LETTER OPEN ACCESS *OREDOLPSOLFDWLRQVRI r&DQG …institutodegeografia.org/wp-content/uploads/2019/10/2018... · 2019-10-28 · Environmental Research Letters LETTER OPEN ACCESS *OREDOLPSOLFDWLRQVRI

Environmental Research Letters

LETTER • OPEN ACCESS

Global implications of 1.5 °C and 2 °C warmerworlds on extreme river flowsTo cite this article: Homero Paltan et al 2018 Environ. Res. Lett. 13 094003

View the article online for updates and enhancements.

Recent citationsIntensified hydroclimatic regime in Koreanbasins under 1.5 and 2°C global warmingJeongBae Kim et al

-

A review of the effects of climate changeon riverine flooding in subtropical andtropical regionsRohan Eccles et al

-

Comparison of extreme temperatureresponse to 0.5 °C additional warmingbetween dry and humid regions overEast–central AsiaMeng Zhang et al

-

This content was downloaded from IP address 192.188.53.214 on 28/10/2019 at 19:22

Environ. Res. Lett. 13 (2018) 094003 https://doi.org/10.1088/1748-9326/aad985

LETTER

Global implications of 1.5°C and 2°C warmer worlds on extremeriver flows

HomeroPaltan1,2 ,MylesAllen3, KarstenHaustein3, Lena Fuldauer3 and SimonDadson1

1 School of Geography and the Environment, University of Oxford,Oxford, United Kingdom2 Instituto deGeografia, Universidad San Francisco deQuito, Quito, Ecuador3 Environmental Change Institute, University ofOxford, Oxford, UnitedKingdom

E-mail: [email protected]

Keywords:Paris Agreement, highflows, climate change

Supplementarymaterial for this article is available online

AbstractTargets agreed to in Paris in 2015 aim to limit global warming to ‘well below 2 °Cand to pursue effortsto limit the temperature increase to 1.5 °Cabove pre-industrial levels’. Despite the far-reachingconsequences of thismulti-lateral climate changemitigation strategy, the implications for global riverflows remain unclear. Here we estimate the impacts of 1.5 °Cversus 2.0 °Cmitigation scenarios onpeakflows by using daily riverflowdata from amulti-model ensemble which follows theHAPPIProtocol (that is specifically designed to simulate these temperature targets).We find agreementbetweenmodels with regard to changing risk of riverflow extremes.Moreover, we find that theresponse at 2.0 °C is not a uniform extension of the response at 1.5°, suggesting a non-linear globalresponse of peak flows to the twomitigation levels. Yet committing to the 2.0 °Cwarming target,rather than 1.5 °C, is projected to lead to an increase in the frequency of occurrence of extreme flows inseveral large catchments. In themost affected areas, predominantly in SouthAsia, while region-specific features such as aerosol loadsmay determine precipitation patterns, we estimate that underour 1.5 °C scenario the historical 1-in-100 yearflowoccurs with a frequency of 1-in-25 years. At2.0 °C, similar increases are observed in several global regions. These shifts are also accompanied bychanges in the duration of rainy seasonswhich influence the occurrence of highflows.

1. Introduction

The Conference of the Parties of the United NationsFramework Convention on Climate Change(UNFCCC) in its Paris Agreement in December 2015agreed to hold ‘the increase in the global averagetemperature to well below 2 °C above pre- industriallevels and to pursue efforts to limit the temperatureincrease to 1.5 °C above pre-industrial levels’. Thesimplicity of these targets led parties to the treaty toembrace a common climate strategy with the aim ofpreventing the risks and impacts associated withunabated climate change (Hulme 2016, Rogeljet al 2016, Schellnhuber et al 2016).

Yet the assumption that a 2 °C limit is a safe bar asfar as climate-change related tipping points are con-cerned, has diverted attention away from the difference

in impacts between worlds in which global tempera-tures are stabilized at 1.5 °C instead of 2.0 °C. Over-looking these thresholds may have importantconsequences given the sensitivity of high flows toradiative forcings (Milly et al 2002). In particular, earlystudies projected an increase in annual runoff and flowpeaks in snow-dominated catchments as a result ofclimate change (Nijssen et al 2001). From here, severalstudies have projected increases in high flow and floodfrequency in Southeast Asia and eastern Africa,although important shifts have also been calculated forother regions such as the northern Andes, North Amer-ica, and eastern Europe (Okazaki 2012, Hirabayashiet al 2013, Dankers et al 2014, Koirala et al 2014). Assuch, the impacts of extreme river flows on humanlives, socio-economic development, and monetary los-ses may be particularly important in a changing climate

OPEN ACCESS

RECEIVED

23November 2017

REVISED

9 July 2018

ACCEPTED FOR PUBLICATION

10August 2018

PUBLISHED

24August 2018

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.

© 2018TheAuthor(s). Published by IOPPublishing Ltd

(Peduzzi et al 2009, Wake 2013, Arnell and Lloyd-Hughes 2014, Jongman et al 2014, Winsemiuset al 2015, Arnell andGosling 2016).

More recently, by downscaling climate projectionsand finding the corresponding year in which variouslevels of atmospheric warming are exceeded, a positiveincrease in future flood risk has been projected (Alfieriet al 2017). While the future increment in flood risks isimportant when compared to its historical baseline,the impacts appear to be more important at 4.0 °C ifcompared to the less evident shifts observed between1.5 °Cand 2.0 °C.

Yet, since climatological biases in current climatemodels experiments, such as the Coupled ModelIntercomparison Project, remain of the order of 0.5 °Cthese efforts may not be sufficient to differentiate therisk associated with these two levels of warming(Hulme 2016, Mitchell et al 2016). For the hydro-logical cycle, in particular, the distribution of globalprecipitation between CMIP scenario experimentsdoes not respond uniformly between CMIP scenarioexperiments due to the changing role of non-CO2 for-cing over the 21st century and also due to the pre-cipitation sensitivity to emission-scenarios. This, inturn, makes it difficult to differentiate whether detec-ted differences result from additional warming oradditional factors. So, extracting anomalies for 1.5 °Cand 2.0 °C levels of warming from the traditionalCMIP scenarios driven experiments may not be scien-tifically robust (Mitchell et al 2016).

With such concerns in mind, the HAPPI protocolwas specifically designed to simulate the specified ParisAgreement temperature targets (1.5 °C and 2.0 °C) asprecisely as possible by separating the impact of anadditional approximately half degree of warming fromuncertainty in climate model responses and internalclimate variability (Mitchell et al 2017). So, unlikeother multi-model exercises, HAPPI does not exclu-sively rely on prescribed radiative forcing protocols toproject the future temperature evolution, but robustlyconstrains it using prescribed future sea surface tem-peratures (SSTs) as well. This effectively reduces theinfluence ofmodel sensitivity and results in a narrowlydefined range of future temperature targets. Thus, theHAPPI Protocol provides an improved framework forthe analysis of the impacts of an additional half degreeof warming.

In this study, we estimate the extent to which glo-bal peak river flows seen under a 1.5 °C scenario differfrom those under a 2 °C scenario. In particular, weinvestigate where shifts in high flow occurrence wouldbe more important under these climate scenarios. Wefollow the HAPPI experimental protocol, to use themulti-model-ensemble daily output of four atmos-phere-only general circulation models (AGCMs).Total runoff values were used to drive a river-flowrouting scheme to calculate daily river flows for the: (a)present baseline (2006–2015), (b) 1.5 °C scenario, and(c) 2 °C scenario. From here we compute river flow

extreme statistics to calculate changes in high flowoccurrence andmagnitude.

2.Methods

2.1.HAPPI experiments and dataTheHAPPI protocol provides three 10 year simulationperiods with prescribed atmospheric forcing, sea-surface temperature and sea-ice coverage. The threescenarios are: (1) the reference or historical periodwhich is the ‘current decade’ from 2006 to 2015, (2) afuture decade that is about 1.5 °C warmer than pre-industrial levels, and (3) a future decade that is about2.0 °C warmer than pre-industrial levels. In thisprotocol, the reference experiment chosen is2006–2015 because it is the most recently observedperiod. This period also contains a range of differentSST patterns over the decade, allowing for an assess-ment of how the ocean conditions vary on inter-annual timescales, including important El Niño andLaNiña events.

For each scenario, we used the output of fourHAPPI AGCMS: CanAM4 (100 ensemble members);CAM4-2degree (100 ensemble members); NorESM1-HAPPI (125 ensemble members), and MIROC5 (50ensemble members). Each simulation (ensemblemember)within an experiment differs from the othersin its initial weather state. So, the use of 50–125 ten-year time slices really provides 500 years of data for theMIROC5 experiment, 1000 years for CanAM4 andCAM4-2degree; and 1250 years for NorESM1-HAPPI.Such extensive record (500–1250 years) in turn pro-vides the basis for robust calculations. More details ofthe HAPPI protocol is found in the supplementaryinformation is available online at stacks.iop.org/ERL/13/094003/mmedia.

2.2. RunoffThe AGCMs used in this study have a land surfacescheme which is coupled to the overlying atmosphere.For two models, NCC/NorESM1-HAPPI and ETH/

CAM4-2degree, runoff was stored in the HAPPIarchive. The other two models (MIROC/MIROC5andCCCma/CanAM4) did not have a runoff-produc-tion scheme activated for use in the present experi-ment and so we applied a comparable runoff-generation scheme to calculate runoff in those cases.This in turn enabled us to use those models alongsidethe members of the ensemble for which runoff datawere available.

So, for these two AGCMs we calculated runoff byusing a simple runoff production model designed tobe comparable with the runoff-production modelstypically embedded within climate models. Ourscheme uses a Rutter–Gash canopy formulation(Gash 1979) together with Penman–Monteith eva-poration calculated using available radiation data(Monteith 1965). Soil moisture was accounted for

2

Environ. Res. Lett. 13 (2018) 094003

using a two-layer model with saturation-excess runoffcomputed using a generalized TOPMODEL (ClarkandGedney 2008).

The snowpack model used here was based on atemperature-based model of accumulation and melt(Moore et al 1999, Hock 2003, Beven 2011). Snowaccumulates when precipitation falls while temper-ature is below a threshold temperature Ta. Whentemperature is above a threshold for melt, Tmmeltingoccurs at a rate proportional to the difference betweenthe current temperature and Tm. This conceptualmodel is widely used (Hock 2003, Zhang et al 2006,Rango andMartinec 2007, Beven 2011) and gives per-formance comparable with that of more parameter-rich energy balance models, despite their greater com-plexity (Parajka et al 2010).

We acknowledge the existence of other moresophisticated runoff schemes yet ours is intended to beapplied directly to atmospheric outputs resulted fromlarge ensembles such as the HAPPI ensemble allowingus to maximize the utility of the HAPPI ensemble bygenerating runoff when it was not provided directly.Comparison of our simulated runoff indicates goodlatitudinal agreement with observations provided bythe World Meteorological Organization Global Run-off Data Center (GRDC) (Fekete et al 2002) (see sup-plementary figure 4).

2.3. Runoff routingTotal runoff derived from NCC/NorESM1-HAPPIand ETH/CAM4models or calculated from CCCma/CanAM4 and MIROC/MIROC5 were routed byimplementing the grid-based hydrological routingscheme as presented by Dadson et al (2011). Thisscheme is based on the discrete approximation to the1D kinematic wave equation with lateral inflow (Okiand Sud 1998). This routing model routes flows alongthe path of steepest descent. Resistance to flow ishandled via a wave celerity parameter. This algorithmhas been widely used in various large-scale hydrologi-cal studies as well as various existing climate and landsurface model configurations. Further details areavailable in Bell et al (2007). We also selected thisscheme as it is computationally tractable when dealingwith a global problem with many hundreds ofensemblemembers, as with theHAPPI experiment.

Also, in this scheme the river network data usedwas constructed by using the network tracing method(NTM) which is a vector-based network scheme thattraces the path of river networks downstream (Oliveraand Raina 2003). This method works by overlaying amesh of a certain grid size over a fine-scale joined rivernetwork to determine the coarse-scale downstreamcell. This method has the advantage that it can be usedin areas where digitized river networks exist but noDTM is available and it is usually spatially closer to thebase river network. AsNTMhas been found to outper-form other raster-based methods, it has been widely

used in previous routing studies (Olivera andRaina 2003, Bell et al 2007, Davies and Bell 2009). Theresolution of the river network used here is halfdegree (0.5°×0.5°).

Also, the spatial resolution of runoffs given by eachAGCM (NCC/NorESM1-HAPPI: 1.875°×0.625°;ETH/CAM4: 2.5°×1.875°, MIROC/MIROC5:1.40°×1.40°, CCCma/CanAM4: 2.81°×2.81°)was downscaled to match the resolution of our rivernetwork. The downscaling was performed by applyinga bilinear interpolation technique. Apart from its sim-plicity, this technique has been found to provide amore realistic spatial gradient instead of patches ofsame runoff values from an AGCM in cells with finerresolution (Koirala et al 2014); as consequence thismethod has been used in a range of global hydrologicalstudies (Qian et al 2006,Materia et al 2009, Ukkola andPrentice 2013, Koirala et al 2014).

The final results of our study are presented at thescale of the routingmodel (0.5°×0.5°). We acknowl-edge that this increase in the spatial representation ofrunoff may result in our overlooking or under-representing runoff processes that operate at finerscales, which are represented by using statistical dis-tributions of sub-grid runoff generation. We note thatour routing scheme does not account for human inter-ventions and flow regulations such as dams and reser-voirs, or changes in land use and river interventions.We acknowledge the importance of these anthro-pogenic interventions, but they are not the subject ofour study. Models that incorporate complex anthro-pogenic processes do not typically offer computationalsimplicity appropriate for use with large ensembles.The comparison of derived river flows with observa-tions derived fromGRDC can be found in supplemen-taryfigure 5.

2.4. Riverflow extreme statisticsThe change in high flowhazards between the historicaland future conditions (1.5 °C and 2 °C warmingscenarios) was calculated from the probability of thehistorical 100 year return period river flow magnitudeto be exceeded at the 1.5 °C and 2 °C warmingscenarios. We selected the 100 year return period asthe reference river flow in order to keep consistencywith previous hydrological studies (Dankers andFeyen 2008, Hirabayashi et al 2008, Ward et al 2013).So, time series (500–1250 years) of simulated annualmaximum daily river flows in the historical scenario(2006–2015) were fitted to a two-parameter Gumbeldistribution (Gumbel 1941). As a result, themagnitudeof the 100 year return period river flowof the historicalscenario was calculated at each grid and for eachAGCM. Lastly, the return period of the historical100 year river flow was calculated for the time series ofthe 1.5 °Cand 2 °Cwarming scenarios.

A two-parameter Gumbel distribution was selec-ted as it provides relatively stable distribution

3

Environ. Res. Lett. 13 (2018) 094003

parameters showing adequate results in similar pre-vious studies (Hosking and Wallis 2005, Dankers andFeyen 2008, Hirabayashi et al 2008, Hirabayashiet al 2013). Also, while these previous studies acknowl-edge that the type of extreme distribution functionselectedmay impact the probabilities of extreme flows,it does not affect the estimation changes in the fre-quency and tendency of them, which is the aim of thisstudy. Similarly, Hirabayashi et al (2013) used theprobability plot correlation coefficient test to find thatglobally 76%of the AGCMs adequately fit to theGum-bel distribution (coefficient>0.96 at 95% of sig-nificance). Grids that did not fit this distribution weremostly found in arid regions which has also beenfound in similar studies (Hirabayashi et al 2008).

3. Results

We detect regions where the shift in occurrence ofglobal historical extreme flows at 1.5 °C and at 2.0 °Cdiffer as calculated by our ensemble of four AGCMs(figure 1). First, we identify regions that show an initialincrease in the frequency of occurrence of historical

high flows at 1.5 °C and then progressively increases at2.0 °C. These regions include South America (Amazoncatchment), central Africa, central-western Europe,the south of the US (Mississippi river area), centralAsia, and Siberian catchments. In these regions, themedian increase of the frequency of the current 1-in-100 year flow is once in 70–90 years at 1.5 °C, and at2.0 °C this frequency increases to at least once in 50years. Moreover, we find that in most of central andeastern China, the southern part of the Amazoncatchment, the Blue Nile, and in northern India, the1-in-100 year flow at 1.5 °C ranges between 1-in-50and 1-in-60 years, whereas at 2 °C such an event rangesbetween 1-in-25 and 1-in-35 years.

Second, we identify areas where the frequency ofthe 1-in-100 year flow increases at 1.5 °C but does notchange appreciably between 1.5 °C and 2.0 °C. Nota-bly these areas are located in South and Southeast Asia(south of the Yangtze river) and the Indus river basin.In these regions under a 1.5 °C scenario, the current1-in-100 year flow increases in frequency to 1-in-50 year return period. However, we also identify someareas, such as Southeast Asia, where the current 100year flood occurs 1-in-25 years in the future. In North

Figure 1.Multi-modelmedian return period (years) for future riverflow corresponding to the historical 100 year flowunder (a) at1.5 °C scenario (b) at 2 °C scenario. Grid boxeswhere the historical averaged annual flow is<10 m3 s−1 were screened out. Bluecolors indicate amore frequent historical 100 year flowwhereas red colors indicate a reduction in the frequency.

4

Environ. Res. Lett. 13 (2018) 094003

America and eastern Europe, the current 100 yearreturn period flow increases in frequency only slightly,to once in 80 years at 1.5 °C.

We also find some regions which see a decrease inthe frequency of high flows between the present dayand 1.5 °C or 2.0 °C or both. These regions are locatedin the northern part of South America, areas of thewestern coast of the United States (Colorado River)and Canada, South Africa (Orange River), Scandina-via, and in some parts of eastern Europe. In most ofthese regions, we note that the current 1-in-100 yearflow decreases in frequency to approximately 1-in-150years, with little further decrease at 2.0 °C. Lastly, wefind that in the area of the Murray-Darling basin inAustralia, there is a shift from historical high flows notchanging in a world 1.5 °C warmer (projected returnperiods between 90 and 125 years), whilst under a2.0 °C scenario the return period decreases to approxi-mately 1-in-70 years.

3.1. Physical characterization of highflowsIn order to understand the physical reasons thatinduced shifts in high flow frequencies, we calculatedthe change in both the number of frost days and thenumber of days where precipitation has occurredbetween the two scenarios in comparison to thehistorical baseline (see figure 2). We find an importantdecrease in the number of frost days at 1.5 °C whichslightly increases at 2.0 °C. This decrease is greater inNorth America, and central and eastern Europe(figures 2(a), (b)). This finding suggests that globally

the role of snowmelt in the generation of high flowswill decrease as snow accumulation reduces. Excep-tions to this pattern are found in the South and EastCoast of the US and in the south of South America. Inthese areas, particularly in the US, where we estimatean increase in the occurrence of high flows, snowdynamics and possibly longer periods of snow accu-mulation may be important drivers of high flowevents.

The change in the number of days with precipita-tion between the two climate thresholds (figures 2(c),(d)) shows a diverse regional response. For example, inmost of the regions where we estimate a progressiveincrease in the frequency of high flows between 1.5 °Cand 2.0 °C, such as central-western Europe or SouthAmerica, the number of days with precipitation typi-cally decreases or does not importantly change at1.5 °C and also at 2.0 °C. This is also accompanied by adecrease in average annual precipitation magnitudesat both scenarios (see supplementary figure 2). Thisresult suggests that in these areas extreme precipita-tion, will likely dominate the occurrence of high flows,despite the reduction in the number of rain days. Animportant exception is observed in Siberia, where wesee an increase in the frequency of high flows accom-panied by a slight decrease in the number of days withprecipitation.

In regions where the increase in high flow occur-rences is important at 1.5 °C but minor at 2.0 °C, suchas the south of Asia, we find that, at this first temper-ature target, the frequency of rain days decreases as

Figure 2.Multi-modelmedian change (%) in the number of frost days between the historical and: (a) 1.5 °C scenario, (b) 2.0 °Cscenario. A frost day is that where estimated temperatures are less than 0 °C. (c) and (b) is as (a) and (b), correspondingly, but for thenumber of days with precipitation.Wedefine a daywith precipitation as that where estimated daily precipitation is greater than2 mm d−1. This threshold is selected as it adequately describes the range of global light precipitation (0.1–5 mm d−1) (Qiaohonget al 2017).

5

Environ. Res. Lett. 13 (2018) 094003

well as total annual precipitation magnitudes. Yet at2.0 °C days with precipitation are either more com-mon (especially in North America) or show a lowerreduction compared with that detected at 1.5 °C.Similarly, annual precipitation magnitudes show aslight increase when compared to the historical refer-ence period. This finding suggests that, in theseregions, the role of shorter wet seasons and storms inhigh flows, which at 1.5 °C seems to be more impor-tant, will decrease at 2.0 °C. So, in these regions, thedifference between the two climate targets, and thusflow occurrences, may be associated with shifts indurations and characteristics of the wet season. In theregions where we observe a decrease in the occurrenceof high flows, we also detect a decrease in the numberof days with precipitation and annual precipitationmagnitudes.

3.2.Multi-model consistency and biases in resultsThe change in high flow frequency at 1.5 °C and2.0 °C, although consistent in sign between modelsacross most parts of the globe, differs in specific

regions (see figure 3). Globally, we find that 49% ofland grid cells at 1.5 °C and 43% at 2.0 °C show a highconsistency in the tendencies described above (consis-tency defined as an agreement in trajectory of three outof four models). This high level of consistency is seenparticularly in regions where we find an increase inhigh flow frequency, such as the south of Asia, India,or East Africa. Conversely, in just about 7% of landgrid cell at 1.5 °C and 4% at 2.0 °C just one modelagrees with the tendency mapped above. Typicalregions with lower consistencies can be found innorthern Russia and parts of the tropics. In these areasour projections are subject to an important spreadacross the AGCMensemble.

Additionally, the examination of the way eachAGCM simulates the changes in high frequency helpsto disseminate further model biases (supplementaryfigure 1). For instance, NCC/NorESM1-HAPPI gen-erally projects greater increases in the frequency ofoccurrence of the reference high flow in the tropics.Thus, it simulates amore pronounced increase in highflow from the historical reference return time of 1-in-100 years to approximately 1-in-40 years at 2.0 °C for

Figure 3.Consistency in projected river flow change for the four AGCMs used in this study. Consistency is expressed as a percentage ofthe total the number ofmodels that agree with the tendency in the direction of change (increase or decrease) of the historical returnperiod.

6

Environ. Res. Lett. 13 (2018) 094003

larger areas (e.g., central South America and SouthAsia). Also,MIROC/MIROC5 andCCCma/CanAM4show greater increases in reference extreme flow fre-quencies in northern regions. We also find that ETH/

CAM4-2degree generally produces lower river flowsoutside the tropics.

Moreover, by examining the way each AGCMrepresents precipitation and then runoff (supplemen-tary figure 3 for precipitation and 4 for runoff), weestimate whether existing biases (such as the lowerriver flow in ETH/CAM4-2degree) are due to defi-ciencies in the AGCM, runoff generated, or the riverrouting model. Using zonal means, we find that eachof the four AGCMs slightly overestimates mean pre-cipitation in the tropics compared with the GlobalPrecipitation Climatology Center dataset, GPCC(Becker et al 2011) (maximum precipitationbias∼30%).

This tropical overestimation persists for meanrunoff in MIROC/MIROC5 and, to a lesser extent, inCCCma/CanAM4 when compared with observedriver flow data obtained from the World Meteor-ological Organization Global Runoff Data Center,GRDC (Fekete et al 2002). ETH/CAM4-2degree andNCC/NorESM1-HAPPI on the other hand, show animportantly reduced runoff in the tropics comparedwith the other two AGCMs. In contrast, extra-tropicaloverestimation of precipitation in MIROC/MIROC5does not lead to equally overestimated runoff in thoseregions. It is therefore not possible to identify the rootcause for biases in river flowwith confidence, given thesubstantial discrepancies in the precipitation-to-run-off ratio in the models. However, the fact that the twomodels with lowest zonal average runoff also show thelowest river flow is indicative that the treatment of thewater budget and the associated hydrological balance

in the AGCMs is crucial in capturing the magnitude ofthe river flow. That said, we emphasize that biases inriver flow are unlikely to affect the change in frequencyof the return time between the different temperaturescenarios. This is also shown in our sampling uncer-tainty estimates derived from resampling our dataset1000 times for the ten largest catchments globally (seetable 1).

3.3.Magnitude of projected changes in the 100 yearriverflowIn contrast to the important increase in high flowoccurrences, described above, we lastly find a diverseresponse in the change of flow magnitudes betweenthe two temperature scenarios for the ten largestcatchments in the world (see table 1). In catchmentswhere we do not detect a large change in the frequencyof occurrence of the historical 1-in-100 years returnperiod flow such as the Amazon and Murray-Darling,we find a reduction of up to 5% in mean annual riverflowunder the 1.5 °C scenario. In contrast, somemeanriver flows such as those at the outlets of northerncatchments (Mackenzie and Ob’) show importantincreases of up to 8% (at 2.0 °C) despite quasi-constant changes in frequency of occurrence of peakflows. At the Yangtze river, apart from the increase inthe frequency of occurrence of extreme river-flows aspresented above, there is an important increase in themean discharge at both, 1.5 °C–2.0 °C. Also, we notethat for theMississippi and the Ganges rivers the smallchange in mean river flow is contrasted by animportant increase in high flow frequencies. Also, wenote that regarding flow magnitude, the change from1.5 °C to 2.0 °C levels of warming are not as clear asthose detected for high river flow frequencies. How-ever, we note that the combined fidelity of the

Table 1.Projected change (expressed as return period) of the 1-in-100 year historical flow at both scenarios and estimated change in themean riverflow (%) from ten selectedmajor global outlets. Uncertainty due to sample size is calculated by a bootstrapmethod. Uncertaintyfor the four AGCMswas averaged. Standard deviation is calculated fromour ensemble of AGCMs.

Return period of historical high flow

Change inmean historical

river flow (%)

1.5 °C 2.0 °C1.5 °C 2.0 °C

River system

R. per-

iod (year)Uncertainty due to sam-

ple size

R. per-

iod (year)Uncertainty due to sam-

ple size Mean Std Mean Std

Amazon 108 ±5 87 ±3 −4.7 1.3 −5.2 2.6

Mississipi 85 ±3 64 ±4 1.0 2.4 1.6 5.1

Mackenzie 61 ±7 49 ±4 6.3 2.4 8.9 3.7

Congo 80 ±6 57 ±5 −0.6 2.2 −0.2 4.2

Nile 92 ±7 55 ±4 −1.2 2.2 2.8 3.4

Rhine 70 ±8 72 ±7 2.6 2.4 1.3 3.2

Danube 54 ±8 63 ±6 2.0 5.4 −1.8 5.1

Murray-

Darling

143 ±5 96 ±5 −9.0 1.7 −4.0 4.8

Ganges 67 ±4 41 ±3 1.7 3.6 5.0 4.7

Indus 130 ±6 127 ±6 −1.2 5.1 −0.3 7.4

Yangtze 58 ±8 26 ±9 7.4 4.9 8.2 5.9

Ob’ 54 ±9 48 ±9 5.7 3.2 7.5 5.1

7

Environ. Res. Lett. 13 (2018) 094003

ensemble of our AGCMs and the river routing modelto reproduce the annual cycle of the river flow,exemplified for a subset of major river catchments islimited in certain areas (see supplementary figure 5).The representation of the water budget and hydro-logical balance discussed in the section above may alsobe affecting the ability to represent changes in flowmagnitudes.

3.4. Flow responses for alternative return periodsWe note that although the four AGCMs ensembledisagree in maximum daily flow magnitudes, they arenot sensitive to the choice of the return time ashighlighted in figure 4, where we have plotted thereturn times for all scenarios and models for fourdifferent rivers which are affecting a large fraction ofglobal population. At the Yangtze as well as at theOrange outlets, we find that the difference between1.5 °C and 2.0 °C remains small for a wide range ofreturn times (figures 4(a)/(d)). As mentioned above,the four models show similar tendencies when simu-lating changes in return periods of high flows for theselected regions, which suggests that our flow fre-quency results are indeed robust, despite biases in themagnitude of the river flow. The detected wet bias inrunoff at certain latitudes in MIROC/MIROC5 reap-pears here, in that we find the river flow magnitude to

be overestimated at the Yangtze and the Ganges(figures 4(a), (b)). Also, the difference in runoffbetween the twowettest anddriest AGCMs ismirroredin the return time plots at the Upper Blue Nile(figure 4(c)), where we find considerably higher riverflow magnitudes between the two model subsets. Inother words, the regionally-overestimated precipita-tion (∼30% in NCC/NorESM1-HAPPI and CCCma/CanAM4) translates directly into higher mean andextreme river flows.

4.Discussion

The results shown here agree with various of theprevious findings that also project shifts in runoff,river flow magnitudes and flood frequencies underrepresentative concentration pathways in variousglobal regions (Hirabayashi et al 2013, Stockeret al 2013, Dankers et al 2014, Koirala et al 2014,Schewe et al 2014, Arnell and Gosling 2016). Forinstance, our findings consistently agree with theprojection of increasedmean river flows and high flowfrequencies for Southeast Asia, eastern Africa, andvarious parts of South America. Similarly, for a world1.5 °C and 2.0 °Cwarmer, Döll et al (2018) also projectincreases in high flows in South and South Asia, andCentral Africa. On the contrary, these studies typically

Figure 4.Maximumone-day river flow as simulated by the four AGCMs ensemble (a) the outlet area of Yangtze River, (b) outlet of theGanges River, (c) the upper BlueNile (EthiopianHighlands), and (d) theOrange River in SouthAfrica.

8

Environ. Res. Lett. 13 (2018) 094003

suggest a general decrease in future high flows incentral-eastern Europe. On the contrary, our resultsproject an increase in high flow frequencies in thisregion (relatively good multi-model consistenciesfound here).

Also, the nonlinearity found between current con-ditions and the two future warming scenarios might aswell be related to counterbalancing effects due to thepresence of different forcing agents that vary greatlyover time. In both HAPPI future scenarios, the aerosolload is reduced to approximately one third of its cur-rent levels. This leads many regions to show higherrainfall increases at 1.5 °C compared to what has beenobserved or simulated in similar experiments as highaerosol loads have been connected to a reducedstrength of the hydrological cycle in several globalregions (Bollasina et al 2011). This may be the case inthe Asian monsoon region which is generally char-acterized with high aerosol loads and our results showan increase in high flow frequencies detected here. Ifthe role these different drivers are playing is not appro-priately analyzed, the attribution results can easily bemisinterpreted. As such, it is also important to under-stand other region-specific features whichmay explainthe different local flow responses between bothscenarios.

Moreover, the consistent decrease in high flow fre-quencies found in our study for Scandinavia andSouth Africa has been previously estimated by Hir-abayashi et al (2013) and Koirala et al (2014) for theRCP8.5 scenario. Similar projections were also foundby Döll et al (2018) for the particular temperature tar-gets agreed in Paris. Also, it is important to note that insome areas where consistently project an increase inhigh flow frequencies, various previous reports esti-mate a future decrease in mean river flow accom-panied by major drought hazard at 2.0 °C (Scheweet al 2014, Lehner et al 2017). These areas include,southern China (such as the Yangtze river), centralEurope, and central South America. Thus, our find-ings suggest an intensification of hydrologicalextremes in these regions. Yet it is important to con-sider the additional role that that the reduction ofaerosols, described previously, may have in showingwetter futures in our estimates in these regions.

5. Conclusions

By using the HAPPI Protocol, which is a frameworkdesigned to reduce the influence of model sensitivity inclimate outputs,we analyze the impacts of an additionalhalf degree of warming on high river flows, as agreed inParis in 2015. We find that historical extreme flows(represented by 1-in-100 year return periods) occurmore frequently at 1.5 °C, but even greater additionalchanges are projected formany regions at 2.0 °C. Theseregions include most of South America, central Africa,

central-western Europe, the south of the US (Missis-sippi river area), central Asia, and Siberian catchments.We detect that in these regions, the median increase ofthe frequency of the current 1-in-100 year flow is oncein 70–90 years at 1.5 °C, and at 2.0 °C this frequencyincreases to at least once in 50 years. Our result suggeststhat in these areas (with the exception of Siberia), thischange is accompanied by shorter, and consequentlymore intense, rainy seasons. As such, committing to a1.5 °C level ofwarming reduces the chanceof highflowsin these regions. In other highly populated areas, suchas South and Southeast Asia (south of the Yangtze river)and the Indus river basin, the current 1-in-100 yearbecomesmore frequent at a 1.5 °C level of warming (1-in-50 year). Then the change from 1.5 °C to 2.0 °C isnot importantly appreciable. Yet it is important to notethat reductions in atmospheric aerosol loading repre-sented in the HAPPI experiments lead to wetter futuresin the areas discussed. As such, the implications thatregional-scale processes may have in our results may beworth noting in a context of global mitigation targets.Lastly, while our analysis highlights limitations con-cerning regional and seasonal river flow projectionsproduced using global models, we have demonstratedthat our river flow results are robust with regard tochanging risks tendencies. Thus, our results may beused to inform adaptation strategies and to guideassessments of socio-economic impacts at the regionalor catchment scale.

ORCID iDs

Homero Paltan https://orcid.org/0000-0001-6952-6850

References

Alfieri L et al 2017Global projections of riverflood risk in awarmerworldEarth’s Future 5 171–82

Arnell NWandGosling SN2016The impacts of climate change onriverflood risk at the global scaleClim. Change 134 387–401

Arnell NWandLloyd-Hughes B 2014The global-scale impacts ofclimate change onwater resources and flooding under newclimate and socio-economic scenariosClim. Change 122127–40

Becker A 2011GPCC full data reanalysis version 7.0 at 0.5:Monthlyland-surface precipitation from rain-gauges built onGTS-basedand historic data (https://doi.org/10.5676/DWD_GPCC/FD_M_V6_050)

Bell VA et al 2007Development of a high resolution grid-based riverflowmodel for usewith regional climatemodel outputHydrol. Earth Syst. Sci. 11 532–49

BevenK J 2012Rainfall-RunoffModelling: The Primer (NewYork:Wiley) (https://doi.org/10.1002/9781119951001)

BollasinaMA,Ming Y andRamaswamyV2011Anthropogenicaerosols and theweakening of the SouthAsian summermonsoon Science 334 502LP–5

ClarkDB andGedneyN 2008Representing the effects of subgridvariability of soilmoisture on runoff generation in a landsurfacemodel J. Geophys. Res. 113D10111

Dadson S J, Bell VA and Jones RG2011 Evaluation of a grid-basedriverflowmodel configured for use in a regional climatemodel J. Hydrol. 411 238–50

9

Environ. Res. Lett. 13 (2018) 094003

Dankers R and Feyen L 2008Climate change impact on flood hazardin Europe: an assessment based on high-resolution climatesimulations J. Geophys. Res. Atmos. 113 1–17

Dankers R et al 2014 First look at changes inflood hazard in theInter-Sectoral ImpactModel Intercomparison ProjectensembleProc. Natl Acad. Sci. 111 3257–61

DaviesHNandBell VA2009Assessment ofmethods for extractinglow-resolution river networks fromhigh-resolution digitaldataHydrol. Sci. J.—J. Sci. Hydrologiques 54 17–28

Döll P et al 2018Risks for the global freshwater system at 1.5 °Cand2 °Cglobal warmingEnviron. Res. Lett. 13 44038

Fekete B, Vörösmarty C andWG2002Global Composite RunoffFields onObserved River Discharge and SimulatedWaterBalances (Koblenz, Germany:Water SystemAnalysis Group,University ofNewHampshire, andGlobal RunoffDataCentre)

Gash JHC1979An analyticalmodel of rainfall interception byforestsQ. J. R.Meteorol. Soc. 105 43–55

Gumbel E J 1941The return period offloodflowsAnn.Math. Stat.12 163–90

Hirabayashi Y et al 2008Global projections of changing risks offloods and droughts in a changing climateHydrol. Sci. J. 53754–72

Hirabayashi Y et al 2013Globalflood risk under climate changeNat.Clim. Change 3 816–21

HockR 2003Temperature indexmeltmodelling inmountain areasJ. Hydrol. 282 104–15

Hosking J RMandWallis J R 2005Regional Frequency Analysis: AnApproach Based on L-Moments (Cambridge: CambridgeUniversity Press)

HulmeM2016 1.5 °Cand climate research after the ParisAgreementNat. Clim. Change 6 222–4

JongmanB et al 2014 Increasing stress on disaster-risk finance dueto large floodsNature Clim. Change 4 1–5

Koirala S et al 2014Global assessment of agreement among streamflowprojections usingCMIP5model outputsEnviron. Res.Lett. 9 64017

Lehner F et al 2017 Projected drought risk in 1.5 °Cand 2 °Cwarmer climatesGeophys. Res. Lett. 44 7419–28

Materia S et al 2009The sensitivity of simulated river discharge toland surface representation andmeteorological forcingsJ. Hydrometeorol. 11 334–51

Milly PCD et al 2002 Increasing risk of great floods in a changingclimateNature 415 514–7

Mitchell D et al 2016Realizing the impacts of a 1.5 °CwarmerworldNat. Clim. Change 6 735–7

Mitchell D et al 2017Half a degree additional warming, prognosisand projected impacts (HAPPI): background andexperimental designGeosci.Model Dev. Discuss. 571–83

Monteith J L 1965Evaporation and environment Symp. Soc. Exp. Biol19 (Cambridge) 205–23

Moore R J et al 1999Methods for snowmelt forecasting in uplandBritainHydrol. Earth Syst. Sci. Discuss. 3 233–46

Nijssen B et al 2001Hydrologic sensitivity of global rivers to climatechangeClim. Change 50 143–75

Okazaki A, Yeh P J F, Yoshimura K,WatanabeM,KimotoMandOki T 2012Changes inflood risk under global warmingestimated usingMIROC5 and the discharge probability indexJ.Meteorolog. Soc. Jpn. Ser. II 90 509–24

Oki T and SudYC1998Design of total runoff integrating pathways(TRIP)—a global river channel networkEarth Interact. 2 1–37

Olivera F andRaina R 2003Development of large scale gridded rivernetworks fromvector streamdata J. Am.Water Resour. Assoc.39 1235–48

Parajka J et al 2010 Evaluation of snow cover and depth simulated bya land surfacemodel using detailed regional snowobservations fromAustria J. Geophys. Res. 115D24117

Peduzzi P et al 2009Assessing global exposure and vulnerabilitytowards natural hazards: the disaster risk indexNat.HazardsEarth Syst. Sci. 9 1149–59

QianT et al 2006 Simulation of global land surface conditions from1948 to 2004: I. Forcing data and evaluationsJ. Hydrometeorol. 7 953–75

Qiaohong S et al 2017A review of global precipitation data sets: datasources, estimation, and intercomparisonsRev. Geophys. 5679–107

RangoA andMartinec J 2007Revisiting the degree day-method forsnowmleat computations JAWRA J. Am.Water Resour. Assoc.31 657–69

Rogelj J et al 2016 Perspective : Paris Agreement climate proposalsneed boost to keepwarmingwell below 2 °CNat. Clim.Change 534 631–9

SchellnhuberH J, Rahmstorf S andWinkelmannR 2016Why theright climate target was agreed in ParisNature Clim. Change 6649–53

Schewe J et al 2014Multimodel assessment of water scarcity underclimate change Proc. Natl Acad. Sci. 111 3245LP–50

StockerTF et al2013Technical summaryClimate change 2013: ThePhysical Science Basis.Contribution ofWorkingGroup I to theFifthAssessmentReport of the Intergovernmental Panel onClimateChange. (Cambridge:CambridgeUniversity Press)pp33–115

Ukkola AMandPrentice I C 2013Aworldwide analysis of trends inwater-balance evapotranspirationHydrol. Earth Syst. Sci.Discuss. 10 4177–87

WakeB 2013 Flooding costsNature 3 778Ward P et al 2013Assessingflood risk at the global scale:model

setup, results, and sensitivity Environ. Res. Lett. 8 44019WinsemiusHC et al 2015Global drivers of future river flood risk

Nature Clim. Change 6 381Zhang Y, Liu S andDing Y 2006Observed degree-day factors and

their spatial variation on glaciers inwesternChinaAnn.Glaciol. 43 301–6

10

Environ. Res. Lett. 13 (2018) 094003