 year floodhazards. A zoomed in section shows the area aroundDhaka. The 1 in 5 year hazard has amuch smallerfloodextent than the 1 in 100 year hazard, but crucially thechange in additional area flooded between 1.5 °C and

the baseline (red regions) is considerably larger for the1 in 5 year hazard than the 1 in 100 year hazard,highlighting the importance of changes in thesefrequently occurring events. The changes to the depthof flood waters for the same simulations are shown infigure 5(c) and (d) with large areas increasing flooddepth by 20 cm and smaller areas increasing by 50 cmor more. The medium 2 °C scenario shows incremen-tally larger changes compared to the 1.5 °C scenario,in bothflood area and flood depth (figure S8).

The changes in flood extent were aggregated overthree sub-regions representing the Ganges, Brahma-putra and Meghna basins. These are shown in figure 6for each of the scenarios and 1 in 5, 1 in 20 and 1 in100 year flood hazard. The more frequent (less severe)flood hazards, show greater percentage increases inflood extent than the more extreme flooding events.This may be expected by the nature of river valleys aslower, flatter areas will flood in the less severe events.However, when the flood waters reach steeper areas inmore extreme events, the relative increase in area willbe less for a given change in water level. The relativechange in the 1 in 5 year flood area (figure 6, blue andred bars) is greater than the corresponding relativeprecipitation change (blue and red shading). There aresmall differences between the basins, but the relativechange in flood area consistently decreases for the 1 in20 and 1 in 100 year floods. The 1 in 100 year floodarea in the Ganges and Brahmaputra rivers show asmaller change than the relative precipitation change.

Figure 6 shows the percentage area relative to thebaseline flood increase, however the absolute change inarea does not show a consistent trend with the returnperiod of the event (figure S9). We also note that the 1in 100 year events experience a greater change in flooddepth, so these particularly extreme events maybecome more destructive due to higher flood waters,evenwithout theflooded area increasing dramatically.

5.Discussion

The Paris Agreement calls to restrict global warming towell below 2 °C and aim for 1.5 °C. The simulationsused here reflect those goals and may give differentresults to evaluating 1.5 °C and 2 °C in high emissionscenarios such as used for CMIP5. The climate modelsemployed here show a significant trend of increasingrainfall in the GBM for all except one of the modelsanalyzed. This trend is stronger for extreme rainfallthan average rainfall which has implications for flood-ing. Even at a low level of global warming of 1.5 °C, thewetting signal in the GBM is clear, and given it isproximity to densely populated regions, this translatesto increased flood risk. There is also a statisticallysignificant increase in monsoon precipitation between1.5 °C and 2 °C despite there being overlap in theuncertainty ranges of changes from present day to1.5 °C and 2 °C (figure 1(c)). This shows there is a clear

8

Environ. Res. Lett. 14 (2019) 074031

benefit in reducing flood risk by keeping temperaturesto the lower target.

The precipitation change between current climateand 1.5 °C is greater than the change between 1.5 °Cand 2 °C. However, due to experimental design, less ofthe models show this trend for change in precipitationper degree of warming. The design of the lowwarming

experiments also results in greater aerosol changebetweenHist and future climates, than between 1.5 °Cand 2 °C, which may not be the case in high emissionscenarios.

In addition to aerosol influence, theremay be non-linear changes in the monsoon circulation in thisregion, which are uncertain based on climate model

Figure 5. Simulated changes in flood extent and depth between 1.5 °Cand the baseline. (a) and (b)Change in flood extent due tofluvialflow, for 1 in 5 year and 1 in 100 year hazards respectively. Regions are separated into ‘Land’ (notflooded), ‘Flooded baseline’and ‘Flooded 1.5 °C’ (additional areas offlooding), and ‘Water body’ (permanent water). (c) and (d)Change in flood depth due tofluvialflow, for 1 in 5 and 1 in 100 year hazards respectively. All panels show the change in the ‘medium’ 1.5 °C scenario compared tothe baseline simulation.

Figure 6.Aggregated changes inflooded area for three sub-basins. Changes inflooded area relative to baseline flooded area for eachscenario, for 1 in 5, 1 in 20 and 1 in 100 year flood hazard. Changes are shown over the three sub-basinsGanges, Brahmaputra andMeghna (figure S1). Themedium estimate is shown as a dot; low andhigh estimates are shown by the extent of the error bars. Theshaded areamarks the range in percentage change in precipitation between the low and high estimates for the 1.5 °C (blue) and 2 °C(red) scenarios.

9

Environ. Res. Lett. 14 (2019) 074031

projections (see figure 3 and [36]). The model meansignal in precipitation may be dominated by the ther-modynamic response, with differences between mod-els due to the representation of changes in themonsoon.

There are advantages and limitations in differentmodeling setups. Coupled atmosphere-ocean models(e.g. CESM-CAM5) sample a wide range of possibleocean variability, and atmosphere only models (e.g.HAPPI) may underestimate this variability [47]. How-ever the use of prescribed SSTs also reduces proble-matic biases present in coupled models [48]. In thisstudy, having consistent projections from both type ofmodels gives us greater confidence that our conclusionsare not skewedby the choice of experimental design.

Increased river flow is expected to have a morenoticeable impact on the flood extent for less extremeevents. This nonlinear response of the flood area to thechange in river flow highlights the importance of thefloodplain topography. Modeling only the change inprecipitation or river discharge may therefore be mis-leading. Using a hydraulic model to map inundationextent is needed to convey the full impacts of climatechange onfloods.

This study applies a scaling factor to river dischargebased on themodeled changes in precipitation. This is asimplification assuming the leading order effects relatedto flood hazard are from the direct precipitationresponse. It also assumes that the shape of distributionsof discharge do not change with global warming, whichwould change the relative magnitudes of floods at dif-ferent return periods. However, any changes in theshape of the distribution are highly uncertain and arenot supported by our climate simulations (figure S6).This approach also does not take into account all of thecatchment hydrology that contributes to river flow andthere are there are influences from changes in temper-ature, glacial melt and rainfall-runoff processes. How-ever, we find that that for the subset of climate modelswhere runoff is calculated, the runoff scales very simi-larly to precipitation (figure S3). There are also uncer-tainties modeling river flow, especially in data sparseregions such as theGBM. So this approach avoids intro-ducing additional uncertainty and complexity, with thecaveat that we are attributing changes in flood risk tochanges in precipitation only and not other catchmenteffects. In addition, the use of RXmonthly for our proxyis most relevant to the downstream river sections asflooding in upstream catchments will have a fasterresponse time. Our analysis showed that for shorterevents (e.g. RXx5day), the climate change influence isstronger, soRXmonthly is a conservative choice.

Another significant contributor to river flow inthis region is glacier melt. In the upper sections of theGanges and Brahmaputra, glacial melt contributesabout 11% and 16% of average runoff respectively (seetable S7 in [49]). For the Upper Brahmaputra, thisincreases to around 20%–25% during months of peakflow (figure S6 in [49]). In the near future, glacier melt

may have a small increase (scenario dependent), how-ever after 2040–2050, glacier melt is projected todecline [49–52]. This is a result of a balance betweenincreased melting rates at warmer temperature andreduced glacier mass. Following this, in the scenariosstabilized at 1.5 °C or 2 °C around the end of the 21stcentury, the contribution of glacier melt to river flowwould be expected to be slightly reduced compared topresent day.

The population in South Asia is highly reliant onwater resources for subsistence agriculture, and isstrongly impacted by floods. We show a clearanthropogenic signal in precipitation change in theGBM basin and a subsequent response in flood areaat 1.5 °C and 2 °C warming. The relative change inflood extent varies with event intensity which isimportant to note for adaptation measures. Thisstudy shows the use of precipitation changes to scaleriver discharge is a justifiable approximation togauge the sensitivity of flood hazard. For future stu-dies, it will be important to investigate a wide rangeof river systems in depth to see what, if any, change isdiscernible due to climate change, even with highlevels ofmitigation.

Acknowledgments

This research used science gateway resources of theNational Energy Research Scientific Computing Cen-ter, a DOEOffice of Science User Facility supported bythe Office of Science of the US Department of Energyunder Contract No. DE-AC02-05CH11231. HAPPIcore support was part funded by Google DeepMind.DM is supported by a NERC Research Fellowship. PBis supported by a LeverhulmeResearch Fellowship anda Royal Society Wolfson Research Merit Award. ASI issupported by a OxfordMartin School Fellowship. TheHAPPI project data is freely available. See http://happimip.org/happi_data/ for details. CESM LowWarming simulations are freely available here:https://doi.org/10.5065/D6RV0MD6. Code for ana-lysis and the floodmodel output can bemade availableon request to corresponding author.

ORCID iDs

PFUhe https://orcid.org/0000-0003-4644-8559DMMitchell https://orcid.org/0000-0002-0117-3486PDBates https://orcid.org/0000-0001-9192-9963A S Islam https://orcid.org/0000-0002-2435-8280

References

[1] Pall P, Aina T, StoneDA, Stott PA,NozawaT,Hilberts AG J,LohmannDandAllenMR2011Anthropogenic greenhousegas contribution toflood risk in England andWales in autumn2000Nature 470 382–7

10

Environ. Res. Lett. 14 (2019) 074031

[2] SchallerN et al 2016Human influence on climate in the 2014southern Englandwinter floods and their impactsNat. Clim.Change 6 627–34

[3] Alfieri L, Bisselink B,Dottori F, NaumannG, de RooA,SalamonP,Wyser K and Feyen L 2017Global projections ofriverflood risk in awarmerworld Earth’s Future 5 171–82

[4] Thober S,KumarR,WandersN,MarxA,PanM,RakovecO,SamaniegoL, Sheffield J,WoodEFandZinkM2018Multi-modelensembleprojectionsof European riverfloods andhighflows at1.5, 2, and3degrees globalwarmingEnviron.Res. Lett.13014003

[5] Clausius R 1850Über die bewegende kraft der wärme und diegesetze, welche sich daraus für die wärmelehre selbst ableitenlassenAnn. Phys., Lpz. 155 368–97

[6] LevyH,Horowitz LW, SchwarzkopfMD,Ming Y,Golaz J-C,NaikV andRamaswamyV2013The roles of aerosol direct andindirect effects in past and future climate change J. Geophys.Res.: Atmos. 118 4521–32

[7] Samset BH, SandM, SmithC J, Bauer S E, Forster PM,Fuglestvedt J S, Osprey S and Schleussner C-F 2018Climateimpacts from a removal of anthropogenic aerosol emissionsGeophys. Res. Lett. 45 1020–9

[8] Luyssaert S et al 2014 Landmanagement and land-coverchange have impacts of similarmagnitude on surfacetemperatureNat. Clim. Change 4 389

[9] UNFCCC, editor. Adoption of the Paris Agreement, volumeFCCC/CP/2015/L.9/Rev. 1, 2015

[10] Mitchell D et al 2017Half a degree additional warming,prognosis and projected impacts (HAPPI): background andexperimental designGeosci.Model Dev. 10 571–83

[11] Sanderson BM et al 2017Community climate simulations toassess avoided impacts in 1.5 and 2 °C futures Earth Syst. Dyn.8 827–47

[12] Mitchell D, James R, Forster PM, Betts RA, ShiogamaH andAllenM2016Realizing the impacts of a 1.5 °CwarmerworldNat. Clim. Change 6 735–7

[13] DewanTH2015 Societal impacts and vulnerability tofloods inBangladesh andNepalWeather Clim. Extremes 7 36–42

[14] Ghosh S andDutta S 2012 Impact of climate change onfloodcharacteristics in Brahmaputra basin using amacro-scaledistributed hydrologicalmodel J. Earth Syst. Sci. 121 637–57

[15] GainAK, ImmerzeelWW, SpernaWeiland FC andBierkensMFP 2011 Impact of climate change on the streamflowof the lower Brahmaputra: trends in high and low flowsbased on discharge-weighted ensemblemodellingHydrol.Earth Syst. Sci. 15 1537–45

[16] AlamS,Mostafa AliM and IslamZ2016 Future streamflowofBrahmaputra river basin under synthetic climate changescenarios J. Hydrol. Eng. 21 05016027

[17] PervezMS andHenebryGM2015Assessing the impacts ofclimate and land use and land cover change on the freshwateravailability in the Brahmaputra River basin J. Hydrol.: Reg.Stud. 3 285–311

[18] MasoodM, Yeh P J-F,HanasakiN andTakeuchi K 2015Modelstudy of the impacts of future climate change on the hydrologyofGanges-Brahmaputra-Meghna basinHydrol. Earth Syst. Sci.19 747–70

[19] MirzaMMQ,Warrick RA and EricksenN J 2003Theimplications of climate change on floods of the ganges,Brahmaputra andMeghnaRivers in BangladeshClim. Change57 287–318

[20] MohammedK, IslamAS, IslamGMT,Alfieri L, Bala SK andKhanM JU2017 Extremeflows andwater availability of thebrahmaputra river under 1.5 and 2°Cglobal warmingscenariosClim. Change 145 159–75

[21] Betts RA et al 2018Changes in climate extremes, freshwateravailability and vulnerability to food insecurity projected at1.5°Cand 2°Cglobal warmingwith a higher-resolution globalclimatemodel Phil. Trans.A 376 20160452

[22] MohammedK, IslamAKMS, IslamGMT,Alfieri L,KhanM JU, Bala S K andDasMK2018 Future floods inBangladesh under 1.5 °C, 2 °C, and 4 °Cglobal warmingscenarios J. Hydrol. Eng. 23 04018050

[23] Pendergrass AG, Lehner F, Sanderson BMandXuY 2015Does extreme precipitation intensity depend on the emissionsscenario?Geophys. Res. Lett. 42 8767–74

[24] vanOldenborghG J, Otto F E L,Haustein K andAchutaRaoK2016The heavy precipitation event of december 2015 inchennai, India [in ‘Explaining Extremes of 2015 from aClimate Perspective’]Bull. Am.Meteorol. Soc. 97 S87–91

[25] BellouinN, Rae J, Jones A, JohnsonC,Haywood J andBoucherO2011Aerosol forcing in the ClimateModelIntercomparison Project (CMIP5) simulations byHadGEM2-ES and the role of ammoniumnitrate J. Geophys. Res.: Atmos.116 D20206

[26] DonlonC J,MartinM, Stark J, Roberts-Jones J, Fiedler E andWimmerW2012TheOperational Sea Surface Temperatureand Sea Ice Analysis (OSTIA) systemRemote Sens. Environ. 116140–58

[27] Taylor KE, Stouffer R L andMeehl GA2012An overview ofCMIP5 and the experiment designBull. Amer.Meteor. Soc. 93485–98

[28] Kay J E et al 2015TheCommunity Earth SystemModel(CESM) Large ensemble project: a community resource forstudying climate change in the presence of internal climatevariabilityBull. Am.Meteorol. Soc. 96 1333–49

[29] Lehner B andGrill G 2013Global river hydrography andnetwork routing: baseline data and new approaches to studytheworld’s large river systemsHydrol. Process. 27 2171–86(www.hydrosheds.org).

[30] Karl T R, Nicholls N and Ghazi A 1999CLIVAR/GCOS/WMOWorkshop on Indices and Indicators for ClimateExtremesWorkshop Summary (Dordrecht: SpringerNetherlands) pp 3–7

[31] Sillmann J, KharinVV, ZhangX, Zwiers FWandBronaughD2013Climate extremes indices in theCMIP5multimodelensemble: Part 1.Model evaluation in the present climateJ. Geophys. Res.: Atmos. 118 1716–33

[32] MénégozM,GalléeH and JacobiHW2013 Precipitation andsnow cover in theHimalaya: from reanalysis to regionalclimate simulationsHydrol. Earth Syst. Sci. 17 3921–36

[33] ImmerzeelWW,WandersN, Lutz A F, Shea JM andBierkensMFP 2015Reconciling high-altitude precipitation inthe upper indus basinwith glaciermass balances and runoffHydrol. Earth Syst. Sci. 19 4673–87

[34] CannonA J, Sobie S R andMurdock TQ2015 Bias correctionof gcmprecipitation by quantilemapping: Howwell domethods preserve changes in quantiles and extremes? J. Clim.28 6938–59

[35] Chevuturi A, KlingamanNP, Turner AG andHannah S 2018Projected changes in the asian-australianmonsoon region in1.5°Cand 2.0°Cglobal-warming scenarios Earth’s Future 145159–75

[36] LeeD,Min S-K, Fischer E, ShiogamaH, Bethke I,Lierhammer L and Scinocca J F 2018 Impacts of half a degreeadditional warming on the asian summermonsoon rainfallcharacteristics Environ. Res. Lett. 13 044033

[37] Saeed F, Bethke I, Fischer E, Legutke S, ShiogamaH,StoneDAand Schleussner C-F 2018Robust changes intropical rainy season length at 1.5°Cand 2°CEnviron. Res. Lett.13 064024

[38] SampsonCC, Smith AM,Bates PD,Neal J C, Alfieri L andFreer J E 2015Ahigh-resolution global flood hazardmodelWater Resour. Res. 51 7358–81

[39] Bates PD,HorrittM S and Fewtrell T J 2010A simple inertialformulation of the shallowwater equations for efficienttwo-dimensional flood inundationmodelling J. Hydrol. 38733–45

[40] YamazakiD, IkeshimaD, Tawatari R, Yamaguchi T,O’Loughlin F,Neal J C, SampsonCC, Kanae S andBates PD2017Ahigh-accuracymap of global terrain elevationsGeophys.Res. Lett. 44 5844–53

[41] Smith A, SampsonC andBates P 2015Regional floodfrequency analysis at the global scaleWater Resour. Res. 51539–53

11

Environ. Res. Lett. 14 (2019) 074031

[42] WingOE J, Bates PD, SampsonCC, Smith AM,JohnsonKAand EricksonTA 2017Validation of a 30 mresolutionflood hazardmodel of the conterminous unitedstatesWater Resour. Res. 53 7968–86

[43] Aljazeera 2017 Floods kill over 1,200 in India, Nepal andBangladesh (http://aljazeera.com/news/2017/08/floods-kill-1200-india-nepal-bangladesh-170826230610924.html)(Accessed: 23October 2017)

[44] TheGuardian 2017While the world’s attention is elsewhere,Bangladesh faces a humanitarian crisis (https://theguardian.com/voluntary-sector-network/2017/sep/12/bangladesh-severe-disaster-flooding) (Accessed: 23October 2017)

[45] Floodlist 2017 Bangladesh—Rivers at RecordHigh, FloodsAffect 1.3Million (http://floodlist.com/asia/bangladesh-floods-august-2017) (Accessed: 20October 2017)

[46] Pekel J-F, CottamA,GorelickN andBelwardA S 2016High-resolutionmapping of global surface water and its long-termchangesNature 540 418

[47] Fischer EM,BeyerleU, SchleussnerCF,KingADandKnuttiR2018BiasedEstimatesofChanges inClimate Extremes FromPrescribedSSTSimulations.Geophys. Lett.458500–09

[48] He J and SodenB J 2016 The impact of SST biases onprojections of anthropogenic climate change: A greater role foratmosphere-onlymodels?Geophys. Res. Lett. 43 7745–50

[49] LutzAF, ImmerzeelWW,ShresthaAB andBierkensMFP2014Consistent increase in highAsia’s runoff due to increasingglaciermelt andprecipitationNat.Clim.Change4 587

[50] Kraaijenbrink PDA, BierkensMFP, Lutz A F andImmerzeelWW2017 Impact of a global temperature rise of1.5 degrees Celsius onAsia‘s glaciersNature 549 257

[51] ImmerzeelWW, Pellicciotti F andBierkensMFP 2013Risingriverflows throughout the twenty-first century in twoHimalayan glacierized watershedsNat. Geosci. 6 742

[52] Shea JM and ImmerzeelWW2016An assessment of basin-scale glaciological and hydrological sensitivities in theHinduKush-HimalayaAnn.Glaciol. 57 308–18

12

Environ. Res. Lett. 14 (2019) 074031