climate research 17:145 · climate research clim res vol. 17: 145–168, 2001 published august 15...
TRANSCRIPT
CLIMATE RESEARCHClim Res
Vol. 17: 145–168, 2001 Published August 15
1. INTRODUCTION
The climates of Africa are both varied and varying:varied because they range from humid equatorialregimes, through seasonally-arid tropical regimes, tosub-tropical Mediterranean-type climates, and varyingbecause all these climates exhibit differing degrees oftemporal variability, particularly with regard to rain-fall. Understanding and predicting these inter-annual,inter-decadal and multi-decadal variations in climatehas become the major challenge facing African andAfrican-specialist climate scientists in recent years.
Whilst seasonal climate forecasting has taken greatstrides forward, in both its development and applica-tion (Folland et al. 1991, Stockdale et al. 1998, Wash-ington & Downing 1999, also see http://www.ogp.noaa.gov/enso/africa.html [SARCOF: Southern AfricaRegional Climate Outlook Forum]), the ultimate causesof the lower frequency decadal and multi-decadalrainfall variability that affects some African climateregimes, especially in the Sahel region, remain uncer-tain (see Rowell et al. 1995 vs Sud & Lau 1996, also Xue& Shukla 1998). This work examining the variability ofAfrican climate, especially rainfall, is set in the widercontext of our emerging understanding of humaninfluences on the larger, global-scale climate. Increas-ing greenhouse gas accumulation in the global atmo-
© Inter-Research 2001
*E-mail: [email protected]
African climate change: 1900–2100
Mike Hulme1,*, Ruth Doherty3, Todd Ngara4, Mark New5, David Lister2
1Tyndall Centre for Climate Change Research and 2Climatic Research Unit, School of Environmental Sciences, University of East Anglia, Norwich NR4 7TJ, United Kingdom
3Environmental and Societal Impacts Group, NCAR, Boulder, Colorado 80307, USA4Climate Change Office, Ministry of Mines, Environment and Tourism, Postal Bag 7753 Causeway, Harare, Zimbabwe
5School of Geography, Mansfield Road, University of Oxford, Oxford OX1 3TB, United Kingdom
ABSTRACT: This paper reviews observed (1900–2000) and possible future (2000–2100) continent-wide changes in temperature and rainfall for Africa. For the historic period we draw upon a newobserved global climate data set which allows us to explore aspects of regional climate changerelated to diurnal temperature range and rainfall variability. The latter includes an investigation ofregions where seasonal rainfall is sensitive to El Niño climate variability. This review of past climatechange provides the context for our scenarios of future greenhouse gas-induced climate change inAfrica. These scenarios draw upon the draft emissions scenarios prepared for the IntergovernmentalPanel on Climate Change’s Third Assessment Report, a suite of recent global climate model experi-ments, and a simple climate model to link these 2 sets of analyses. We present a range of 4 climatefutures for Africa, focusing on changes in both continental and regional seasonal-mean temperatureand rainfall. Estimates of associated changes in global CO2 concentration and global-mean sea-levelchange are also supplied. These scenarios draw upon some of the most recent climate modellingwork. We also identify some fundamental limitations to knowledge with regard to future African cli-mate. These include the often poor representation of El Niño climate variability in global climatemodels, and the absence in these models of any representation of regional changes in land cover anddust and biomass aerosol loadings. These omitted processes may well have important consequencesfor future African climates, especially at regional scales. We conclude by discussing the value of thesort of climate change scenarios presented here and how best they should be used in national andregional vulnerability and adaptation assessments.
KEY WORDS: African climate · Climate scenarios · Rainfall variability · Climate modelling · Seasonalforecasting · Land cover changes
Resale or republication not permitted without written consent of the publisher
Clim Res 17: 145–168, 2001
sphere and increasing regional concentrations ofaerosol particulates are now understood to havedetectable effects on the global climate system (Santeret al. 1996). These effects will be manifest at regionalscales although perhaps in more uncertain terms(Mitchell & Hulme 1999, Giorgi & Francisco 2000).
Africa will not be exempt from experiencing thesehuman-induced changes in climate. Much work re-mains to be done, however, in trying to isolate thoseaspects of African climate variability that are ‘natural’from those that are related to human influences.African climate scientists face a further challenge inthat on this continent the role of land cover changes—some natural and some human-related—in modifyingregional climates is perhaps most marked (Xue 1997).This role of land cover change in altering regional cli-mate in Africa has been suggested for several decadesnow. As far back as the 1920s and 1930s, theoriesabout the encroachment of the Sahara and the desicca-tion of the climate of West Africa were put forward(Stebbing 1935, Aubreville 1949). These ideas havebeen explored over the last 25 yr through modellingstudies of tropical north African climate (e.g. Charney1975, Cunnington & Rowntree 1986, Zheng & Eltahir1997). It is for these 2 reasons—large internal climatevariability driven by the oceans and the confoundingrole of human-induced land cover change—that cli-mate change ‘predictions’ (or the preferable term sce-narios) for Africa based on greenhouse gas warmingremain highly uncertain. While global climate models(GCMs) simulate changes to African climate as a re-sult of increased greenhouse gas concentrations,these 2 potentially important drivers of African climatevariability—for example El Niño/Southern Oscillation(ENSO) (poorly) and land cover change (not at all)—are not well represented in the models.
Nevertheless, it is of considerable interest to try andexplore the magnitude of the problem that theenhanced greenhouse effect may pose for Africanclimate and for African resource managers. Are thechanges that are simulated by GCMs for the next cen-tury large or small in relation to our best estimates of‘natural’ climate variability in Africa? How well doGCM simulations agree for the African continent? Andwhat are the limitations/uncertainties of these modelpredictions? Answering these questions has a verypractical relevance in the context of national vulnera-bility and adaptation assessments of climate changecurrently being undertaken by many African nationsas part of the reporting process to the UN FrameworkConvention on Climate Change. This paper makes acontribution to these assessments by providing anoverview of future climate change in Africa, particu-larly with regard to simulations of greenhouse gaswarming over the next 100 yr.
We start the paper (Section 2) by reviewing someprevious climate change scenarios and analyses forregions within Africa. Such studies have been far fromcomprehensive. Section 3 explains the data, modelsand approaches that we have taken in generating ouranalyses and constructing our climate change scenar-ios for Africa. In Section 4 we consider the salientfeatures of African climate change and variability overthe last 100 yr, based on the observational record ofAfrica climate. Such a historical perspective is essen-tial if the simulated climates of the next century areto be put into their proper context. Section 5 thenpresents our future climate change scenarios forAfrica, based on the draft Special Report onEmissions Scenarios range of future greenhouse gasemissions (http://sres.ciesin.org/index.html [SRES])and the GCM results deposited with the Intergovern-mental Panel on Climate Change (IPCC) Data Distrib-ution Centre (http://ipcc-ddc.cru.uea.ac.uk/index.html[DDC]). Changes in mean seasonal climate are shownas well as some measures of changed interannual vari-ability. Section 6 then discusses these future climatesimulations in the light of modelling uncertainties andin the context of other causes of African climate vari-ability and change. We consider how much useful andreliable information these types of studies yield andhow they can be incorporated into climate changeimpacts assessments. Our key conclusions are pre-sented in Section 7.
2. REVIEW OF PREVIOUS AFRICAN CLIMATECHANGE SCENARIO WORK
There has been relatively little work published onfuture climate change scenarios for Africa. The variousIPCC assessments have of course included globalmaps of climate change within which Africa has fea-tured, and in Mitchell et al. (1990) the African Sahelwas one of 5 regions for which a more detailed analy-sis was conducted. Kittel et al. (1998) and Giorgi &Francisco (2000) also identify African regions withintheir global analysis of inter-model differences in cli-mate predictions, but no detailed African scenarios arepresented.
Tyson (1991) published one of the first scenarioanalyses specifically focused on an African region. Inthis case some climate change scenarios for southernAfrica were constructed using results from the firstgeneration GCM equilibrium 2 × CO2 experiments. Ina further development, Hulme (1994a) presented amethod for creating regional climate change scenarioscombining GCM results with the newly publishedIPCC IS92 emissions scenarios and demonstrated theapplication of the method for Africa. In this study mean
146
Hulme et al.: African Climate Change: 1900–2100
annual temperature and precipitation changes from1990 to 2050 under the IS92a emission scenario werepresented.
Some more recent examples of climate scenarios forAfrica use results from transient GCM climate changeexperiments. Hernes et al. (1995) and Ringius et al.(1996) constructed climate change scenarios for theAfrican continent that showed land areas over theSahara and semi-arid parts of southern Africa warmingby the 2050s by as much as 1.6°C and the equatorialAfrican countries warming at a slightly slower rate ofabout 1.4°C. These studies, together with Joubert et al.(1996), also suggested a rise in mean sea-level aroundthe African coastline of about 25 cm by 2050. A moreselective approach to the use of GCM experiments wastaken in Hulme (1996a). They described 3 future cli-mate change scenarios for the Southern African Devel-opment Community (SADC) region of southern Africafor the 2050s on the basis of 3 different GCM experi-ments. These experiments were selected to deliber-ately span the range of precipitation changes for theSADC region as simulated by GCMs. Using thesescenarios, the study then described some potentialimpacts and implications of climate change for agri-culture, hydrology, health, biodiversity, wildlife andrangelands. A similar approach was adopted by Con-way et al. (1996) for a study of the impacts of climatechange on the Nile Basin. More recently, the Africachapter (Zinyowera et al. 1998) in the IPCC Assess-ment of Regional Impacts of Climate Change (IPCC1998) also reported on some GCM studies that relatedto the African continent.
Considerable uncertainty exists in relation to large-scale precipitation changes simulated by GCMs forAfrica (Hudson 1997, Hudson & Hewitson 1997, Jou-bert & Hewitson 1997, Feddema 1999). Joubert &Hewitson (1997) nevertheless conclude that, in gen-eral, precipitation is simulated to increase over muchof the African continent by the year 2050. These GCMstudies show, for example, that parts of the Sahel couldexperience precipitation increases of as much as 15%over the 1961–90 average by 2050. A note of cautionis needed, however, concerning such a conclusion.Hulme (1998) studied the present-day and futuresimulated inter-decadal precipitation variability in theSahel using the HadCM2 GCM. These model resultswere compared with observations during the 20thCentury. Two problems emerge. First, the GCM doesnot generate the same magnitude of inter-decadalprecipitation variability that has been observed overthe last 100 yr, casting doubt on the extent to whichthe most important controlling mechanisms are beingsimulated in the GCM. Second, the magnitude of thefuture simulated precipitation changes for the Sahelis not large in relation to ‘natural’ precipitation vari-
ability for this region. This low signal:noise ratio sug-gests that the greenhouse gas-induced climate changesignals are not well defined in the model, at least forthis region. We develop this line of reasoning in thispaper and illustrate it in Section 5 with further exam-ples from Africa.
Although there have been studies of GCM-simu-lated climate change for several regions in Africa, thedownscaling of GCM outputs to finer spatial and tem-poral scales has received relatively little attention inAfrica. Hewitson & Crane (1998) and Hewitson &Joubert (1998) have applied empirical downscalingmethods to generate climate change scenarios forSouth Africa using artificial neural networks and pre-dictors relating to upper air circulation and tropos-pheric humidity. The usual caveats, however, apply tothese downscaled scenarios (Hulme & Carter 1999)—they are still dependent on the large-scale forcing fromthe GCMs and they still only sample one realisation ofthe possible range of future possible climates, albeitwith higher resolution. The application of regionalclimate models is still in its infancy, although someinitiatives are now under way for East Africa (Sun etal. 1999), West Africa (Wang & Eltahir 2000) and south-ern Africa (B. Hewitson pers. comm.). These initiativeshave not yet generated experimental results fromregional climate change simulations for use in scenarioconstruction.
3. DATA AND METHODS
For our analyses of observed climate variability inAfrica we use the global gridded data sets of Jones(1994, updated; mean temperature), Hulme (1994b,updated; precipitation), and New et al. (1999, 2000;10 surface climate variables). These data sets are allpublic domain and are available, along with somedocumentation on their construction, from the follow-ing Web sites: GCM results were taken from the IPCCData Distribution Centre (DDC) (http://ipcc-ddc.cru.uea.ac.uk); most of the observed data sets used herecan be obtained from the Climatic Research Unit(http://www.cru.uea.ac.uk); the draft (February 1999;non-IPCC approved, but used with permission) SRESemissions scenarios were obtained from the SRES(http://sres.ciesin.org/index.html). The data sets ofJones (1994) and Hulme (1994b) exist on a relativelycoarse grid (5° latitude/longitude and 2.5° latitude by3.75° longitude respectively), while the data set of Newet al. (1999, 2000) exists with a 0.5° latitude/longituderesolution. These observed data are resolved only tomonthly time steps and we therefore undertake nooriginal analyses of observed daily climate variability.For Ethiopia and Zimbabwe we analyse unpublished
147
Clim Res 17: 145–168, 2001
monthly mean maximum and minimum temperaturedata for a number of stations in each country. Thesedata originate from the respective national meteoro-logical agencies. For the index of the Southern Oscilla-tion we use the updated index of Ropelewski & Jones(1987), calculated as the normalised mean sea-levelpressure difference between Tahiti and Darwin andavailable from Climate Monitor online (http://www.cru.uea.ac.uk/cru/climon).
Other climate-related and continent-wide data setsalso have value for some climate analyses, whetherthese data are derived from satellite observations (e.g.Normalised Difference Vegetation Index or satellite-derived precipitation estimates) or from numericalweather prediction model re-analyses (e.g. the NCEP[National Centers for Environmental Prediction] re-analysis from 1948 to present). Although these alterna-tive data sets have some real advantages in particularenvironmental or modelling applications (e.g. model-ling malaria; Lindsay et al. 1998; evaluating dust forc-ing; Brooks 1999), we prefer to limit our analysis hereto the use of conventional observed climate data setsderived from surface observations.
The GCM results used in this study are mainlyextracted from the IPCC DDC archive. This archivecontains results from climate change experimentsperformed with 7 coupled ocean-atmosphere globalclimate models (Table 1). All these experiments wereconducted using similar greenhouse gas or green-house gas plus aerosol forcing. In this study only theresults from the greenhouse gas-forced simulations areused for reasons outlined below. We also use resultsfrom the 1400 yr control simulation of the HadCM2 cli-mate model (Tett et al. 1997) to derive model-basedestimates of natural multi-decadal climate variability.The data were re-gridded using a Gaussian space-filter onto a common grid, namely the HadCM2 grid.Later results are presented on this common grid.
Climate can be affected by a number of other agentsin addition to greenhouse gases; important amongstthese are small particles (aerosols). These aerosols aresuspended in the atmosphere and some types (e.g. sul-phate aerosols derived from sulphur dioxide) reflectback solar radiation; hence they have a cooling effecton climate. Although there are no measurements toshow how these aerosol concentrations have changedover the past 150 yr, there are estimates of how sul-phur dioxide emissions (one of the main precursors foraerosol particles) have risen and scenarios of suchemissions into the future. A number of such scenarioshave been used in a sulphur cycle model to calculatethe future rise in sulphate aerosol concentrations (Pen-ner et al. 1998). When one of these scenarios was used,along with greenhouse gas increases, as input to theDDC GCMs, the global-mean temperature rise to 2100was reduced by between a quarter and a third. Thereductions over Africa were less than this.
These are very uncertain calculations, however, dueto a number of factors. First, the old 1992 IPCC emis-sions scenario on which it was based (IS92a; Leggett etal. 1992) contains large rises in sulphur dioxide emis-sions over the next century. Newer emissions scenar-ios, including the draft SRES scenarios, estimate only asmall rise in sulphur dioxide emissions over the nextcouple of decades followed by reductions to levelslower than today’s by 2100 (SRES). Over Africa, sul-phur emissions remain quite low for the whole of thiscentury. The inclusion of such modest sulphur dioxideemissions scenarios in GCM experiments would actu-ally produce a small temperature rise relative to modelexperiments that excluded the aerosol effect (Schles-inger et al. 2000). Results from GCM experimentsusing these revised sulphur scenarios are not yetwidely available. Second, more recent sulphur cyclemodels generate a lower sulphate burden per tonne ofsulphur dioxide emissions and the radiative effect of
148
Table 1. Characteristics of the 7 global climate models available at the IPCC Data Distribution Centre from which experimentalresults were used in this study. Only the greenhouse gas-forced integrations were used here. The climate sensitivity describes theestimated equilibrium global-mean surface air temperature change of each model following a doubling of atmospheric carbon
dioxide concentration
Country of Approximate resolution Climate sensitivity Integration Sourceorigin (lat. × long.) (°C) length
CCSR-NIES Japan 5.62° × 5.62° 3.5 1890–2099 Emori et al. (1999)CGCM1 Canada 3.75° × 3.75° 3.5 1900–2100 Boer et al. (2000) CSIRO-Mk2 Australia 3.21° × 5.62° 4.3 1881–2100 Hirst et al. (2000)ECHAM4 Germany 2.81° × 2.81° 2.6 1860–2099 Roeckner et al. (1996)GFDL-R15 USA 4.50° × 7.50° 3.7 1958–2057 Haywood et al. (1997)HadCM2a UK 2.50° × 3.75° 2.5 1860–2099 Mitchell & Johns (1997)NCAR1 USA 4.50° × 7.50° 4.6 1901–2036 Meehl & Washington (1995)
aAn ensemble of 4 climate change simulations were available from the HadCM2 model
Hulme et al.: African Climate Change: 1900–2100
the sulphate particles in more sophisticated radiationmodels is smaller than previously calculated. Third, inaddition to their direct effect, sulphate aerosols canalso indirectly cool climate by changing the reflectivityand longevity of clouds (Schimel et al. 1996). Theseindirect effects are now realised as being at least asimportant as the direct effect, but were not included inthe present DDC GCM climate change simulations.Fourth, there are other types of aerosols (e.g. carbon orsoot) which may also have increased due to humanactivity, but which act to warm the atmosphere. Finallyand above all, the short lifetime of sulphate particles inthe atmosphere means that they should be seen as atemporary masking effect on the underlying warmingtrend due to greenhouse gases. For all these reasons,model simulations of future climate change using bothgreenhouse gases and sulphate aerosols have not beenused to develop the climate change scenarios illus-trated in this paper.
The future greenhouse gas forcing scenario used inthe DDC experiments approximated a 1% annum–1
growth in greenhouse gas concentrations over theperiod from 1990 to 2100. Since the future growth inanthropogenic greenhouse gas forcing is highly uncer-tain, it is important that our climate scenarios for Africareflect this uncertainty; it would be misleading to con-struct climate change scenarios that reflected just onefuture emissions growth curve. We therefore adopt the4 draft marker emissions scenarios of the IPCC SRES:B1, B2, A1 and A2. None of these emissions scenariosassume any climate policy implementation; the differ-ences result from alternative developments in globalpopulation, the economy and technology. Our methodof climate change scenario construction follows thatadopted by Hulme & Carter (2000) in their generationof climate change scenarios for Europe as part of theACACIA assessment of climate impact in Europe. Fulldetails may be found there, but we provide a shortsummary of the method in Section 5 below.
4. TWENTIETH CENTURY CLIMATE CHANGE
4.1. Temperature
The continent of Africa is warmer than it was 100 yrago. Warming through the 20th century has been atthe rate of about 0.5°C century–1 (Fig. 1), with slightlylarger warming in the June–August (JJA) and Sep-tember–November (SON) seasons than in December–February (DJF) and March–May (MAM). The 6warmest years in Africa have all occurred since 1987,with 1998 being the warmest year. This rate of warm-ing is not dissimilar to that experienced globally, andthe periods of most rapid warming—the 1910s to 1930s
and the post-1970s—occur simultaneously in Africaand the rest of the world.
Few studies have examined long-term changes in thediurnal cycle of temperature in Africa. Here, we showresults for 4 countries for which studies have been pub-lished or data were available for analysis—for Sudanand South Africa as published by Jones & Lindesay(1993) and for Ethiopia and Zimbabwe (unpubl.). Whilea majority of the Earth’s surface has experienced a de-cline in the mean annual diurnal temperature range(DTR) as climate has warmed (Nicholls et al. 1996), ourexamples here show contrasting trends for these 4African countries. Mean annual DTR decreased by be-tween 0.5 and 1°C since the 1950s in Sudan andEthiopia, but increased by a similar amount in Zim-babwe (Fig. 2). In South Africa, DTR decreased duringthe 1950s and 1960s, but has remained quite stablesince then. Examination of the seasonal variation inthese trends (not shown) suggests that different factorscontribute to DTR trends in different seasons and in dif-ferent countries. For example, in Sudan DTR shows an
149
Annual
-0.6
0
0.6
DJF
-0.6
0
0.6
MAM
-0.6
0
0.6
Me
an
tem
per
atur
e a
nom
aly
(de
gC)
JJA
-0.6
0
0.6
SON
-0.6
0
0.6
1900 1920 1940 1960 1980 2000
1900 1920 1940 1960 1980 2000
Fig. 1. Mean surface air temperature anomalies for the Africancontinent, 1901–98, expressed with respect to the 1961–90 average;annual and 4 seasons—DJF, MAM, JJA, SON. The smooth curves
result from applying a 10 yr Gaussian filter
Clim Res 17: 145–168, 2001
increasing trend during the July–September wetseason, probably caused by trends towards reducedcloudiness, while DTR decreased during the rest of theyear, probably due to trends for increased dustiness(Brooks 1999). Both of these factors are related to themulti-decadal drought experienced in Sudan since the1950s (Hulme 2001). The long-term increase in annualDTR in Zimbabwe is due almost entirely to increasesduring the November–February wet season; trendsduring the rest of the year have been close to zero. Weare not aware of published analyses of diurnal tem-perature trends in other African countries.
4.2. Rainfall
Interannual rainfall variability is large over most ofAfrica and for some regions, most notably the Sahel,multi-decadal variability in rainfall has also beensubstantial. Reviews of 20th Century African rainfallvariability have been provided by, among others,Janowiak (1988), Hulme (1992) and Nicholson (1994).
To illustrate something of this variability we present ananalysis for the 3 regions of Africa used by Hulme(1996b)—the Sahel, East Africa and southeast Africa(domains shown in Fig. 4). These 3 regions exhibit con-trasting rainfall variability characteristics (Fig. 3): theSahel displays large multi-decadal variability withrecent drying, East Africa a relatively stable regimewith some evidence of long-term wetting, and south-east Africa also a basically stable regime, but withmarked inter-decadal variability. In recent years Sahelrainfall has been quite stable around the 1961–90annual average of 371 mm, although this 30 yr periodis substantial drier (about 25%) than earlier decadesthis century. In East Africa, 1997 was a very wet yearand, as in 1961 and 1963, led to a surge in the level ofLake Victoria (Birkett et al. 1999). Recent analyses (Sajiet al. 1999, Webster et al. 1999) have suggested theseextreme wet years in East Africa are related to a dipole
150
Sudan
-1
0
1
-1
0
1
Ethiopia
-1
0
1
-1
0
1
deg
C a
nom
aly
Zimbabwe
-1
0
1
-1
0
1
South Africa
-1
0
1
-1
0
1
1940 1950 1960 1970 1980 1990 2000
1940 1950 1960 1970 1980 1990 2000
Fig. 2. Mean annual diurnal temperature range (Tmax – Tmin)for a number of African countries: Sudan (data end 1987),Ethiopia (1990), Zimbabwe (1997) and South Africa (1991).The smooth curves result from applying a 10 yr Gaussianfilter. Data for Sudan and South Africa are from Jones &Lindesay (1993), while data for Ethiopia and Zimbabwe are
unpublished
The Sahel
0.0
+0.2
+0.4
-0.2
-0.4
-80
-60
-40
-20
0
20
40
60
80
East Africa
0.0
+0.2
+0.4
-0.2
-0.4
-80
-60
-40
-20
0
20
40
60
80
A Ann
ual t
empe
ratu
re a
nom
aly
(deg
C)
nnu
al r
ain
all
ano
mal
y (p
er c
ent)
Southeast Africa
0.0
+0.2
+0.4
-0.2
-0.4
-80
-60
-40
-20
0
20
40
60
80
1880 1900 1920 1940 1960 1980 2000
1880 1900 1920 1940 1960 1980 2000
Fig. 3. Annual rainfall (1900–98; histograms and bold line) andmean temperature anomalies (1901–98; dashed line) for 3 Africanregions, expressed with respect to the 1961–90 average: Sahel,East Africa and southeastern Africa (regional domains marked onFig. 4). Note: for southeast Africa year is July to June. The smoothcurves for rainfall and temperature result from applying a 10 yr
Gaussian filter
Hulme et al.: African Climate Change: 1900–2100
mode of variability in the Indian Ocean. In southeastAfrica, the dry years of the early 1990s were followedby 2 very wet years in 1995/96 and 1996/97. Mason etal. (1999) report an increase in recent decades in thefrequency of the most intense daily precipitation overSouth Africa, even though there is little long-termtrend in total annual rainfall amount.
Fig. 3 also displays the trends in annual temperaturefor these same 3 regions. Temperatures for all 3 regionsduring the 1990s are higher than they were earlier in thecentury (except for a period at the end of the 1930s in theSahel) and are currently between 0.2 and 0.3°C warmerthan the 1961–90 average. There is no simple correlationbetween temperature and rainfall in these 3 regions, al-though Hulme (1996b) noted that drying in the Sahelwas associated with a moderate warming trend.
4.3. Spatial patterns
Our analysis is summarised further in Fig. 4, wherewe show mean linear trends in annual temperatureand precipitation during the 20th century. This ana-lysis first filters the data using a 10-point Gaussianfilter to subdue the effects on the regression analysisof outlier values at either end of the time period.While warming is seen to dominate the continent (seeFig. 1 above), some coherent areas of cooling arenoted, around Nigeria/Cameroon in West Africa andalong the coastal margins of Senegal/Mauritania andSouth Africa. In contrast, warming is at a maximumof nearly 2°C century–1 over the interior of southernAfrica and in the Mediterranean countries of north-west Africa.
151
Mean Annual Temperature
15W 0 15E 30E 45E
30S
15S
0
15N
30N
-3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0oC
15W 0 15E 30E 45E
30S
15S
0
15N
30N
Annual Precipitation
15W 0 15E 30E 45E
30S
15S
0
15N
30N
-60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60
Percent
Fig. 4. Mean linear trends in annual temperature (°C century–1) and annual rainfall (% century–1), calculated over the period1901–95 from the New et al. (1999, 2000) data set. Data were filtered with a 10-point Gaussian filter before being subject to
regression analysis. The 3 regions shown are those depicted in Figs 3 & 13
Clim Res 17: 145–168, 2001
The pattern of rainfall trends (Fig. 4) reflects theregional analysis shown in Fig. 3, with drying of up to25% century–1 or more over some western and easternparts of the Sahel. More moderate drying—5 to 15%century–1—is also noted along the Mediterraneancoast and over large parts of Botswana and Zimbabweand the Transvaal in southeast Africa. The modestwetting trend noted over East Africa is seen to be partof a more coherent zone of wetting across most ofequatorial Africa, in some areas of up to 10% century–1
or more. Regions along the Red Sea coast have alsoseen an increase in rainfall, although trends in thisarid/semi-arid region are unlikely to be very robust.
4.4. ENSO influence on rainfall
With regard to interannual rainfall variability inAfrica, the ENSO is one of the more important control-ling factors, at least for some regions (Ropelewski &Halpert 1987, 1989, 1996, Janowiak 1988, Dai &Wigley 2000). These studies have established that the2 regions in Africa with the most dominant ENSOinfluences are in eastern equatorial Africa during theshort October-November rainy season and in south-eastern Africa during the main November–Februarywet season. Ropelewski & Halpert (1989) also exam-ined Southern Oscillation and rainfall relationshipsduring La Niña or high index years. We have con-ducted our own more general analysis of SouthernOscillation rainfall variability for the African regionover the period 1901–98 using an updated and morecomprehensive data set (Hulme 1994b) than was usedby these earlier studies. We also use the Southern Os-cillation Index (SOI) as a continuous index of SouthernOscillation behaviour rather than designating discrete‘warm’ (El Niño; low index) and ‘cold’ (La Niña; highindex) Southern Oscillation events as was done byRopelewski & Halpert (1996).
We defined an annual average SOI using the June–May year, a definition that maximises the coherence of
individual Southern Oscillation events, and correlatedthis index against seasonal rainfall in Africa. We per-formed this analysis for the 4 conventional seasons (notshown) and also for the 2 extended seasons of June toOctober (Year 0) and November (Year 0) to April (Year1; Fig. 5a). This analysis confirms the strength of thepreviously identified relationships for equatorial eastAfrica (high rainfall during a warm ENSO event) andsouthern Africa (low rainfall during a warm ENSOevent). The former relationship is strongest during theSeptember–November rainy season (the ‘short’ rains;not shown), with an almost complete absence of ENSOsensitivity in this region during the February–Aprilseason (‘long’ rains) as found by Ropelewski & Halpert(1996). The southern African sensitivity is strongestover South Africa during December–February beforemigrating northwards over Zimbabwe and Mozam-bique during the March–May season (not shown).There is little rainfall sensitivity to ENSO behaviourelsewhere in Africa, although weak tendencies forSahelian June–August drying (Janicot et al. 1996) andnorthwest African March–May drying (El Hamly etal. 1998) can also be found.
5. TWENTY-FIRST CENTURY CLIMATE CHANGE
For a comprehensive assessment of the impact andimplications of climate change, it is necessary to applya number of climate change scenarios that span a rea-sonable range of the likely climate change distribution.The fact that there is a distribution of future climatechanges arises not only because of incomplete under-standing of the climate system (e.g. the unknown valueof the climate sensitivity, different climate model re-sponses, etc.), but also because of the inherent unpre-dictability of climate (e.g. unknowable future climateforcings and regional differences in the climate systemresponse to a given forcing because of chaos). The‘true’ climate change distribution is of course un-known, but we can make some sensible guesses as to
152
Table 2. The 4 climate scenarios. Estimates shown here are for the 2050s (i.e., 2055), but the values for the 2020s and 2080s werealso calculated. Temperature and sea-level changes assume no aerosol effects and are calculated from a 1961–90 baseline usingthe MAGICC climate model (Wigley & Raper 1992, Raper et al. 1996, Wigley et al. 2000). C is annual carbon emissions from fossilenergy sources, S is annual sulphur emissions, ∆T is change in mean annual temperature, ∆SL is change in mean sea-level and
pCO2 is the atmospheric carbon dioxide concentration
Scenario/ Population C emissions from Total S Global ∆T Global ∆SL pCO2
Climate sensitivity (billions) energy (GtC) emissions (TgS) (°C) (cm) (ppmv)
B1-low / 1.5°C 8.76 9.7 51 0.9 13 479B2-mid / 2.5°C 9.53 11.3 55 1.5 36 492A1-mid / 2.5°C 8.54 16.1 58 1.8 39 555A2-high / 4.5°C 11.67 17.3 96 2.6 68 559
Hulme et al.: African Climate Change: 1900–2100
its magnitude and shape and then make some choicesso as to sample a reasonable part of its range.
We have done this at a global scale by making choicesabout future greenhouse gas forcings and about theclimate sensitivity (see Table 1 for definition). We fol-low Hulme & Carter (2000) and Carter et al. (2001) inthis procedure, yielding the 4 global climate scenariosshown in Table 2. We have chosen the SRES A2 emis-
sions scenario combined with a high climate sensitivity(4.5°C), SRES A1 and SRES B2 combined with mediumclimate sensitivities (2.5°C) and SRES B1 combinedwith a low climate sensitivity (1.5°C). These 4 scenariosare subsequently termed A2-high, A1-mid, B2-midand B1-low, respectively, and yield a range of globalwarming by the 2050s of 0.9 to 2.6°C. We chose the2 middle cases deliberately because, even though the
153
-1
-0.5
-0.4
-0.3
-0.26
0.26
0.3
0.4
0.5
1
0 30E
30S
EQ
30N
Time series correlations of SOI (Jun-May) vs. Precip (Jun-Oct)
-1
-0.5
-0.4
-0.3
-0.26
0.26
0.3
0.4
0.5
1
0 30E
30S
EQ
30N
Time series correlations of SOI (Jun-May) vs. Precip (Nov-Apr)
a b
-1
-0.5
-0.4
-0.3
-0.17
0.17
0.3
0.4
0.5
1
0 30E
30S
EQ
30N
Time series correlations of SOI (Jun-May) vs. Precip (Jun-Oct)
-1
-0.5
-0.4
-0.3
-0.17
0.17
0.3
0.4
0.5
1
0 30E
30S
EQ
30N
Time series correlations of SOI (Jun-May) vs. Precip (Nov-Apr)
c d
Fig. 5. Correlation between annual (June–May) Southern Oscillation Index (SOI) and seasonal rainfall; (a,c) June–October(Year 0) rainfall; (b,d) November–April (Year 0 to +1) rainfall. (a,b) Observed relationship over the period 1901–98; (c,d) HadCM2model-simulated relationship over a 240 yr unforced simulation. Correlations are only plotted where they are significant at 95%
and in regions where the respective seasonal rainfall is greater than 20 mm and greater than 20% of annual total
a b
c d
Clim Res 17: 145–168, 2001
global warming is similar, the worlds which underliethe B2 and A1 emissions scenarios are quite different(SRES). The impacts on Africa of what may be rathersimilar global and regional climate changes could bequite different in these 2 cases. For example, global(and African) population is lower in the A1 world thanin the B2 world, but carbon and sulphur emissions andCO2 concentrations are higher (Table 2).
Having defined these 4 global climate scenarios, wenext consider the range of climate changes for Africathat may result from each of these 4 possible futures.Again, we have a distribution of possible regional out-comes for a given global warming. We use results fromthe 7 GCM experiments (see Table 1) to define thisrange. (Note: for HadCM2 there are 4 simulations forthe same scenario thus the total GCM sample availableto us is 10; this gives more weight in our final scenariosto the HadCM2 responses than to the other 6 GCMs.)We present the scenario results for seasonal mean tem-perature and precipitation for the 2020s, 2050s and2080s in 2 different ways: Africa-wide maps andnational-scale summary results for 4 representativecountries within Africa.
5.1. African scenario maps
The construction of the scenario maps follows theapproach of Hulme & Carter (2000) and Carter et al.(2001). We first standardise the 2071–2100 climateresponse patterns—defined relative to the 1961–90model average—in the DDC GCMs using the globalwarming values in each respective GCM. These stan-dardised climate response patterns are then scaled bythe global warming values for our 4 scenarios and 3time periods calculated by the MAGICC climate model(see Table 2). Scaling of GCM response patterns in thisway assumes that local greenhouse gas-induced cli-mate change is a linear function of global-mean tem-perature. (See Mitchell et al. 1999 for a discussion ofthis asumption). Only a selection of the full set of mapsis shown here. For each scenario, season, variable andtime slice we present 2 maps representing the changein mean seasonal climate for the respective 30 yrperiod (Figs 6 to 11). One map shows the Medianchange from our sample of 10 standardised and scaledGCM responses (left panels) and the other map showsthe absolute Range of these 10 model responses (rightpanels).
We also introduce the idea of signal:noise ratios bycomparing the Median GCM change with an estimateof natural multi-decadal climate variability. In themaps showing the Median change we only plot thesevalues where they exceed the 1 standard deviationestimate of natural 30 yr time-scale climate variability.
These estimates were extracted from the 1400 yrunforced simulation of the HadCM2 model (Tett et al.1997). We use a climate model simulation to quantifythe range of natural climate variability rather thanobservations because the model gives us longer andmore comprehensive estimates of natural climate vari-ability. This has the disadvantage that the climatemodel may not accurately simulate natural climatevariability, although at least for some regions and onsome time-scales, HadCM2 yields estimates of naturalvariability quite similar both to observations (Tett et al.1997) and to climatic fluctuations reconstructed fromproxy records over the past millennium (Jones et al.1998). We discuss this problem further in Section 6.
The resulting African scenario maps are thereforeinformative at a number of levels:
• Africa-wide estimates are presented of mean sea-sonal climate change (mean temperature and pre-cipitation) for the 4 adopted climate change sce-narios;
• These estimates are derived from a sample (apseudo-ensemble) of 10 different GCM simulations,rather than being dependent on any single GCM orGCM experiment;
• Only Median changes that exceed what may reason-ably be expected to occur due to natural 30 yr time-scale climate variability are plotted;
• The extent of inter-model agreement is depictedthrough the Range maps.
For our scenarios, future annual warming acrossAfrica ranges from below 0.2°C decade–1 (B1-low sce-nario; Fig. 6) to over 0.5°C decade–1 (A2-high; Fig. 7).This warming is greatest over the interior semi-aridtropical margins of the Sahara and central southernAfrica, and least in equatorial latitudes and coastalenvironments. The B2-mid and A1-mid scenarios (notshown) fall roughly in between these 2 extremes. All ofthe estimated temperature changes exceed the 1 sigmalevel of natural temperature variability (as defined byunforced HadCM2 simulation), even in the B1-low sce-nario. The inter-model range (an indicator of the extentof agreement between different GCMs) is smallestover northern Africa and the Equator and greatest overthe interior of central southern Africa. For example, theinter-model range falls to less than 25% of the modelmedian response in the former regions, but rises toover 60% of the model median response in the latterareas.
Future changes in mean seasonal rainfall in Africaare less well defined. Under the B1-low scenario, rela-tively few regions in Africa experience a change ineither DJF or JJA rainfall that exceeds the 1 sigmalevel of natural rainfall variability simulated by theHadCM2 model (Figs 8 & 9). The exceptions are parts
154
Hulme et al.: African Climate Change: 1900–2100 155
0.90.91.01.01.01.01.01.01.01.00.90.8
0.90.90.91.01.01.01.01.01.00.90.90.8
0.80.90.91.01.01.01.01.01.00.90.80.80.7
0.80.80.90.91.01.01.01.00.90.80.80.80.70.80.7
0.80.80.90.91.00.90.90.90.90.80.80.80.80.80.80.80.80.90.90.90.80.80.8
0.80.80.80.90.90.90.90.90.90.80.80.80.80.80.90.91.01.01.01.11.01.00.90.80.80.7
0.70.80.80.90.90.90.90.90.90.80.80.80.80.80.90.91.01.01.11.11.11.11.11.00.90.8
0.70.80.80.90.90.90.90.90.90.80.80.80.80.80.90.90.91.01.01.01.01.01.01.00.90.8
0.70.80.90.90.90.90.90.90.90.80.80.80.70.70.80.80.90.90.91.01.01.00.90.80.8
0.80.90.90.90.90.90.90.90.90.80.70.70.70.70.70.80.80.80.90.90.90.80.8
0.80.80.90.90.9
0.90.90.90.80.80.70.60.60.60.60.70.70.70.70.7
0.90.90.91.01.01.01.0
0.80.80.70.70.60.6
0.90.90.90.90.91.01.00.90.90.80.80.70.70.6
0.70.6
0.90.9
0.90.91.00.90.8
0.70.7
0.60.60.6
0.80.80.80.80.8
0.80.90.91.01.01.00.90.90.8
0.90.90.91.01.01.01.01.01.00.90.80.7
1.01.01.01.01.01.01.01.01.01.00.90.80.7
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2020s
0.50.30.20.30.30.40.40.50.50.50.50.4
0.40.30.20.20.30.40.40.50.50.50.50.4
0.30.30.20.20.30.30.40.50.50.50.40.40.4
0.20.20.20.30.30.40.50.40.40.40.40.40.40.40.4
0.20.20.30.30.40.50.50.50.40.40.50.50.50.50.50.50.50.50.50.50.50.40.3
0.20.20.30.30.30.50.50.50.50.50.50.40.40.50.50.50.50.60.60.60.60.60.50.40.40.4
0.30.30.30.30.30.50.60.50.50.50.40.40.40.40.50.50.50.60.70.70.70.60.50.50.50.5
0.30.30.20.20.30.60.60.60.50.40.40.40.40.40.40.50.50.60.70.70.70.60.50.40.40.4
0.40.30.20.20.30.60.60.50.50.40.50.40.50.50.60.50.60.60.70.70.70.60.40.40.3
0.30.20.20.30.40.60.60.60.40.50.50.50.50.50.40.50.50.60.70.70.60.50.4
0.30.30.20.20.2
0.60.70.60.40.40.50.50.40.20.20.30.40.50.50.5
0.40.40.20.10.30.50.5
0.70.60.60.40.30.3
0.40.40.20.20.30.50.60.70.70.70.70.50.30.3
0.40.4
0.40.4
0.40.60.60.60.5
0.50.4
0.40.40.4
0.50.40.30.30.4
0.40.50.50.40.40.50.60.50.3
0.40.40.40.40.40.30.40.50.60.50.40.4
0.50.40.40.40.40.30.30.40.50.60.50.40.4
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
1.41.41.51.51.61.61.61.61.61.51.41.2
1.41.41.51.51.51.61.61.61.51.51.31.2
1.31.41.41.51.51.51.51.51.51.41.31.21.1
1.31.31.41.51.51.51.51.51.41.31.31.21.21.21.2
1.21.31.41.51.51.51.51.41.31.31.31.21.21.31.31.31.31.41.41.41.31.31.2
1.21.31.31.41.51.51.51.41.31.31.31.21.21.31.41.51.51.61.61.61.61.51.41.31.21.1
1.11.21.31.41.41.41.51.41.31.31.31.21.21.31.41.41.61.61.71.71.71.71.61.61.41.3
1.11.21.31.41.41.41.41.51.41.31.21.21.21.21.31.41.51.51.61.61.61.61.61.51.41.2
1.21.31.31.41.41.41.51.51.41.31.31.21.11.21.21.31.41.41.51.51.51.51.41.31.3
1.31.41.41.41.41.51.41.41.31.21.11.11.01.01.11.21.31.31.31.41.41.31.2
1.31.31.41.51.5
1.41.41.31.31.21.11.01.01.01.01.01.11.11.21.2
1.41.41.41.51.51.51.5
1.31.31.21.11.01.0
1.41.51.41.41.51.61.61.51.41.31.21.21.11.0
1.11.0
1.51.4
1.41.51.61.51.3
1.11.1
0.91.01.0
1.21.31.31.31.3
1.31.31.51.61.61.51.51.41.2
1.41.41.51.51.61.61.61.61.51.41.31.2
1.51.51.51.51.61.61.61.61.61.51.41.31.2
30S
20S
10S
EQ
10N
20N
30N
2050s
0.70.50.30.40.50.60.60.80.80.90.80.7
0.60.40.40.40.50.60.70.70.80.80.80.6
0.50.40.40.30.40.50.60.70.70.80.70.60.6
0.30.40.30.40.50.60.70.70.70.70.60.70.70.70.6
0.40.40.40.50.60.70.70.70.70.70.70.70.70.70.80.70.70.80.90.80.80.70.5
0.30.40.40.50.50.80.80.80.70.70.70.70.70.70.80.80.80.90.91.01.00.90.80.60.60.6
0.40.50.40.40.50.80.90.80.80.70.70.70.60.60.70.80.80.91.01.11.01.00.80.80.70.7
0.50.50.30.30.50.90.90.90.80.70.70.60.60.60.70.70.81.01.11.11.11.00.80.70.70.7
0.60.40.30.40.50.90.90.80.80.70.80.70.80.80.90.80.91.01.11.11.00.90.70.60.5
0.50.30.30.50.60.91.00.90.60.70.80.80.80.80.70.70.81.01.01.00.90.80.6
0.50.50.30.20.3
0.91.10.90.60.60.80.70.60.40.30.50.60.80.80.8
0.60.60.30.20.40.80.8
1.10.90.90.60.40.5
0.70.60.30.30.50.91.01.11.11.01.10.70.40.4
0.70.7
0.70.6
0.60.90.90.90.8
0.80.6
0.70.70.6
0.80.70.50.50.6
0.70.80.70.60.60.80.90.70.5
0.60.60.70.60.50.50.70.80.90.80.60.6
0.70.70.60.60.60.50.50.70.70.90.80.60.6
30S
20S
10S
EQ
10N
20N
30N
1.81.81.91.91.92.02.02.01.91.81.71.5
1.71.81.81.91.91.91.91.91.91.81.71.5
1.61.71.81.91.91.91.91.91.91.81.61.51.4
1.61.61.71.81.91.91.91.81.81.61.61.51.41.51.4
1.51.61.71.81.81.81.81.81.61.61.61.51.51.61.61.61.61.71.71.71.61.61.5
1.51.61.61.81.81.81.81.81.61.61.61.51.51.61.71.81.92.02.02.02.01.91.81.61.51.3
1.41.51.61.71.81.81.81.81.71.61.61.51.51.61.71.81.92.02.02.12.12.12.01.91.71.6
1.41.51.61.71.71.71.81.81.71.61.51.51.51.51.71.81.81.91.92.02.02.02.01.91.71.5
1.41.61.71.71.71.71.81.81.71.61.61.51.41.41.51.61.71.71.81.91.91.91.81.61.5
1.61.71.71.71.71.81.81.71.71.51.41.31.31.31.41.51.61.61.71.71.71.61.5
1.61.61.71.81.8
1.71.71.71.61.51.31.21.21.21.21.31.41.41.41.4
1.71.71.81.81.91.91.8
1.61.61.41.31.21.2
1.81.81.81.71.81.91.91.81.71.61.51.41.31.2
1.31.2
1.81.8
1.71.81.91.81.6
1.41.3
1.11.21.2
1.51.61.61.61.6
1.61.71.81.91.91.91.81.71.5
1.81.81.81.82.02.02.02.01.91.81.61.4
1.91.91.91.91.92.02.02.02.01.91.81.61.4
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2080s
oC change
0.5 1 1.5 2 3 4 5
0.90.60.40.50.60.70.80.91.01.11.00.8
0.80.60.40.50.60.70.80.90.91.00.90.8
0.60.50.40.40.50.70.80.90.90.90.90.80.7
0.40.50.40.50.60.70.90.80.90.80.80.90.80.80.8
0.40.40.50.60.70.90.90.90.80.90.90.90.90.90.90.90.91.01.11.01.00.80.7
0.40.50.50.60.60.91.00.90.90.90.90.90.90.90.91.01.01.11.11.21.21.11.00.80.80.7
0.50.60.50.50.61.01.11.01.00.90.80.80.80.80.90.91.01.11.31.31.31.21.01.00.90.9
0.70.60.40.40.61.11.21.11.00.90.80.70.70.70.80.91.01.21.31.41.31.20.90.90.80.8
0.70.50.40.50.71.11.21.01.00.91.00.81.01.01.11.01.21.21.41.41.31.10.90.80.7
0.60.40.40.60.71.21.21.10.70.91.00.91.00.90.90.91.01.21.31.31.10.90.8
0.70.60.40.30.3
1.21.41.10.80.71.00.90.70.40.40.60.71.01.01.0
0.70.70.40.30.50.91.0
1.31.21.10.70.50.6
0.80.70.40.40.61.11.21.31.31.31.30.90.60.5
0.80.8
0.80.7
0.71.11.11.11.0
1.00.7
0.80.80.8
0.90.80.60.60.7
0.91.00.90.80.70.91.20.90.6
0.70.70.80.80.70.60.80.91.21.00.70.7
0.90.90.80.80.70.70.70.80.91.10.90.70.7
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
oC change
0.5 1 1.5 2 3 5
Fig. 6. Left panels: change in mean annual temperature for the 2020s, 2050s and 2080s (with respect to 1961–90) for the B1-lowscenario; median of 7 GCM experiments. Right panels: inter-model range in mean annual temperature change. See text for
further explanation. Selected domains in the top left panel are the 4 ‘national’ regions used in Fig. 12
Clim Res 17: 145–168, 2001156
2.12.12.22.22.32.32.32.32.32.12.01.8
2.02.02.12.22.22.32.32.22.22.11.91.7
1.92.02.12.22.22.22.22.22.22.11.81.81.6
1.81.92.02.12.22.22.22.12.11.91.81.71.71.71.7
1.81.92.02.12.22.12.12.11.91.81.81.81.81.91.91.91.92.02.02.01.91.81.7
1.71.81.92.12.12.12.12.11.91.91.81.81.81.92.02.12.22.32.42.42.32.22.11.91.71.6
1.61.81.92.02.12.12.12.11.91.81.81.81.81.82.02.12.22.32.42.42.42.42.42.22.01.8
1.61.81.92.02.02.02.12.12.01.91.81.81.81.81.92.12.12.22.32.32.42.42.32.22.01.8
1.71.81.92.02.02.02.12.12.01.91.81.71.61.71.81.92.02.02.12.22.22.22.11.91.8
1.82.02.02.02.02.12.12.01.91.81.61.61.51.51.61.71.81.91.92.02.01.91.8
1.91.92.02.12.1
2.02.01.91.91.71.51.41.41.41.41.51.61.71.71.7
2.02.02.12.22.22.22.2
1.91.81.71.51.51.4
2.12.12.12.02.12.22.32.12.01.81.71.71.51.4
1.51.5
2.12.1
2.02.12.32.11.9
1.61.6
1.31.41.4
1.81.91.91.91.8
1.91.92.12.22.32.22.12.01.7
2.12.12.12.12.32.32.42.32.22.11.91.7
2.22.22.22.22.32.32.32.32.32.22.11.91.7
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2020s
1.00.80.50.60.70.80.91.11.11.21.21.0
0.90.60.50.60.70.80.91.11.11.11.10.9
0.80.60.50.50.60.80.91.01.11.11.00.90.9
0.50.60.50.60.70.91.01.01.00.90.91.00.91.00.9
0.50.50.60.70.81.11.11.01.01.01.11.01.01.11.11.01.01.21.21.21.11.00.8
0.50.60.60.70.71.11.21.11.11.01.01.01.01.01.11.11.21.31.31.41.41.31.10.90.90.8
0.60.70.60.60.71.21.31.21.21.01.00.90.90.91.01.11.21.31.51.51.51.41.21.21.11.1
0.80.70.50.50.71.31.41.21.21.01.00.90.80.91.01.11.21.41.51.61.51.41.11.01.01.0
0.80.60.40.50.81.31.41.21.11.01.21.01.11.21.21.21.41.41.61.61.51.31.00.90.8
0.70.40.40.70.91.41.41.30.91.01.11.11.21.11.01.01.21.41.51.51.31.10.9
0.80.80.50.30.4
1.41.61.30.90.81.11.10.80.50.50.70.91.11.21.1
0.90.80.40.30.61.11.2
1.51.41.20.80.60.7
0.90.80.50.40.71.21.41.51.51.51.51.00.60.6
0.91.0
1.00.8
0.91.31.31.31.1
1.10.8
0.91.00.9
1.10.90.70.70.8
1.01.11.10.90.91.11.41.00.7
0.80.81.00.90.80.71.01.11.41.20.80.9
1.11.00.90.90.90.80.81.01.11.31.10.90.9
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
3.93.94.14.24.34.44.34.34.34.13.83.4
3.73.94.04.24.24.34.34.24.24.03.63.3
3.53.73.94.14.14.24.24.14.13.93.53.33.1
3.43.63.84.04.14.14.14.13.93.63.43.33.23.23.2
3.43.63.74.04.14.04.03.93.63.53.53.33.43.63.63.53.63.73.73.73.63.43.3
3.23.53.63.94.04.04.03.93.63.63.53.43.43.63.84.04.14.34.54.54.44.23.93.53.33.0
3.13.43.63.83.93.94.03.93.73.43.43.43.43.43.73.94.24.44.54.64.64.64.54.23.83.4
3.13.43.63.73.83.83.94.03.73.53.43.33.33.43.63.94.04.14.34.44.54.44.34.23.83.3
3.23.53.73.73.83.84.04.03.83.63.43.33.13.13.43.63.73.84.04.14.24.13.93.63.4
3.53.83.83.83.83.93.93.83.63.43.12.92.82.83.03.23.43.63.63.73.83.63.3
3.63.63.84.04.0
3.83.83.73.63.22.92.72.72.72.72.83.03.13.13.2
3.73.83.94.14.24.14.1
3.63.43.22.92.72.7
3.94.03.93.84.04.24.24.03.73.53.33.22.92.7
2.92.7
4.03.9
3.74.04.34.03.5
3.02.9
2.52.62.7
3.33.53.63.63.5
3.53.64.04.24.34.24.03.73.3
3.93.94.04.04.34.44.54.44.23.93.63.1
4.14.14.24.14.34.44.44.44.34.23.93.53.2
30S
20S
10S
EQ
10N
20N
30N
2050s
2.01.40.91.11.31.61.72.12.22.32.21.9
1.81.21.01.01.21.61.82.02.12.12.11.8
1.41.11.00.91.21.51.71.92.02.11.91.71.6
0.91.10.91.11.31.61.91.81.91.81.71.91.81.81.8
1.01.01.21.41.62.02.01.91.81.92.02.01.92.02.11.92.02.22.32.32.11.81.5
0.91.01.21.31.42.12.22.12.02.01.91.91.91.92.12.12.22.42.52.72.62.52.21.81.71.6
1.11.31.11.11.22.22.42.32.22.01.91.81.71.81.92.12.22.52.82.92.82.72.32.22.02.0
1.41.30.90.91.32.42.62.32.21.91.91.61.51.61.82.02.22.62.93.02.92.62.11.91.81.8
1.51.10.81.01.52.42.62.32.11.92.21.82.12.22.32.22.62.73.03.02.82.51.91.71.5
1.20.80.81.31.62.52.72.41.61.92.12.12.22.11.92.02.22.72.82.82.52.11.7
1.51.40.90.70.8
2.63.02.51.71.62.12.01.51.00.91.31.62.12.32.2
1.61.50.80.61.12.12.3
2.92.52.41.61.11.2
1.81.60.90.81.42.32.62.92.92.82.92.01.21.2
1.81.8
1.81.6
1.62.52.42.52.2
2.11.6
1.81.81.7
2.11.81.31.41.5
1.92.12.01.71.62.12.61.91.4
1.51.61.81.71.51.41.82.02.62.31.61.6
2.01.91.71.71.61.51.51.92.02.42.01.61.6
30S
20S
10S
EQ
10N
20N
30N
6.06.06.36.46.66.76.66.66.66.25.85.2
5.75.96.26.46.56.66.66.56.46.15.65.0
5.45.76.06.36.36.46.46.46.25.95.35.14.7
5.35.55.96.16.36.36.36.26.05.55.25.04.84.94.9
5.25.55.86.26.26.26.26.05.65.35.35.15.25.55.55.45.55.75.75.75.55.25.0
4.95.35.55.96.16.16.15.95.65.55.35.25.25.55.86.16.36.66.96.96.86.55.95.45.04.5
4.75.15.55.85.95.96.16.05.65.25.35.25.25.35.76.06.56.76.97.07.07.06.96.55.85.3
4.75.25.55.75.85.86.06.15.85.45.25.15.15.25.65.96.26.36.56.76.86.86.76.45.95.1
4.95.35.65.75.85.96.16.15.85.55.35.04.74.85.25.55.75.96.16.36.46.36.05.55.2
5.35.85.95.85.86.16.15.95.65.24.84.54.34.44.65.05.25.45.65.75.85.55.1
5.55.55.86.16.1
5.85.85.65.54.94.54.24.14.14.14.44.64.84.84.9
5.75.86.06.26.46.36.2
5.55.34.94.54.24.1
6.06.16.05.86.16.56.56.15.75.35.04.84.54.2
4.44.2
6.16.0
5.76.16.56.15.4
4.64.5
3.84.04.1
5.15.45.55.55.3
5.45.66.16.56.66.46.15.75.0
6.06.06.16.26.66.86.96.76.46.05.54.8
6.36.36.46.36.56.76.86.86.76.56.05.44.9
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2080s
oC change
0.5 1 1.5 2 3 4 5
3.02.21.41.61.92.42.73.23.33.63.42.8
2.71.91.51.61.92.42.73.13.23.33.22.7
2.21.71.51.51.82.32.62.93.13.12.92.62.5
1.41.61.41.72.02.53.02.82.92.72.72.92.72.82.7
1.51.51.82.12.43.13.12.92.82.93.13.02.93.13.23.03.03.43.63.53.32.82.2
1.31.61.92.02.13.23.33.23.13.03.02.92.93.03.23.33.43.63.94.14.03.93.32.72.52.4
1.71.91.71.71.93.43.73.53.33.02.92.72.62.73.03.23.43.94.44.54.34.13.53.43.13.0
2.22.01.41.32.03.73.93.63.42.92.82.52.32.52.83.13.44.04.44.64.44.03.22.92.82.8
2.31.71.21.52.33.73.93.53.22.93.42.83.23.43.63.43.94.14.64.64.33.82.92.62.3
1.91.21.32.02.53.94.23.62.52.93.23.23.43.22.93.03.44.14.44.33.83.22.5
2.32.21.41.01.2
4.04.63.92.72.43.23.12.41.51.41.92.53.33.53.3
2.52.31.30.91.73.23.5
4.43.93.62.41.71.9
2.72.41.31.22.13.64.04.54.44.34.43.01.91.8
2.72.8
2.82.4
2.53.93.73.83.3
3.22.4
2.72.82.5
3.22.72.12.12.4
2.93.23.12.72.53.24.03.02.1
2.42.42.82.62.32.12.83.13.93.52.42.5
3.02.92.62.62.52.22.22.83.03.63.12.52.5
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
oC change
0.5 1 1.5 2 3 4 5
Fig. 7. Left panels: Change in mean annual temperature for the 2020s, 2050s and 2080s (with respect to 1961–90) for the A2-highscenario; median of 7 GCM experiments. Right panels: Inter-model range in mean annual temperature change. See text for
further explanation
Hulme et al.: African Climate Change: 1900–2100 157
2 321
4542
6553
101110
-6
1013
12
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2020s
1824327662117615866504127
183331476662325655483925
174029433848294448103352526
131643333626379882423130211117
13275235235777403131301778
16222818109
101618
14213132515455612530301454
121311978768
131160
16573336349299632130291476131399878671079
153232346143
1186620272715107111178876679710
1432414141985940192221121010998898787108
1430292968
1267635171414191916151311111010979
1314141829
644920182231463144412013151313
13151618172936
553948866444
1214171814182740388265825968
1314
1012
1517286046
3024
999
1917223440
293628223737483024
222624402938435040312725
16204237476451455945383525
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
19 20
5
23
7
4
5
3
6
4532
-3
767642
-3
-3
89885
-3-3-5
141617151173
-9-8
162121181614136
222419
18
145
30S
20S
10S
EQ
10N
20N
30N
2050s
283850118971839691103786442
2852487410496508886756239
276345685974466975161554040
21266751564158154128664846321727
21438155378912063484848261013263444281515162528
21334950808585963847462286
192117141012111013201793
2589525653
144155983347452211920201415121113912161014
235050549667
18510331424224151017171212121191012141116
2350646464
15493633034321915161315131314131112121512
22484645
10619711954272221302925232117171616141115
2022232845
1017631283448724869643220232021
20232628264556
876075
13410068
19222628222842625912810212893107
2022
1619
2427439472
4637
131514
2926355362
465643355857754838
344138624559687863484339
253165587310079719271605539
30S
20S
10S
EQ
10N
20N
30N
23 25 21 2613
6
281184
4
6
4
3
-6
11734642
-3
-4-5
979853
-3
-2-4
8101110106
-2
-4-4-6
13172021191494
-11-10
13202627232017168
283023242317
30
176
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2080s
Median of % changes
-50 -30 -20 -10 0 10 20 30 50
344661146121226119112127978052
3564599112911962109107937649
347856847492578593199684949
26328363695072190159825957402033
2653100684511014977596059331316324254341918203134
264160629910510611947585727107
2426211813151412162521116
31110657066
178191121405955271411252517181514161114201317
28626266
11983
228127385252291913222114151513111214171320
2862798079
191115783742402319201718161618161415151915
28595755
13124314767332726373631292522211919171418
2527283556
1249439354259895985794024292525
25283235325669
1077593
16612385
242833342734527773158126158115132
2427
2024
29335311789
5746
161818
3633436577
576954437271935947
425146775673849678605348
3139807291124988811488746848
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
Range of % changes
10 20 30 40 50 100
Fig. 8. Left panels: Change in mean DJF rainfall for the 2020s, 2050s and 2080s (with respect to 1961–90) for the B1-low scenario;median of 7 GCM experiments. For areas with no change shown the model median response fails to exceed the 1 sigma level ofnatural rainfall variability as defined by HadCM2. Right panels: Inter-model range in mean annual temperature change. See text
for further explanation
Clim Res 17: 145–168, 2001158
-9-8
3 3 2 -2-3-5
4-3-5
-5-6
-7-4
-5-3
7
-8
5
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2020s
131624568068221615109
10
141834107877521141310109
1415331181048524131111101114
17401379388301712121213171699
541861059145241613131214151414181441415312311791156
4420710010460352010111111121416242730414758775765102299
6593
11210767452488899101422294259696463674938429
1141191049159392310989101010182638718068535448393622
176133100835631181310119910121316494030422745292321
1461399675462219121099111816164122231722293131
31531417313998
3422151291620272314161716911
173222144133886744
242020191934
11014912111075614528252228241917
1312
120119
7053422523
2325
171815
2715212322
2232258
1214141313
29222927311279
13131319
16292428383217101113131217
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
-15-12
-4-33
-12-11
3
-11
5 42
3
-5-6
2-3-4-8
6-5-8
-8-9
-12
-11-12-11-73
-7-8-5311
-15-12-13
-7-4
8
30S
20S
10S
EQ
10N
20N
30N
2050s
20253787125106352524151416
222853168136117332220161515
22245218416313337201716151821
266221414613847261918192026251414
85291164143703825202019222322222822646583193183142244
6932315616294553116171716192225384246647391120891021594514
10214517516710570381312121414152235456692
10710099
10577606514
17818616314393613516141314161516284159
111126107848475615734
275208157130874929201517151415182024776347664271463634
22921815011873353020151414182826256434352634464948
493492271218153
533424191525324235212527251416
27134722520813810669
383132302953
172233190172118967043393443383026
2018
188187
10983653937
3538
272923
4324333534
345040121922222021
453545434818121420202030
25463745595026151721201926
30S
20S
10S
EQ
10N
20N
30N
-18-15-13
-5-44
-15-14
4
-14
6
9
53
41
-7-7
2-4-6-10
7-6-10
-5-10-9
-11-11
-15-14
-13-14-15-14-84
-17
-9-9-6413
-19-15-16
-8-5
10
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2080s
Median of % changes
-50 -30 -20 -10 0 10 20 30 50
253146108155132433230191720
273566208169145412724201918
28296422820216446252220192226
337726418117058332423232532311717
1053602031778747302525232729272735287980103239226176302
8540019320011668392021212024273147515779901131481101261975618
126180216206129874716151517181927435681
11413312412313095748117
220230202177115764420171617191920355173
13715513210410493767042
3402581941601076036251921181819232430967758825288574442
28327018614690433724191818223432317942443242576159
610608335269190
664230231831395244263133311820
33642927925817113186
473939373665
2132892352131451198753484353473732
2522
233231
135102814945
4447
343629
5329414442
426249152427282425
564356536022141725252538
31564655736132192126242433
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
Range of % changes
10 20 30 40 50 75 100
Fig. 9. Left panels: Change in mean JJA rainfall for the 2020s, 2050s and 2080s (with respect to 1961–90) for the B1-low scenario;median of 7 GCM experiments. For areas with no change shown the model median response fails to exceed the 1 sigma level ofnatural rainfall variability as defined by HadCM2. Right panels: Inter-model range in mean annual temperature change. See text
for further explanation
Hulme et al.: African Climate Change: 1900–2100 159
276
29 25 301677
331395
5
21
8
34
3
-6
12845743
-4
-3-5-6
1111911963
-4
-3-5
-7
101213111272-3
-5-5-7
162023252216114
-13-12
15233031272320199
323527282620
-7
35
2924
4207
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2020s
4054721711412641391311491139361
407569107150139721271251088957
399165988610766100108232805758
30379774815884222185956967472438
3062116805312817390696969381519374963402221243640
3148707311612212313955676732128
2830242115181614192924135
37129758177
208223141476865321613292920211816191317231520
33727277
13997
267148456161342215252417171815131417201623
3373939392
222134914349462722241921191921181617172218
32696665
15328417178383131444236343025242323201621
2931334165
14511045414969
1046999934628342930
30333741386580
12587
10819414499
283238403240618985184147185134155
2831
2328
343962136104
6754
192121
4238507690
6781635084831086954
4959549065869811391696156
37459484106145114103133102868056
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
5211
-11
485510
464721
4256291312919
6224189510
8
53401214558236
-3
-5-12-12
-16
423323157101385
-7-3-3
-6-10-11-11-17
423421201620171062-7-4
-5-9
-4-13
221822252222133-5-3
-9-9-12
-23
293743474130208
-25-22-19
18284457585043373617
-6
-14-17-19
616652535038
-13
44675759544427
-6
2825
1574
-10
27
213714
30S
20S
10S
EQ
10N
20N
30N
2050s
76102135322265497261247280212176114
76141131201283263136240235204168107
74171123184162203125188204437150108109
5770182139152110158418349180130126884572
561162191501002413271701301311307228367092119764140446876
58911321372182302332621041271266023155356463928333027365446254
6924314215314539242126689
129122603025545538393430352532442937
6213613714626218350327984
115115644128474631333329242731382943
611371751751744192531728193875142443640363539343033324234
61129125122288535323148725958828067635647464342383040
55596177
123
274208857793
1311971311871758753645556
5662707771
123152
236164204366271186
536172766075115168160347277348252291
5459
4453
6473118257197
126102
364039
807295143169
12615211894159156205130102
93112102170122161184212171131116105
6985177159200273216193251193162151105
30S
20S
10S
EQ
10N
20N
30N
-9
197918
-16
738515
-10
707232-6
65864520191430
94372714715
12
-20
826119227812459
-4
-8-19-18-15
-24
64513624111520128
-11-4-5
-9-15-16-17-26
-18
655332312531261693
-11-6
-8-13
-6-20
-24
48
342834383334205-8-443
-14-14-19
-35
675553384557677263463112-3
-38-34-29
64
27446887897767575525
-10
-22-26-29
4231
9310179817658
-19
681038790836842
-10
2228465151
3747
-9-5
23
7
27433827
23116
-15
4442
335721
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2080s
Median of % changes
-50 -30 -20 -10 0 10 20 30 50
116157207493407763401379430326269176
117216200308434403209369361312258164
114263189282248311191288313671230166167
8810727921423416924264253527620019313569111
861783372301533705022612002011991104455107142183116646168104116
891392032103353533574011601941949335248186716043514641558370390
106372218235223602645409136198187924638838459615246543949674457
95209210224401280771428129177177986343737148505144374148594567
942112682692676433882641241431347864685661555460534650496452
931981911874428214952271119189
126123103978673706565584762
849194
118189
42031913111914220130220128726813382988586
8596
107118109188233
362252313561416286
819411011692115176258246533425534387447
8291
6781
98113180394302
194156
566260
123110146220260
193233181145244240314199157
143172157261188248282326263201178161
106130272244306418331297385296249231162
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
Range of % changes
10 20 30 40 50 75 100
Fig. 10. Left panels: Change in mean DJF rainfall for the 2020s, 2050s and 2080s (with respect to 1961–90) for the A2-high sce-nario; median of 7 GCM experiments. For areas with no change shown the model median response fails to exceed the 1 sigmalevel of natural rainfall variability as defined by HadCM2. Right panels: Inter-model range in mean annual temperature change.
See text for further explanation
Clim Res 17: 145–168, 2001160
-21-18-16
-6-55
-18-16
-35
-16
33 7
10
6
12
3
12
42
3
3
-8-8
-43-5-6-12
8-7-12
-6-12-11
-13-12
-11
-18-17
-15-16-16-17-16-105
-20-14
-11-11-11-8416
-22-17-18
-9-6
12
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2020s
293653126181154503734222023
314177243197169483129232221
32347526623619253292524222631
389030821119967382727272938362020
12342023720610154362929273134323241329293120278264206352
99466225234136804523252424283136556067931051311731291472296521
1482092522411511015519171820212232506595
133155145143151111869420
257268236206134885124201820222223416085
160181154121121108888249
3973012271871257042292225212122272835
11290689661
102665148
331315217170105504328222121254037369349513849667169
711710391314221
774934272136466051313638362124
392500325301199152100
554546434276
24933727424817013810162565062554337
2926
272269
158119945753
5155
394234
6234475149
497257172831322830
655065617026172029292944
36665364857137222430292838
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
-40-33-29-25
-12-99
-33-31-26-24-29-20
9
-6-69
-31-23-26-25-28-20
179
557
-23-25-25-28
1712
145
-28-33-26
2220
114
-25
-26
22
6244
27
-17
23838
5712
-8
3
547
-7-45
-2-15-15
-85-9-12-23-10
-34
-416-14-22-16-15
-26
-17-11-22-20-20
-37
-28-32-21
-24-23-21
-15-19
-21
-22
-15
-23-20-20
-32-33-31
-28-29-30-32-30-189
-37-27-26-26-20-13-20-20-21-14829
-41-32-35-28-26-18-18
-8-18-11423
30S
20S
10S
EQ
10N
20N
30N
2050s
5668101237340290946965423743
5976145457371318905954434140
6164142501445362101554845424958
72170582398374127725250515471683837
23179244638919010367545552596460607761174176227525497388665
186880425441256150854347454553596810411312617519924832624327843312239
2783954764542851901043532343739426094
12217825129227327028520916317737
484505445389253167964437343742414377
11316130234229122822820416715492
74856742735323613279554247403942505466
2101701281811151931259791
62359340932119894815341393948757068
17592967192
125133130
13411339738592417
145926550406987
11496586872673945
738943613567376287188
10386868280
144
469635517468320261191118106941181038171
5549
513508
298225178107100
96104
747864
11664899693
93136108325259615356
1239512211613249313855545483
6912410112116013570424656545272
30S
20S
10S
EQ
10N
20N
30N
-61-51-45-38-26-19-14
-9-18-1314
-51-47-40-37-44-31
14
-10-913
-47-35-39-38-43-30
2613
7810
-35-38-39-43-21
261953217
-10-21
-44-50-40
343054176
-10-19
24
-26-39-30-40
223489365610
2341
-26
-30
-42
-35
3612413-2471118
-12
14
538610
-59
-10-784-3-22-24
-18
-138
-14-19-35-15
-32
-25
-52-37
-52
-724-21-34-25-23-26-20
-39-27
-40-40
-37-50-44-41
-26-17-34-31-31
-21
-45-56
-43-49-32
-37-36-32
-23-29
-33-32
-35-24-25-34
-23
-35-31-30
-25-48-51-48
-43-45-46-49-46-2814
-57-42-39-40-31-20-31-31-32-221245
-63-49-53-42-39-28-27-15-12-27-17535
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
2080s
Median of % changes
-50 -30 -20 -10 0 10 20 30 50
85105154364522444145106100645766
911172227025694881399083666362
9398218769682555154857369657489
111261892610574194110797779831091045857
355121468459629215710383857991989392119932672703488057635951020
286134965167739223013066727069819010416017319326830538050037242666418760
42760673069743729215954495256606491
14518827438544841841543832125027257
74277568259638925614768575357656366
118173247463525446349350313256236141
11478706555413632031218564716160657782
102323261197277176296192149140
9569106274933041441248163595973
115108105268141147110141192205200
205720531132908640
2231421007761
10613317514889
1041111036069
11331447941870576440289
158131132126123221
719974794719491400293181162144180158124108
8476
787779
457345273164153
148160
11312098
17899137148142
142208165498091938286
189145188178202764858858383127
1051901541862462071086570868281110
10W 0 10E 20E 30E 40E 50E
30S
20S
10S
EQ
10N
20N
30N
Range of % changes
10 20 30 40 50 75 100
Fig. 11. Left panels: Change in mean JJA rainfall for the 2020s, 2050s and 2080s (with respect to 1961–90) for the A2-high sce-nario; median of 7 GCM experiments. For areas with no change shown the model median response fails to exceed the 1 sigmalevel of natural rainfall variability as defined by HadCM2. Right panels: Inter-model range in mean annual temperature change.
See text for further explanation
Hulme et al.: African Climate Change: 1900–2100
of equatorial East Africa where rainfall increases by 5to 30% in DJF and decreases by 5 to 10% in JJA. Someareas of Sahelian West Africa and the Mahgreb alsoexperience ‘significant’ rainfall decreases in JJA sea-son under the B1-low scenario. The inter-model rangefor these rainfall changes is large and in the cases citedabove always exceeds the magnitude of the Medianmodel response. Over the seasonally arid regions ofAfrica, the inter-model range becomes very large(>100%) because of relatively large percent changesin modelled rainfall induced by very small baselineseasonal rainfall quantities.
With more rapid global warming (e.g. the B2-mid,A1-mid and A2-high scenarios), increasing areas ofAfrica experience changes in DJF or JJA rainfall thatdo exceed the 1 sigma level of natural rainfall vari-ability. Thus for the A2-high scenario, large areas ofequatorial Africa experience ‘significant’ increases inDJF rainfall of up to 50 or 100% over parts of EastAfrica (Fig. 10), while rainfall decreases ‘significantly’in JJA over parts of the Horn of Africa and northwestAfrica (Fig. 11). Some ‘significant’ JJA rainfall increasesoccur over the central Sahel region of Niger and Chad,while ‘significant’ decreases in DJF rainfall (15 to 25%)occur over much of South Africa and Namibia andalong the Mediterranean coast. The inter-model rangefor these rainfall changes remains large, however, andwith very few exceptions exceeds the magnitude of theMedian model response. Even for the seasonally wetJJA rainfall regime of the Sahel, inter-model rangescan exceed 100%, suggesting that different GCM sim-ulations yield (sometimes) very different regional rain-fall responses to a given greenhouse gas forcing. Thislarge inter-model range in seasonal mean rainfallresponse is not unique to Africa and is also found overmuch of south and southwest Asia and parts of CentralAmerica (Carter et al. 2001).
5.2. National scenario graphs
To condense this scenario information further, wealso constructed ‘national’-scale summary graphs for 4smaller regions—centred on the countries of Senegal,Tunisia, Ethiopia and Zimbabwe. These chosen do-mains are shown in Fig. 6 (top left panel) and reflectthe diversity of existing climate regimes across thecontinent from north to south and from west to east.Each country graph shows, for the 2050s, the distribu-tion of the mean annual changes in mean temperatureand precipitation for each GCM simulation and foreach of our 4 scenarios (Fig. 12). As with the continen-tal maps, these changes are compared with the naturalmulti-decadal variability of annual-mean temperatureand precipitation extracted from the HadCM2 1400 yr
unforced simulation. These graphs provide a quickassessment at a ‘national’ scale of the likely range andsignificance of future climate change and again showsthe extent to which different GCMs agree in theirregional response to a given magnitude of globalwarming.
For each country there is a spread of results relatingto inter-model differences in climate response. Forexample, in Tunisia the change in annual rainfall ispredominantly towards drying (only ECHAM4 dis-plays wetting), although the magnitude of the dryingunder the A2-high scenario is between 1 and 30%.Natural climate variability is estimated to lead to dif-ferences of up to ±10% between different 30 yr meanclimates; therefore the more extreme of these scenariooutcomes would appear to be ‘significant’ for Tunisia.The picture would appear at first sight to be less clearfor Zimbabwe, where 4 of the GCMs suggest wettingand 3—including the HadCM2 ensemble of 4 simula-tions—suggest drying. However, the range of naturalvariability in annual rainfall when averaged over 30 yris shown to be about ±6% and most of the wetting sce-narios fall within this limit. It is the drying responsesunder the more extreme A2-high, B2-mid and A1-midscenarios that would appear to yield a more ‘signifi-cant’ result.
It is also important to point out that inter-ensembledifferences in response at these national scales canalso be large. The 4-member HadCM2 ensemble forTunisia yields differences in rainfall change of 15% ormore, while for Ethiopia inter-ensemble differencescan lead to a sign change in the rainfall scenario. Inthis latter case, however, few of these HadCM2 rainfallchanges are larger than the estimate of natural rainfallvariability for Ethiopia. It is also worth noting that therelative regional response of different GCMs is notalways the same. Thus, for Ethiopia, the CCSR-NIESGCM generates the most extreme wetting scenario,whereas for Tunisia the same model yields the mostextreme drying scenario. We discuss the significanceof some of these differences and similarities betweendifferent GCMs in our discussion of uncertainties inSection 6.
5.3. Changes in ENSO-related rainfall variability
Given the important role that ENSO events exert oninterannual African rainfall variability, at least in someregions, determining future changes in interannualrainfall variability in Africa can only be properly con-sidered in the context of changes in ENSO behaviour.There is still ambiguity, however, about how ENSOevents may respond to global warming. This is partlybecause GCMs only imperfectly simulate present
161
Clim Res 17: 145–168, 2001
ENSO behaviour. Tett et al. (1997) demonstrate thatHadCM2 simulates ENSO-type features in the PacificOcean, but the model generates too large a warmingacross the Tropics in response to El Niño events. Tim-mermann et al. (1999), however, have recently arguedthat their ECHAM4 model (see Table 1) has sufficientresolution to simulate ‘realistic’ ENSO behaviour. Theyanalysed their greenhouse gas-forced simulations andsuggested that in the future there will be more frequentand more intense ‘warm’ and ‘cold’ ENSO events, aresult also found in the HadCM2 model (Collins 2000).
What effects would such changes have on interan-nual African rainfall variability? This not only dependson how ENSO behaviour changes in the future, butalso upon how realistically the models simulate theobserved ENSO-rainfall relationships in Africa. Smith& Ropelewski (1997) looked at Southern Oscillation-rainfall relationships in the NCEP atmospheric GCM,where the model is used to re-create observed climatevariability after being forced with observed sea surfacetemperatures (SSTs). Even in this most favourable ofmodel experiments, the model relationships do not
162
-1 0 1 2 3 4 5Temperature change ( oC)
-40
-20
0
20
Pre
cip
itatio
n ch
ange
(%)
Senegal
-1 0 1 2 3 4 5Temperature change ( oC)
-40
-20
0
20
Pre
cip
itatio
n ch
ange
(%)
Tunisia
-1 0 1 2 3 4 5Temperature change ( oC)
-40
-20
0
20
Pre
cip
itatio
n ch
ange
(%)
Ethiopia
-1 0 1 2 3 4 5Temperature change ( oC)
-40
-20
0
20
Pre
cip
itatio
n ch
ange
(%)
Zimbabwe
CCSR CGCMI CSIRO ECHAM GFDL HadCM2 NCAR
Fig. 12. Change in mean annual temperature and precipitation for the 2050s (with respect to 1961–90) for regions centred onSenegal, Tunisia, Ethiopia and Zimbabwe (see Fig. 6, top left panel, for selected domains). Results from the 7 DDC GCMs areshown, scaled to reflect the 4 climate change scenarios adopted in this study: A2-high, A1-mid, B2-mid and B1-low. Note: theHadCM2 GCM has 4 results reflecting the 4-member ensemble simulations completed with this GCM. The bold lines centred onthe origin indicate the 2 standard deviation limits of natural 30 yr time-scale natural climate variability defined by the 1400 yr
HadCM2 control simulation
Hulme et al.: African Climate Change: 1900–2100
always reproduce those observed. Over southeasternAfrica, the simulated rainfall percentiles are consistentwith the observations reported by Ropelewski &Halpert (1996), but over eastern equatorial Africa themodel simulates an relationship opposite to thatobserved. The recently elucidated role of the IndianOcean dipole (Saji et al. 1999, Webster et al. 1999) inmodulating eastern African rainfall variability may beone reason simple ENSO-precipitation relationshipsare not well replicated by the GCMs in this region.
We analysed 240 yr of unforced simulated climatemade using the HadCM2 GCM (see Table 1) to see towhat extent this model can reproduce observed rela-tionships. We performed the identical analysis to thatperformed on the observed data in Section 4 and theresults are plotted in Fig. 5c,d. The 2 strongest ENSOsignals in African rainfall variability are only imper-fectly reproduced by the model. The East African neg-ative correlation in November–April is rather too weakin the model and also too extensive, extending west-wards across the whole African equatorial domain.The positive correlation over southern Africa is tooweak in HadCM2 and displaced northwards by some10° latitude. The absence of any strong and coherentrelationship during the June–October season is repro-duced by the model (Fig. 5b).
On the basis of this analysis, and our assessment ofthe literature, we are not convinced that quantifyingfuture changes to interannual rainfall variability inAfrica due to greenhouse gas forcing are warranted.At the very least, this issue deserves a more thoroughinvestigation of ENSO-rainfall relationships in theGCMs used here and how these relationships changein the future. Such an analysis might also be useful indetermining the extent to which seasonal rainfall fore-casts in Africa that rely upon ENSO signatures mayremain valid under scenarios of future greenhouse gasforcing.
6. UNCERTAINTIES AND LIMITATIONSTO KNOWLEDGE
In the introduction to this paper we alluded to someof the limitations of climate change scenarios for Africaand those shown in this paper are no exception. Theselimitations arise because of, inter alia, (1) the problemof small signal:noise ratios in some scenarios for pre-cipitation and other variables, (2) the inability of cli-mate model predictions to account for the influence ofland cover changes on future climate, and (3) the rela-tively poor representation in many models of someaspects of climate variability that are important forAfrica (e.g. ENSO). Some of these limitations havebeen revealed by analyses presented earlier.
Even though we have presented a set of 4 climatefutures for Africa, where the range reflects unknownfuture global greenhouse gas emissions and 3 differentvalues for the global climate sensitivity, we cannotplace probability estimates on these 4 outcomes withmuch confidence. While this conclusion may wellapply for most, or all, world regions, it is particularlytrue for Africa, where the roles of land cover changeand dust and biomass aerosols in inducing regional cli-mate change are excluded from the climate changemodel experiments reported here.
This concern is most evident in the Sahel region ofAfrica. None of the model-simulated present or futureclimates for this region displays behaviour in rainfallregimes that is similar to that observed over recentdecades. This is shown in Fig. 13 where we plot theobserved regional rainfall series for 1900–98, as usedin Fig. 3, and then append the 10 model-simulatedevolutions of future rainfall for the period 2000–2100.These future curves are extracted directly from the 10GCM experiments reported in Table 1 and have notbeen scaled to our 4 scenario values (this scaling wasperformed in the construction of Figs 8 to 11 as dis-cussed in Section 5). One can see that none of themodel rainfall curves for the Sahel displays multi-decadal desiccation similar to what has been observedin recent decades. This conclusion also applies to themulti-century unforced integrations performed withthe same GCMs (Brooks 1999).
There are a number of possible reasons for this. Itcould be that the climate models are poorly replicating‘natural’ rainfall variability for this region. In particularthe possible role of ocean circulation changes in caus-ing this desiccation (Street-Perrott & Perrott 1990) maynot be well simulated in the models. It could also bethat the cause of the observed desiccation is some pro-cess that the models are not including. Two candidatesfor such processes would be the absence of a dynamicland cover/atmosphere feedback process and the ab-sence of any representation of changing atmosphericdust aerosol concentration. The former of these feed-back processes has been suggested as being very im-portant in determining African climate change duringthe Holocene by amplifying orbitally induced Africanmonsoon enhancement (Kutzbach et al. 1996, Claussenet al. 1999, Doherty et al. 2000). This feedback mayalso have contributed to the more recently observeddesiccation of the Sahel (Xue 1997). The latter processof elevated Saharan dust concentrations may also beimplicated in the recent Sahelian desiccation (Brooks1999).
Without such a realistic simulation of observed rain-fall variability, it is difficult to define with confidencethe true magnitude of natural rainfall variability inthese model simulations and also difficult to argue
163
Clim Res 17: 145–168, 2001
that these greenhouse gas-induced attributed rainfallchanges for regions in Africa will actually be those thatdominate the rainfall regimes of the 21st century.Notwithstanding these model limitations due to omit-ted or poorly represented processes, Fig. 13 alsoillustrates the problem of small signal:noise ratios inprecipitation scenarios. The 10 individual model simu-lations yield different signs of precipitation change forthese 3 regions as well as different magnitudes. Howmuch of these differences are due to model-generated
natural variability is difficult to say. Inour scenario maps (Figs 8 to 11) we pre-sented the median precipitation changefrom these 10 (scaled) model simula-tions, implying that we can treat theseclimate change simulations as individualmembers of an ensemble. The ensem-ble-mean or median therefore yields our‘best’ estimate of the true response togreenhouse gas forcing; much as innumerical weather prediction the en-semble-mean forecast is often taken asthe ‘best’ short-range weather forecast.In our example, for the Sahel and south-ern African the median response wasannual drying, whereas for East Africathe median response was wetting(Fig. 13).
One other concern about the applica-bility in Africa of climate change sce-narios such as those presented here isthe relationship between future climatechange predictions and seasonal rainfallforecasts. There is increasing recogni-tion (e.g. Downing et al. 1997, Ringius1999, Washington & Downing 1999) thatfor many areas in the tropics one ofthe most pragmatic responses to theprospect of long-term climate change isa wish to strengthen the scientific basisof seasonal rainfall forecasts. Whereforecasts are feasible, this should beaccompanied by improvements in themanagement infrastructure to facilitatetimely responses. Such a research andadaptation strategy focuses on the short-term realisable goals of seasonal climateprediction and the near-term and quan-tifiable benefits that improved forecastapplications will yield. At the same time,the strengthening of these institutionalstructures offers the possibility that themore slowly emerging signal of climatechange in these regions can be bettermanaged in the decades to come. It is
therefore an appropriate form of climate change adap-tation. This means that 2 of the objectives of climatechange prediction should be: (a) to determine theeffect global warming may have on seasonal predict-ability (will forecast skill levels increase or decrease orwill different predictors be needed); and (b) to deter-mine the extent to which predicted future climatechange will impose additional strains on natural andmanaged systems over and above those that arecaused by existing seasonal climate variability. For
164
The Sahel
-80
-60
-40
-20
0
20
40
60
80
East Africa
-80
-60
-40
-20
0
20
40
60
80
An
nual
ra
ina
ll a
nom
aly
(per
cen
t)
Southeast Africa
-80
-60
-40
-20
0
20
40
60
80
1880 1900 1920 1940 1960 1980 2000 2020 2040 2060 2080 2100
1880 1900 1920 1940 1960 1980 2000 2020 2040 2060 2080 2100
Fig. 13. Observed annual rainfall anomalies for 3 African regions, 1900–98(see Fig. 3), and model-simulated anomalies for 2000–2099. Model anomaliesare for the 10 model simulations derived from the 7 DDC GCM experiments—the 4 HadCM2 simulations are the dashed curves (see Table 1). All anomaliesare expressed with respect to either observed or model-simulated 1961–90average rainfall. The model curves are extracted directly from the GCM ex-periments and the results are not scaled to the 4 scenarios used in this paper.
The smooth curves result from applying a 20 yr Gaussian filter
Hulme et al.: African Climate Change: 1900–2100
both of these reasons we need to improve our predic-tions of future climate change and in particular toimprove our quantification of the uncertainties.
7. CONCLUSIONS
The climate of Africa is warmer than it was 100 yrago. Although there is no evidence for widespreaddesiccation of the continent during this century, insome regions substantial interannual and multi-de-cadal rainfall variations have been observed and nearcontinent-wide droughts in 1983 and 1984 had somedramatic impacts on both the environment and someeconomies (Benson & Clay 1998). The extent to whichthese rainfall variations are related to greenhouse gas-induced global warming, however, remains undeter-mined. A warming climate will nevertheless placeadditional stresses on water resources, whether or notfuture rainfall is significantly altered.
Model-based predictions of future greenhouse gas-induced climate change for the continent clearly sug-gest that this warming will continue and, in most sce-narios, accelerate so that the continent on averagecould be between 2 and 6°C warmer in 100 yr time.While these predictions of future warming may be rel-atively robust, there remain fundamental reasons whywe are much less confident about the magnitude, andeven direction, of regional rainfall changes in Africa.Two of these reasons relate to the rather ambiguousrepresentation in most GCMs of ENSO-type climatevariability in the tropics (a key determinant of Africanrainfall variability) and the omission in all currentGCMs of any representation of dynamic land cover-atmosphere interactions and dust and biomass aero-sols. Such interactions have been suggested to beimportant in determining African climate variabilityduring the Holocene and may well have contributed tothe more recently observed desiccation of the Sahel.
We suggest that climate change scenarios, such asthose presented here, should nevertheless be used toexplore the sensitivity of a range of African environ-mental and social systems, and economically valuableassets, to a range of future climate changes. Someexamples of such exploration were presented in Dixonet al. (1996), although in these studies there was littleco-ordinated and quantified use of a coherent set ofclimate futures. Further work can be done to elaborateon some of the higher order climate statistics associ-ated with the changes in mean seasonal climate shownhere—particularly daily temperature and precipita-tion extremes. It may also be worthwhile to explorethe sensitivity of these model predictions to the spatialresolution of the models—i.e. explore the extent towhich downscaled scenarios differ from GCM-scale
scenarios—although such downscaling techniques donot remove the fundamental reasons why we are un-certain about future African rainfall changes.
The exploration of African sensitivity to climatechange must also be undertaken in conjunction withthe more concrete examples we have of sensitivity toshort-term (seasonal time scale) climate variability.These estimates may be based on observed recon-struction of climate variability over the last century oron the newly emerging regional seasonal rainfall fore-casts now routinely being generated for southern, east-ern and western Africa (e.g. NOAA 1999, SARCOF,see also http://iri.ldeo.columbia.edu [InternationalResearch Institute for Climate Prediction]). Because ofthe uncertainties mentioned above about futureregional climate predictions for Africa, initial steps toreduce vulnerability should focus on improved adapta-tion to existing climate variability (Downing et al. 1997,Adger & Kelly 1999, Ringius 1999). Thus, emphasiswould be placed on reducing vulnerability to adverseclimate events and increasing capacity to adapt toshort-term and seasonal weather conditions and cli-matic variability. The likelihood of significant eco-nomic and social benefits from adaptation to short-term climate variability in Africa justifies this activity.Additionally, and importantly, lessons from adaptationto short-term climate variability would build capacityto respond incrementally to longer-term changes inlocal and regional climates.
Acknowledgements. The Climate Impacts LINK Project(funded by DETR Contract No. EPG 1/1/68) contributed com-puting facilities and staff time for RMD. T.N. was supportedby an IGBP/START Fellowship. The MAGICC climate modelwas used with the permission of Tom Wigley and SarahRaper.
LITERATURE CITED
Adger WN, Kelly PM (1999) Social vulnerability to climatechange and the architecture of entitlements. Mitig AdaptStrat Global Change 4:253–266
Aubreville A (1949) Climats, foréts et désertification del’Afrique tropicale. Société d’Editions Géographiques,Maritimes et Coloniales, Paris
Benson C, Clay EJ (1998) The impact of drought on sub-Saharan economies. World Bank Tech Paper No. 401,World Bank, Washington, DC
Birkett C, Murtugudde R, Allan T (1999) Indian Ocean cli-mate event brings floods to East Africa’s lakes and theSudd Marsh. Geophys Res Lett 26:1031–1034
Boer GJ, Flato GM, Reader MC, Ramsden D (2000) Transientclimate change simulation with greenhouse gases andaerosol forcing: projected climate to the twenty-first cen-tury. Clim Dyn 165:427–450
Brooks N (1999) Dust-climate interactions in the Sahel-Saharazone with particular reference to late twentieth centurySahel drought. PhD thesis, University of East Anglia
Carter TR, Hulme M, Tuomenvirta H, New MG, Osborn TJ,
165
Clim Res 17: 145–168, 2001
Crossley J, Doherty RM, Jones PD (2001) Interim charac-terizations of regional climate and related changes up to2100 associated with the preliminary SRES emissions sce-narios. FEI Report, Helsinki
Charney JG (1975) Dynamics of deserts and drought in theSahel. Q J R Meteorol Soc 101:193–202
Claussen M, Kubatzki C, Brovkin V, Ganopolski A (1999)Simulation of an abrupt change in Saharan vegetation inthe mid-Holocene. Geophys Res Lett 26:2037–2040
Collins M (2000) The El Niño-Southern Oscillation in thesecond Hadley Centre coupled model and its response togreenhouse warming. J Clim 13:1299–1312
Conway D, Krol M, Alcamo J, Hulme M (1996) Future wateravailability in Egypt: the interaction of global, regionaland basin-scale driving forces in the Nile Basin. Ambio25:336–342
Cunnington WM, Rowntree PR (1986) Simulations of the Sa-haran atmosphere: dependence on moisture and albedo.Q J R Meteorol Soc 112:971–999
Dai A, Wigley TML (2000) Global patterns of ENSO-inducedprecipitation. Geophys Res Lett 27:1283–1286
Dixon RK, Guill S, Mkanda FX, Hlohowskyj I (eds) (1996) Vul-nerability and adaptation of African ecosystems to globalclimate change. CR SPECIAL 3. Clim Res 6(2)
Doherty RM, Kutzbach J, Foley J, Pollard D (2000) Fully-coupled climate/dynamical vegetation model simulationsover Northern Africa during the mid-Holocene. Clim Dyn16:561–573
Downing TE, Ringius L, Hulme M, Waughray D (1997) Adapt-ing to climate change in Africa. Mitig Adapt Strat GlobalChange 2:19–44
El Hamly M, Sebbari R, Lamb PJ, Ward MN, Portis DH (1998)Towards the seasonal prediction of Moroccan precipita-tion and its implications for water resources management.In: Servat E, Hughes D, Fritsch JM, Hulme M (eds) Waterresources variability in Africa during the 20th century.IAHS Publication No. 252, Wallingford, p 79–88
Emori S, Nozawa T, Abe-Ouchi A, Numaguti A, Kimoto M,Nakajima T (1999) Coupled ocean-atmosphere modelexperiments of future climate change with an explicit rep-resentation of sulfate aerosol scattering. J Meteorol SocJpn 77:1299–1307
Feddema JJ (1999) Future African water resources: interac-tions between soil degradation and global warming. ClimChange 42:561–596
Folland CK, Owen JA, Ward MN, Colman AW (1991) Predic-tion of seasonal rainfall in the Sahel region using empiricaland dynamical methods. J Forecast 10:21–56
Giorgi F, Francisco R (2000) On the predictability of regionalclimate change. Clim Dyn 16:169–182
Haywood JM, Stouffer RJ, Wetherald R, Manabe S,Ramaswamy V (1997) Transient response of a coupledmodel to estimated changes in greenhouse gas and sul-phate concentrations. Geophys Res Lett 24:1335–1338
Hernes H, Dalfelt A, Berntsen T, Holtsmark B, Otto Naess L,Selrod R, Aaheim A (1995) Climate strategy for Africa.CICERO Report 1995:3, University of Oslo
Hewitson BC, Crane RG (1998) Regional scale daily precipita-tion from downscaling of data from the GENESIS andUKMO GCMs. In: Proceedings of 14th conference onprobability and statistics in the atmospheric sciences,Phoenix, Arizona, 11–16 January 1998. American Meteo-rological Society, Boston, p J48–J50
Hewitson BC, Joubert A (1998) Climate downscaling: currentSouth African projections. http://www.egs.uct.ac.za/fccc/
Hirst AC, O’Farrell SP, Gordon HB (2000) Comparison of acoupled ocean-atmosphere model with and without oceanic
eddy-induced advection. 1. Ocean spin-up and controlintegrations. J Clim 13:139–163
Hudson DA (1997) Southern African climate change simu-lated by the GENESIS GCM. S Afr J Sci 93:389–403
Hudson DA, Hewitson BC (1997) Mid-latitude cyclones southof South Africa in the GENESIS GCM. Int J Climatol 17:459–473
Hulme M (1992) Rainfall changes in Africa: 1931–60 to1961–90. Int J Climatol 12:685–699
Hulme M (1994a) Regional climate change scenarios basedon IPCC emissions projections with some illustrations forAfrica. Area 26:33–44
Hulme M (1994b) Validation of large-scale precipitation fieldsin General Circulation Models. In: Desbois M, DésalmandF (eds) Global precipitations and climate change. Springer-Verlag, Berlin, p 387–405
Hulme M (ed) (1996a) Climate change and southern Africa:an exploration of some potential impacts and implicationsin the SADC region. CRU/WWF, Norwich
Hulme M (1996b) Climatic change within the period of mete-orological records. In: Adams WM, Goudie AS, Orme AR(eds) The physical geography of Africa. Oxford UniversityPress, Oxford, p 88–102
Hulme M (1998) The sensitivity of Sahel rainfall to globalwarming: implications for scenario analysis of future cli-mate change impact. In: Servat E, Hughes D, Fritsch JM,Hulme M (eds) Water resources variability in Africa dur-ing the 20th century. IAHS Publication No. 252, Walling-ford, p 429–436
Hulme M (2001) The years of drought: the Sahel 1973–1998.Global Environ Change 11(1):19–29
Hulme M, Carter TR (1999) Representing uncertainty in cli-mate change scenarios and impact studies. In: Carter T,Hulme M, Viner D (eds) Representing uncertainty in cli-mate change scenarios and impact studies. Proceedings ofthe ECLAT-2 Helsinki Workshop, 14–16 April 1999. Cli-matic Research Unit, Norwich, p 11–37
Hulme M, Carter TR (2000) The changing climate of Europe,Chapter 3. In: Parry ML (ed) An assessment of the poten-tial effects of climate change in Europe. The Europe ACA-CIA project. UEA, Norwich, p 47–84
IPCC (1998) The regional impacts of climate change: anassessment of vulnerability. Cambridge University Press,Cambridge
Janicot S, Moron V, Fontaine B (1996) Sahel droughts andENSO dynamics. Geophys Res Lett 23:515–518
Janowiak JE (1988) An investigation of interannual rainfallvariability in Africa. J Clim 1:240–255
Jones PD (1994) Hemispheric surface air temperature vari-ability—a reanalysis and an update to 1993. J Clim 7:1794–1802
Jones PD, Lindesay J (1993) Maximum and minimum temper-ature trends over South Africa and the Sudan. In: Pro-ceedings of the 4th international conference on southernhemisphere meteorology and oceanography, 29 March to2 April 1993, Hobart, Australia. American MeteorologicalSociety, Boston, p 359–360
Jones PD, Briffa KR, Barnett TP, Tett SFB (1998) High-resolu-tion palaeoclimatic records for the last millennium: inter-pretation, integration and comparison with GCM control-run temperatures. The Holocene 8:455–471
Joubert AM, Hewitson BC (1997) Simulating present andfuture climates of southern Africa using general circula-tion models. Prog Phys Geogr 21:51–78
Joubert AM, Mason SJ, Galpin JS (1996) Droughts oversouthern Africa in a doubled-CO2 climate. Int J Climatol16:1149–1156
166
Hulme et al.: African Climate Change: 1900–2100
Kittel TGF, Giorgi F, Meehl GA (1998) Intercomparison ofregional biases and doubled CO2-sensitivity of coupledatmosphere-ocean general circulation model experiments.Clim Dyn 14:1–15
Kutzbach J, Bonan G, Foley J, Harrison SP (1996) Vegetationand soil feedbacks on the response of the Africa monsoonto orbital forcing in the early to middle Holocene. Nature384:623–626
Leggett J, Pepper WJ, Swart RJ (1992) Emissions scenarios forthe IPCC: an update. In: Houghton JT, Callander BA,Varney SK (eds) Climate change 1992: the supplementaryreport to the IPCC scientific assessment. Cambridge UnivPress, Cambridge, p 75–95
Lindsay SW, Parson L, Thomas CJ (1998) Mapping the rangesand relative abundance of the two principal Africanmalaria vectors, Anopheles gambiae sensu stricto and An.Arabiensis, using climate data. Proc R Soc Lond B 265:847–854
Mason SJ, Waylen PR, Mimmack GM, Rajaratnam B, Harri-son JM (1999) Changes in extreme rainfall events in SouthAfrica. Clim Change 41:249–257
Mearns LO, Bogardi I, Giorgi F, Matyasovszky I, Palecki M(1999) Comparison of climate change scenarios generatedfrom regional climate model experiments and statisticaldownscaling. J Geophys Res 104:6603–6621
Meehl GA, Washington WM (1995) Cloud albedo feedbackand the super-greenhouse effect in a global complexGCM. Clim Dyn 11:399–411
Mitchell JFB, Johns TC (1997) On the modification of globalwarming by sulphate aerosols. J Clim 10:245–267
Mitchell JFB, Manabe S, Meleshko V, Tokioka T (1990) Equi-librium climate change—and its implications for thefuture. In: Houghton JT, Jenkins GJ, Ephraums JJ (eds)Climate change: the IPCC scientific assessment. Cam-bridge University Press, Cambridge, p 137–164
Mitchell T, Hulme M (1999) Predicting regional climatechange: living with uncertainty. Prog Phys Geogr 23:57–78
New M, Hulme M, Jones PD (1999) Representing twentiethcentury space-time climate variability. Part 1: develop-ment of a 1961–90 mean monthly terrestrial climatology.J Clim 12:829–856
New M, Hulme M, Jones PD (2000) Representing twentiethcentury space-time climate variability. Part 2: develop-ment of 1901–96 monthly grids of terrestrial surface cli-mate J Clim (in press)
Nicholls N, Gruza GV, Jouzel J, Karl TR, Ogallo LA, ParkerDE (1996) Observed climate variability and change. In:Houghton JT, Meiro Filho LG, Callendar BA, KattenburgA, Maskell K (eds) Climate change 1995: the science ofclimate change. Cambridge University Press, Cambridge,p 133–192
Nicholson SE (1994) Recent rainfall fluctuations in Africa andtheir relationship to past conditions over the continent.The Holocene 4:121–131
NOAA (1999) An experiment in the application of climateforecasts: NOAA-OGP activities related to the 1997–98 ElNiño event. OGP/NOAA/US Dept of Commerce, Wash-ington, DC
Penner JE, Chuang CC, Grant K (1998) Climate forcing bycarbonaceous and sulfate aerosols. Clim Dyn 14:839–852
Raper SCB, Warrick RA, Wigley TML (1996) Global sea levelrise: past and future. In: Milliman JD, Haq BU (eds) Sea-level rise and coastal subsidence: causes, consequencesand strategies. Kluwer Academic Publishers, Dordrecht,p 11–45
Ringius L (1999) The climatic variability and climate change
initiative for Africa. Processed report for the World Bank,Washington, DC
Ringius L, Downing TE, Hulme M, Waughray D, Selrod R(1996) Climate change in Africa—issues and regionalstrategy. CICERO Report No. 1996:8, CICERO, Oslo
Roeckner E, Arpe K, Bengtsson L, Christoph M, Claussen M,Dümenil L, Esch M, Giorgetta M, Schlese U, Schulz-weida U (1996) The atmospheric general circulation modelECHAM-4: model description and simulation of present-day climate. Max Planck Institute for Meteorology, ReportNo. 218, Hamburg
Ropelewski CF, Halpert MS (1987) Global and regional scaleprecipitation patterns associated with the El Niño/South-ern Oscillation. Mon Weather Rev 115:1606–1626
Ropelewski CF, Halpert MS (1989) Precipitation patternsassociated with the high phase of the Southern Oscillation.J Clim 2:268–284
Ropelewski CF, Halpert MS (1996) Quantifying SouthernOscillation-precipitation relationships. J Clim 9:1043–1059
Ropelewski CF, Jones PD (1987) An extension of the Tahiti-Darwin Southern Oscillation Index. Mon Weather Rev115:2161–2165
Rowell DP, Folland CK, Maskell K, Ward MN (1995) Variabilityof summer rainfall over tropical north Africa (1906–92): ob-servations and modelling. Q J R Meteorol Soc 121:669–704
Saji NH, Goswami BN, Vinayachandran PN, Yamagata T(1999) A dipole mode in the tropical Indian Ocean. Nature401:360–363
Santer BD, Taylor KE, Wigley TML, Johns TC, Jones PD,Karoly DJ, Mitchell JFB, Oort AH, Penner JE, RamaswamyV, Schwarzkopf MD, Stouffer RJ, Tett S, Boyle JS, ParkerDE (1996) Human effect on global climate. Nature 384:523–524
Schimel D and 26 others (1996) Radiative forcing of climatechange. In: Houghton JT, Meiro Filho LG, Callendar BA,Kattenburg A, Maskell K (eds) Climate change 1995: thescience of climate change. Cambridge University Press,Cambridge, p 65–131
Schlesinger ME, Malyshev S, Rozanov EV, Yang F, Andro-nova NG (2000) Geographical distributions of temperaturechange for scenarios of greenhouse gas and sulfur dioxideemissions. Technol Forecast Social Change 65:167–193
Smith TM, Ropelewski CF (1997) Quantifying Southern Os-cillation-precipitation relationships from an atmosphericGCM. J Clim 10:2277–2284
Stebbing EP (1935) The encroaching Sahara: the threat to theWest African colonies. Geogr J 85:506–524
Stockdale TN, Anderson DLT, Alves JOS, Balmaseda MA(1998) Global seasonal rainfall forecasts using a coupledocean-atmosphere model. Nature 392:370–373
Street-Perrott FA, Perrott RA (1990) Abrupt climate fluctua-tions in the tropics: the influence of Atlantic Ocean circu-lation. Nature 343:607–612
Sud YC, Lau WKM (1996) Comments on paper ‘Variability ofsummer rainfall over tropical North Africa (1906–1992):observations and modelling.’ Q J R Meteorol Soc 122:1001–1006
Sun LQ, Semazzi FHM, Giorgi F, Ogallo L (1999) Applicationof the NCAR Regional Climate Model to eastern Africa—2. Simulation of interannual variability of short rains.J Geophys Res 104:6549–6562
Tett SFB, Johns TC, Mitchell JFB (1997) Global and regionalvariability in a coupled AOGCM. Clim Dyn 13:303–323
Timmermann A, Oberhuber J, Bacher A, Esch M, Latif M,Roeckner E (1999) Increased El Niño frequency in a cli-mate model forced by future greenhouse warming. Nature398:694–697
167
Clim Res 17: 145–168, 2001
Tyson PD (1991) Climatic change in southern Africa: past andpresent conditions and possible future scenarios. ClimChange 18:241–258
Wang GL, Eltahir EAB (2000) Ecosystem dynamics and theSahel drought. Geophys Res Lett 27:795–798
Washington R, Downing TE (1999) Seasonal forecasting of Afri-can rainfall: problems and prospects. Geogr J 165:255–274
Webster PJ, Moore AM, Loschnigg JP, Leben RR (1999) Cou-pled ocean-atmosphere dynamics in the Indian Oceanduring 1997–98. Nature 401:356–360
Wigley TML, Raper SCB (1992) Implications of revised IPCCemissions scenarios Nature 357:293–300
Wigley TML, Raper SCB, Salmon M, Hulme M (2000)MAGICC: Model for the Assessment of Greenhouse-gas
Induced Climate Change: Version 2.4. Climatic ResearchUnit, Norwich
Xue Y (1997) Biosphere feedback on regional climate intropical North Africa. Q J R Meteorol Soc 123:1483–1515
Xue Y, Shukla J (1998) Model simulation of the influence ofglobal SST anomalies on Sahel rainfall. Mon Weather Rev126:2782–2792
Zheng X, Eltahir AB (1997) The response to deforestation anddesertification in a model of West African monsoon. Geo-phys Res Lett 24:155–158
Zinyowera MC, Jallow BP, Maya RS, Okoth-Ogendo HWO(1998) Africa. In: IPCC (eds) The regional impacts of cli-mate change: an assessment of vulnerability. CambridgeUniversity Press, Cambridge, p 29–84
168
Submitted: July 18, 1999; Accepted: October 11, 2000 Proofs received from author(s): February 12, 2001