Rainfall and Water Resources Variability in Sub-Saharan Africa During the Twentieth Century

Home , African Great Lakes, Dam, Rift Valley lakes

FEBRUARY 2009 C O N W A Y E T A L . 41

Rainfall and Water Resources Variability in Sub-Saharan Africa during the Twentieth Century

DECLAN CONWAY School of Development Studies, and Tyndall Centre for Climate Change Research, University of East Anglia, Norwich, United Kingdom

AURELIE PERSECHINO School of Development Studies, University of East Anglia, Norwich, United Kingdom

SANDRA ARDOIN-BARDIN UMR HydroSciences Montpellier, Montpellier, France

HAMISAI HAMANDAWANA Department of Geography and Environmental Sciences, North West University, Republic of South Africa

CLAUDINE DIEULIN AND GIL MAHE´ UMR HydroSciences Montpellier, Montpellier, France

(Manuscript received 21 December 2007, in ﬁnal form 16 May 2008)

ABSTRACT

River basin rainfall series and extensive river flow records are used to characterize and improve understanding of spatial and temporal variability in sub-Saharan African water resources during the last century. Nine major international river basins were chosen for examination primarily for their extensive, good quality flow records. A range of statistical descriptors highlight the substantial variability in rainfall and river flows [e.g., differences in rainfall (flows) of up to 214% (251%) between 1931–60 and 1961–90 in West Africa], the marked regional differences, and the modest intraregional differences. On decadal time scales, sub-Saharan Africa exhibits drying across the Sahel after the early 1970s, relative stability punctuated by extreme wet years in East Africa, and periodic behavior underlying high interannual variability in southern Africa. Central Africa shows very modest decadal variability, with some similarities to the Sahel in the adjoining basins. No consistent signals in rainfall and river flows emerge across the whole of the region. An analysis of rainfall–runoff relationships reveals varying behavior including strong but nonstationary relationships (particularly in West Africa); many basins with marked variations (temporal and spatial) in strength; weak, almost random behavior (particularly in southern Africa); and very few strong, temporally stable relationships. Twenty-year running correlations between rainfall and river flow tend to be higher during periods of greater rainfall station density; however, there are situations in which weak (strong) relationships exist even with reasonable (poor) station coverage. The authors conclude for sub-Saharan Africa that robust identification and attribution of hydrological change is severely limited by data availability, conflicting behavior across basins/regions, low signal-to-noise ratios, sometimes weak rainfall–runoff relationships, and limited quantification of the magnitude and effects of land use change.

Corresponding author address: Declan Conway, School of Development Studies, University of East Anglia, Norwich NR4 7TJ, United Kingdom. E-mail: [email protected]

DOI: 10.1175/2008JHM1004.1

Ó 2009 American Meteorological Society Unauthenticated | Downloaded 10/01/21 03:31 AM UTC 42 JOURNAL OF HYDROMETEOROLOGY VOLUME 10

1. Introduction scale of large river basins [see Hulme et al. (2001) for African overview]. Above average and sometimes ex- a. The significance of water resources variability in treme rainfall in East Africa tends to be associated with Africa periodic circulation dipole events in the Indian Ocean Rainfall and river flows in Africa display high levels and complex interaction with the El Nin˜ o–Southern of variability across a range of spatial and temporal Oscillation (ENSO), particularly during the short Oc- scales, with important consequences for the manage- tober–December rains (Saji et al. 1999; Webster et al. ment of water resource systems (Sutcliffe and Knott 1999). The large decline in many West African river 1987; Grove 1996; Laraque et al. 2001; Conway 2002; flows is primarily related to the effects of the prolonged Ogutunde et al. 2006; Hamandawana 2007). Through- drying in the Sahel (late 1950s–late 1980s), with condi- out Africa, this variability brings significant implica- tions still drier than during the humid 1950s (L’Hoˆ te et tions for society and causes widespread acute human al. 2002; Dai et al. 2004)—though there are exceptions suffering and economic damage. Examples of variabil- (Nicholson 2005). ity include prolonged periods of high flows for rivers Despite the large influence of rainfall fluctuations on draining large parts of East and central Africa (Conway river flow variability, the response may be influenced 2002), and multidecadal anomalies in river flow regimes by other factors such as changes in land use or land in parts of West Africa where long-term mean yields of cover for the Sahel/West Africa (Mahe´ et al. 2005; Li et freshwater into the Atlantic Ocean fell by 18% between al. 2007; Leblanc et al. 2008) and southern Africa (Troy 1951–70 and 1971–89 (Mahe´ and Olivry 1999). There et al. 2007; Woyessa et al. 2006; Lørup et al. 1998). are many examples of the challenges posed by water Human abstractions and reservoir construction also resources variability in Africa: Lake Chad fisheries play a role (Vo¨ ro¨ smarty and Sahagian 2000; Haman- (Sarch and Allison 2000), reservoir management on the dawana et al. 2007) along with land surface to atmo- Senegal River (Magistro and Lo 2001), balancing sup- sphere feedbacks (Savenije 1996). ply and demand for Nile water in Egypt (Conway Depending on the actual hydrological conditions, 2005), irrigation management in the Greater Ruaha the effects of rainfall variability on the hydrologic re- River in Tanzania (Lankford and Beale 2007), and hy- sponse will generally translate into smooth and de- dropower generation in the Kafue (Sutcliffe and Knott layed responses in lake and wetland systems, whereas 1987) and Lake Victoria basins (Tate et al. 2004). semiarid river basins often exhibit low runoff coeffi- As anthropogenic climate change becomes increas- cients and high sensitivity to rainfall fluctuations ingly manifest, the prospect of shifts in flows and vari- (Nemec and Schaake 1982; Li et al. 2005; McMahon ability underscores the need for better understanding of et al. 2007). In addition, the nature of the land surface the drivers of variability and rainfall–runoff interac- itself may increase the variability of river flow re- tions. It is likely that extreme events are going to be the sponses to rainfall fluctuations. Peel et al. (2001, 2004) greatest socioeconomic challenge. Although sub- examined differences in the temporal behavior of Saharan Africa is generally associated with drought- rainfall and river flow between continents and showed related influences, anecdotally there appears to be that variability of annual river flows is higher for greater frequency and spatial extent of damaging temperate Australia, arid southern Africa, and tem- floods, particularly in East Africa and Ethiopia (e.g., perate southern Africa, than for other continents 2006 and 2007). Extreme floods have caused substantial with similar climatic zones. In addition to rainfall socioeconomic disruption in Mozambique (2000; variability, they found that the distribution of ever- Christie and Hanlon 2001) and East Africa (1961, 1978, green and deciduous vegetation in temperate regions and 1997; Conway 2002), whereas smaller floods may was a potential cause of greater river flow variability. be somewhat overlooked but locally significant, for ex- Vo¨ ro¨ smarty et al. (2005) found interesting features ample, in Nigeria (Tarhule 2005). Late 2006–07 saw of African river systems by combining biophysical and major floods of unprecedented spatial extent (and tim- social datasets to show that population distribution is ing) across Somalia, Ethiopia, and other parts of East strongly concentrated in regions exposed to high levels Africa, which is broadly in line with projections in the of interannual variability in rainfall and runoff. Intergovernmental Panel on Climate Change’s Fourth b. Aims Assessment Report for increases in autumn and winter rainfall (Christensen et al. 2007). Although much evidence exists for high interannual The main driver of much of the observed variability and decadal variability in rainfall and river flows in in river flows is—of course—rainfall, particularly at the sub-Saharan Africa, there are few detailed studies of

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC FEBRUARY 2009 C O N W A Y E T A L . 43 their spatial and temporal covariability. Previous at- et al. (2005) for the Okavango. Other river flow series tempts to estimate runoff at this scale have been based and some recent updates to the above series were ob- on the limited data available through international bod- tained from respective national agencies. ies; none have undertaken detailed rainfall–runoff For rainfall, we use the University of East Anglia’s analysis at the basin and subbasin scale, or taken a long Climate Research Unit (CRU) TS 2.1 0.58 resolution historical perspective to incorporate natural variability time series for 1901–2001 from New et al. (2001) and over decadal and century time scales. Rather surpris- updated in Mitchell and Jones (2005). River basin ingly, sub-Saharan Africa provides one of the best op- boundaries upstream of gauging stations were delin- portunities globally to do this type of analysis because eated using the Shuttle Radar Topography Mission many of the very large river basins possess long, rela- (SRTM) from the National Aeronautics and Space Ad- tively natural river flow records. We combine extensive ministration (NASA)/U.S. Geological Survey (USGS) databases held by international and national agencies as the digital elevation model, and ESRI’s Digital Chart to build a comprehensive picture of hydrometeorolog- of the World drainage files to delineate the catchments. ical variability during the twentieth century in large Basin rainfall series were calculated as the average of river basins comprising roughly 32% of sub-Saharan all 0.58CRU TS 2.1 grid boxes within the basin bound- Africa’s area. Our overarching aim is to contribute to ary. Detailed information on CRU TS 2.1 quality control and notes on data interpretation can be found in the scientific understanding of variability in large water relevant publications, but we note here that Africa has resource systems. First, we characterize the spatial and generally poor spatial and temporal coverage of rainfall temporal dimensions of rainfall and river flow covari- stations (Hulme 1996; Nicholson 1996), and this is true ability across the region and second, we examine the for the CRU TS 2.1 data—particularly before the 1930s characteristics and stability of rainfall–runoff relation- and after about 1980. We use gridded time series of the ships over time. number of stations within range of a grid box (Mitchell and Jones 2005; available online at www.cru.uea.ac.uk/ 2. Data sets: Identifying river basins and regions ;timm/grid/stns.html) and calculate a basin average se- for the analysis ries from all grid boxes in the basin. Range is defined as the correlation decay length (450 km for rainfall), so a. Data sources that the series represent the average number of stations We use river flow observations recorded over the with data upon which the grid boxes in the basin may period 1901–2002. These were provided mainly by the draw to calculate rainfall anomalies. Because these se- Institute de Recherche pour le De´veloppement (IRD) ries do not record the actual number of stations that of Montpellier for the West and central African regions have been used to generate rainfall values, we concen- and from a range of international and national sources trate on changes in their relative—rather than their ab- for the East and southern African regions. The data solute— number. provided by IRD come primarily from the SIEREM b. River flow records in major African river basins database (UMR HydroSciences of Montpellier; Boyer et al. 2006; available online at www.hydrosciences.fr/ The study concerns sub-Saharan Africa, divided for sierem). The SIEREM project gathers hydrological and means of presentation and analysis into four regions: climatic data collected by national networks, various west, central, east, and south. Observations are used for international organizations, and research bodies such 26 gauging stations in total, irregularly spaced between as the Food and Agriculture Association (FAO) and and within the regions (Fig. 1). The period of record IRD. IRD updated the data until 1980 (Ardoin-Bardin and details of river gauge and basin characteristics are 2004). Thereafter, updates have been obtained within summarized in Table 1. The river basins range in size the framework of the United Nations Educational, from ;30 200 to 3 475 000 km2. We use three loosely Scientific and Cultural Organization’s (UNESCO’s) Flow applied criteria to select river basins for analysis, which Regimes from International Experimental and Network are in decreasing order of priority: availability of long, Data, West and Central Africa (FRIEND–AOC) project verifiably good quality river flow record; large in area and cooperative undertakings with national agencies. (.10 000 km2); and spatial coverage across sub- Data for East Africa come primarily from Hurst (1933) Saharan Africa. In some situations, we relaxed the cri- for rivers in the Nile basin [Kagera, Lake Victoria out- teria to maximize spatial coverage. Figure 1 shows that flows, Blue Nile, Equatorial Lakes, Sobat and Atbara in general, the gauges represent the key upstream con- rivers; see Conway and Hulme (1993); Sutcliffe and Parks tributing areas of these large river basins and in some (1999)], UNESCO (1995) for the Tana, and Hamandawana situations, combinations of upstream and downstream

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC 44 JOURNAL OF HYDROMETEOROLOGY VOLUME 10

FIG. 1. The main drainage basins analyzed and location of the gauging stations (refer to Table 1). gauges (Niger and Ogooue´). In all situations at annual construction of the Owen Falls dam in 1954, the out- time scales, the river flow records are—for the most flows from Lake Victoria have been regulated to follow part—unaffected by human influences in the form of the natural relationship between lake level and out- upstream dams and major abstractions. Because of the flows (an ‘‘agreed curve’’; Tate et al. 2004). After a rise

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC FEBRUARY 2009 C O N W A Y E T A L . 45

TABLE 1. General characteristics of the river basins included in the analysis (locations in Fig. 1).

Basin area Period for river Basin Gauging station River Lat (8N) Lon (8E) (km2) flow analysis West Africa Niger Koulikoro Niger 12.87 27.55 120 332 1907–99 Douna Niger 13.22 25.9 101 226 1922–97 Dire Niger 16.27 23.38 341 066 1924–98 Niamey Niger 13.39 2.18 631 381 1947–96 Makurdi Benue 7.75 8.53 303 637 1955–92 Onitsha Niger 6.18 6.77 1 388 334 1950–87 Seńe´gal Oualia Bakoy 13.6 210.38 78 155 1904–89 Bakel Senegal 14.9 212.45 220 818 1904–2000 Volta Dapola Black Volta 10.57 22.92 86 559 1952–94 Senshi Halcrow Volta 6.2 0.09 388 154 1936–79 Chari N’djamena Chari 12.12 15.03 601 984 1952–2000 Central Africa Congo Bangui Oubangui 4.37 18.58 485 478 1936–98 Ouesso Sangha 1.62 16.05 159 016 1948–96 Kinshasa Congo 24.30 15.30 3 475 000 1903–96 Ogooue´ Makokou Ivindo 0.57 12.86 48 912 1955–83 Fougamou N’Gounie 21.22 10.59 48 912 1954–83 Lambareńe´ Ogooue 20.71 10.23 205 418 1930–89 East Africa Nile Nyaka Ferry Kagera 21.20 31.25 30 200 1940–78 Owen Falls L. Victoria 0.43 33.23 258 000 1901–89 Owen Falls* Equatorial Lakes 0.43 33.23 293 000 1905–82 Hillet Doleib Sobat 9.20 31.38 231 000 1905–84 Kilo 3 Atbara 17.68 34.02 188 200 1903–82 El Deim Blue Nile 11.23 34.98 195 000 1912–2002 Tana Garissa Tana 20.45 39.70 42 220 1934–75 Southern Africa Zambezi Victoria Falls Zambezi 217.95 25.90 360 683 1907–90 Okavango Mohembo Okavango 218.12 21.68 238 700 1933–99

* Equatorial Lakes represents the difference between White Nile river flows measured at Owen Falls and downstream at Mongalla (see Conway and Hulme 1993). in the lake in the late 1990s, outflow departed from the de la Recherche Scientifique et Technique d’Outre- agreed curve to alleviate flooding around Lake Kyoga Mer (ORSTOM), now IRD]. In the Nile River basin, downstream (Goulden 2006; Sutcliffe and Petersen Egyptian and Sudanese interests have maintained ex- 2007), but we only use data up to 1989. There is a major tensive hydrological records, although conflict in south- dam on the Senegal River, the Manantali dam, so we ern Sudan has undermined these efforts since 1983. use reconstructed natural discharges for the down- There are only a few large international river basins in stream series at Bakel, Senegal (Bader 1990, 1992). The East Africa that drain into the Indian Ocean because Niger River has some dams but these have only minor the region is dominated by the complex internal hydrol- effects on the annual time scales used in this anlaysis. ogy of the Rift Valley Lakes system. Extensive, reliable The discharge series of the Senshi Halcrow gauging lake level records exist for many of these lakes and point is influenced by the Akosombo dam mainly during have been described in detail by Nicholson (1998, the first years of filling (1964–67). Its interannual vari- 1999). Rather surprisingly, southern Africa has few ability remains similar to that of neighboring rivers, but long duration records for its larger river basins, partly the monthly regime has been modified as a result of because the effects of human influence (dams and ab- regulation and the total discharge has been reduced stractions) very early on in the Limpopo and Orange because of evaporation (Moniod et al. 1977). Rivers have restricted the compilation of such records. Interestingly, West and East Africa are well served Our coverage is, therefore, limited to the Zambezi by long river flow series primarily as a result of French, (measured at Livingstone, upstream of Lake Kariba) English, and Anglo-Egyptian interests in water re- and Okavango Rivers. We also analyzed the Olifants, a sources development from the early colonial period in major tributary of the Limpopo, but decided the flow Africa (circa 1880s) to independence (1960s onwards). record was too difficult to naturalize. Finally, the Horn Since independence, western and central African coun- of Africa, Ethiopia, and Somalia are poorly represented tries have tended to receive greater support for coor- because of limited data availability, especially in east- dinated data collection [particularly through the Office ern and southern Ethiopia where some very large river

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC 46 JOURNAL OF HYDROMETEOROLOGY VOLUME 10

TABLE 2. Long-term conditions for rainfall and river flows: annual mean, CV, runoff coefficient (RC), and percentage change in rainfall and river flow between 1931 and 1960, and between 1961 and 1990. Refer to Table 1 and Fig. 1 for locations.

1931–60 1961–90 % change Gauge Rainfall CV River flow CV RC Rainfall CV River flow CV RC River location (mm) (%) (m3 s21) (%) (%) (mm) (%) (m3 s21) (%) (%) Rainfall flow West Africa Koulikoro 1619 7 1529 20 25 1436 10 1213 33 22 211 221 Douna 1230 10 — — — 1066 14 332 64 9 213 251 Dire 1170 8 1101 20 9 1020 12 870 29 8 213 221 Niamey 807 9 — — — 693 13 804 29 6 214 220 Makurdi 1268 8 — — — 1175 11 3152 31 27 27— Onitsha 992 8 — — — 898 11 5580 22 14 210 — Bakel 953 12 764 32 11 808 17 549 48 9 215 241 Oualia 927 12 170 34 7 793 16 116 68 5 214 234 Senshi Hal. 1080 8 1204 38 9 992 12 1044 68 8 28 213 Dapola 963 9 — — — 836 15 89 46 4 213 — N’Djamena 1047 8 — — — 942 12 892 42 5 210 — Central Africa Kinshasa 1567 4 40584 6 24 1567 6 42418 13 25 0 15 Ouesso 1537 9 — — — 1550 7 1571 17 20 11— Bangui 1555 10 4265 11 18 1517 8 3740 28 16 22 212 Lambarene 1762 9 4761 16 42 1764 9 4587 13 40 0 24 Makokou 1597 10 — — — 1616 9 — — — 11— Fougamou 1839 15 — — 1862 16 712 19 39 11— East Africa Nyaka Ferry. 1058 9 412 18 42 1084 9 662 21 63 12 161 Owen Falls 1129 12 672 15 7 1222 12 1190 17 12 18 177 Owen Falls* 1147 9 105 67 1 1167 12 372 21 3 12 1254 Hillet Doleib 1039 11 408 14 5 974 21 448 21 6 26 110 El Deim 1070 10 1626 13 25 1010 10 1454 20 23 26 211 Kilo3 644 22 390 26 10 582 21 294 36 8 210 225 Garissa 615 19 131 41 16 706 26 — — — 115 — Southern Africa Victoria F. 860 17 1171 33 12 857 15 1183 38 12 0 11 Mohembo 777 20 763 23 13 743 17 869 22 16 24 114

* Owen Falls represents the Equatorial Lakes, the difference between White Nile river flows measured at Owen Falls and downstream at Mongalla. basins (e.g., Omo and Wabe Shebelle) remain sparsely of 20-yr moving average correlations. Annual rainfall– instrumented and understudied. runoff plots are used to identify shifts and spatial differences in relationships. The runoff coefficient represents the ratio (expressed as a percentage) of rainfall to 3. Methods of analysis runoff—that is, the fraction of total rainfall that be- We characterize rainfall and runoff for different pe- comes river flow. riods; World Meteorological Organization (WMO) normals (1901–30, 1931–60, and 1961–90) and periods before and after notable breakpoints were identified 4. Rainfall and river flow variability using statistical tests. Means, coefficients of variation a. Long-term conditions (CV), and selected indicators of trend (based on linear regression) and temporal variability are presented. Table 2 shows descriptive statistics for rainfall and Breakpoints in time series are identified using Khrono- river flows based on the 1961–90 WMO period along stat 1.0 software (Lube`s-Niel et al. 1998a) for nonpara- with percent differences from the previous 30-yr period metric tests and segmentation tests, including Hubert’s from which data are available. The West African rivers segmentation, Pettitt, Lee and Heghinian, and Buis- are mainly strongly seasonal and humid with fairly mod- hand tests (Lubes-Niel et al. 1998b). The relationship est interannual rainfall variability. In all cases, river between rainfall and river flow is examined using plots flows show much greater CVs mainly because of the

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC FEBRUARY 2009 C O N W A Y E T A L . 47

FIG. 2. Annual rainfall (black line) and river flow (gray line) series for West Africa stations (river in parentheses): Douna (Niger), Dapola (Volta), Koulikoro (Niger), Oualia and Bakel (Senegal), and N’Djamena (tributary of Lake Chad). Note different vertical scales and record lengths for river flow (refer to Table 1) and standard record lengths for rainfall (1901–2002). heterogeneity and nonlinear response of runoff to mogeneity, with all rivers studied exhibiting broadly changes in rainfall, especially to variations in rainfall similar temporal behavior. Changes in rainfall and intensity. Groundwater interactions also contribute: runoff between 1901–30 and 1931–60 (not shown) are variability of groundwater levels is linked to cumulative modest, with rainfall ranges from 22% to 8% and river rainfall anomalies and can affect runoff over prolonged flow from 21% to 2% (only five rivers have data for periods independently of the rainfall anomaly of a both periods). specific year (Mahe´ et al. 2000). Runoff coefficients Central Africa is dominated by the Congo River and are fairly low and show considerable variation, ranging its major tributary, the Bangui. Rainfall and river flows from around 4% to 27%. The period is marked by the are fairly stable from year to year, and annual means large negative trend in rainfall and river flows, which show little variation between 1931–60 and 1961–90, ex- occurs in all the West African rivers and has been cept for the decrease of flows of the Bangui. widely documented (Janicot 1992; Paturel et al. 1997, East and southern Africa show greater heterogene- 1998; Mahe´ and Olivry 1999; Mahe´ et al. 2001; Leduc ity, both within and between regions. Interannual vari- et al. 2001; Le Barbe´ and Lebel 1997). Time series show ability tends to be highest in the drier basins—higher this event is characterized by a shift rather than a trend. than in West African basins with an equivalent annual Proportionally, the shift is much greater in river flows rainfall. River flows are generally less variable than (from 213% to 251%) than rainfall (from 27% to in West Africa, with the exception of the Atbara. Rain- 214%). West Africa shows strong intraregional ho- fall and river flows in all basins show decreasing

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC 48 JOURNAL OF HYDROMETEOROLOGY VOLUME 10

FIG. 3. Same as Fig. 2 but for central Africa: Kinshasa (Congo) and Oubangui (Bangui).

trends from 1961–90. Runoff coefficients range consid- 1980s and have stabilized, and in most cases recovered erably—from 3% to 63%—because of the effects of somewhat. high evaporative demand and transmission losses (At- The Chari and Logone Rivers are the main tributar- bara River), transmission losses (Sobat River and ies of Lake Chad, and they join at N’Djamena, Chad, to Sudd swamps), and lake evaporation (Lake Victoria). form the N’Djamena River just before entering the Changes between 1931–60 and 1961–90 are mixed: lake. Although it is the most easterly of this group of three basins show modest decreases in rainfall, three rivers, it shows similar temporal behavior, with its de- show almost no change, and two show modest to large crease leading to the dramatic lowering and shrinking increases. These trends are associated with some very of Lake Chad since the 1970s (Lemoalle 2004). Under large river flow responses, not easily explained by the present climatic conditions, these two rivers contribute rainfall changes and are explored in more detail in ;90% of the lake’s water, with the remaining 10% section 5. The two southern African rivers possess coming from local rainfall and deliveries by the El Beid slightly different climatic conditions: the upper Zam- and Komadougou Yobe Rivers (Birkett 2000). Severe bezi is humid seasonal and the Okavango is closer to droughts occurred during the 1970s and 1980s, leading semiarid seasonal, and both rivers have modest inter- to widespread sinking of boreholes and to the develop- annual rainfall and river flow variability and quite low ment of a large-scale irrigation system that promoted runoff coefficients. Mean rainfall between 1931–60 and water loss through evaporation (Birkett 2000). The 1961–90 is fairly stable as is the river flow in the Zam- more ‘‘recent’’ reduction in the lake’s surface area is the bezi, whereas a 14% increase was recorded in the Oka- outcome of sustained decrease in river flow as a result vango. of persistent rainfall failures, human-induced increases in evaporation losses, and poor management of irrigation water. b. Decadal and interannual variability The two longest river flow records for central Africa A sample of rainfall and river flow records for West are shown in Fig. 3. The region is dominated by the Africa is shown in Fig. 2. These highlight strong re- Congo and by the Bangui, which forms its main north- gional homogeneity in temporal behavior because all ern tributary—draining large parts of the Central Afri- the series show the marked downturn in rainfall and can Republic, south of Lake Chad (in fact, the decrease river flow around 1970 that characterizes the climate of in rainfall and flows post-1970 is similar to the West the Sahel during the last century. Many series also show African region). The Congo’s rainfall and river flows humid conditions during the 1950s and 1960s. These have been quite stable, with only the 1960s (wet) and features have been analyzed in detail, for example, 1980s (dry) showing any decadal patterns of variability. rainfall by Lamb (1982), Nicholson (1983), and Hulme The other river basins (series not shown) exhibit quite (1992); river flow by Sircoulon (1976, 1985); and both marked interannual variability (high CVs) and modest rainfall and river flow by Mahe´ and Olivry (1999). Un- decadal or trend-like patterns. fortunately, it has not been possible to update most Figure 4 shows four examples from East Africa, of these series beyond the 1990s; however, recent again chosen for display because of their long river studies by L’Hoˆ te et al. (2002) and Dai et al. (2004) flow records. Temporal behavior in river flows—and to note that rainfall in the Sahel has not returned to those a lesser extent rainfall—is regionally less homogeneous. conditions prior to the early 1970s. Rainfall and river The Blue Nile (and Atbara, but not shown) display flows certainly reached their lowest point in the mid- some features similar to West Africa: humid 1950s and

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC FEBRUARY 2009 C O N W A Y E T A L . 49

FIG. 4. Same as Fig. 2 but for East Africa: Blue Nile, Sobat, Lake Victoria outflows and Tana. dry 1970s and 1980s, but it has recovered more than decadal variability in river flows, which is discussed be- West African river flows have. The Sobat River is more low. stable and shows intermediate behavior to the Blue Across the whole of sub-Saharan Africa, maximum Nile and Atbara to the north and Lake Victoria and 20-yr trends expressed as percent of long-term means other rivers to the south. The marked rise in outflows have ranged from 63% per annum for rainfall and from Lake Victoria after 1961 has been explained using from 215% to 111% per annum for river flow (not the lake’s water balance to show it resulted from a se- shown). ries of extremely wet years in the 1960s and a slight increase in the short rains after the 1960s, combined with lake storage effects (Piper et al. 1986; Sene and 5. Rainfall–runoff relationships Plinston 1994; Conway 2002). Many other East African a. Regional relationships lakes show marked increases in level in 1961 (and in other years such as 1968, 1978, 1982, and 1997), but Figure 6 shows the strength of regression relation- these increases have been much shorter in duration. ships between annual rainfall and runoff during 1961– The smaller rivers such as Tana (Fig. 4) and Kagera 90 and 1931–60. Rivers in West Africa generally display show the short-lived effects of major rainfall extremes very strong relationships, with rainfall accounting for that produce high levels of interannual variability with around 60%–70% of river flow variability. In central modest decadal variability. The two rivers in southern Africa, relationships are slightly weaker but still quite Africa (Fig. 5) have stable rainfall series but quite high robust (around 50% variance explained). Relationships

FIG. 5. Same as Fig. 2 but for southern Africa: Zambezi and Okavango.

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC 50 JOURNAL OF HYDROMETEOROLOGY VOLUME 10

2 FIG. 6. Strength of regression relationships (R ) between rainfall and runoff for the 30-yr periods 1931–60 and 1961–90 (results shown for all rivers with .20 yr flow data). in East and southern Africa are substantially weaker, 1986; Ba and Nicholson 1998). Rivers with less complex with the exception of the Blue Nile, which drains much hydrology show better results such as the Blue Nile (Fig. of central and northern Ethiopia. To help explain some 7) and Tana (not shown). The two rivers in southern of these differences, Fig. 7 shows the nature of the re- Africa are shown in Fig. 7 to have very weak relation- lationships that are typical of each region. The Niger at ships as a result of their complex runoff response to Koulikoro, Mali (upstream of the Niger Inland Delta), rainfall. Further exploration of the basin’s physiography is fairly typical of West Africa. Nearly all the rivers may shed light on this; although, data quality for the show much weaker (and generally lower gradient) re- Zambezi may be a critical factor. We have been unable lationships during the period 1901–30, with stronger re- to trace any metadata for this river flow record and, lationships (and generally higher gradient) across both therefore, its accuracy is open to question (see next 1931–60 and 1961–90. In central Africa the results section). In addition, the upper basin, which drains parts are less consistent: fairly weak relationships in the of Angola, is unlikely to have had many rain gauges Ogooue´ River of Gabon (two flow series have wide present throughout the whole period. scatter and no obvious patterns; plots not shown) and b. Analysis of nonstationary behavior stronger but highly unstable relationships in the Congo and Bangui. Both of these rivers only produced strong It is clear from the previous section that significant relationships during 1961–90; Fig. 7 shows almost ran- shifts in rainfall, runoff, and rainfall–runoff relation- dom patterns for the time prior to that period, al- ships have occurred across sub-Saharan Africa. To ex- though a moderately positive relationship exists for the plore this in more detail, Table 3 lists breakpoint years Congo. The other rivers in central Africa generally and whether series show evidence of breakpoints using show reasonable linear relationships but data are only four statistical tests (described in section 2). The shift/ available for 1961–90, so it is not possible to comment discontinuity in West African rainfall and river flow on their temporal stability. series around 1970 has been previously documented In East Africa, many of the rivers possess complex (Paturel et al. 1998; Mahe´ and Olivry 1999; Mahe´ et al. drainage systems as a result of the Rift Valley and other 2001) and is common to all the series presented here. physiographical features. For example, the Sobat River What this analysis identifies is the possible existence of and Equatorial Lakes experience, respectively, the non- changes in rainfall–runoff relationships as highlighted linear effects of overbank losses and lake level–area by large differences between intercepts (smaller dif- effects on inflow–outflow relationships, resulting in weak ferences in slope), which are very clear in the cases relationships with significant nonstationary behavior (see of stations Douna (Fig. 8), Niamey, Dire, Bakel, and below). Outflows from Lake Victoria show a very weak N’Djamena (not shown). Some rivers show a greater relationship to basin rainfall—even during periods of change in the slope factor such as the Niger (Koulikoro; stationary conditions. Lake outflows are constricted so Fig. 8), Makurdi, and Oualia. In the Sudano–Guinean that their response to wet years is attenuated, leading to area, the prolonged rainfall decline since the beginning a smoothed response. Difficulties in estimating lake of the 1970s led to a persistent deepening of ground- rainfall have been identified in efforts to model the water levels. The percentage of baseflow in the annual lake’s water balance (the lake has an area of about 78 discharge of all rivers in West Africa is, therefore, 000 km2 out of a basin area of 258 000 km2; Piper et al. correspondingly lower because the drought exacerbates

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC FEBRUARY 2009 C O N W A Y E T A L . 51

FIG. 7. Rainfall–runoff relationships for up to three different 30-yr periods, with data permitting for West Africa, Niger; central Africa, Congo and Bangui; East Africa, Lake Victoria and Blue Nile: and southern Af- rica, Zambezi and Okavango.

the effects on river flows (except for Sahelian rivers, Figure 9 shows time series of 20-yr running correla- where the groundwater contribution to surface runoff is tions between rainfall and river flows for three of the insignificant; Mahe´ et al. 2005). This is visible via the West African rivers and the total number of rainfall increase of the depletion coefficient (Bricquet et al. stations within range of all the grid boxes in the basins. 1997; Orange et al. 1997; Olivry et al. 1998; Mahe´ et al. The temporal pattern for the Niger at Koulikoro is 2000), which means that the groundwater resources similar to that of the Senegal River at Bakel, showing have declined, rapidly draining out since the 1970s. highest correlations between the 1950s and 1980s, These shifts are likely to primarily reflect nonlinear which is the period with the best rainfall station cov- dynamics in runoff response; however, because of the erage. Correlations tend to strengthen moderately from prolonged duration of the change in rainfall patterns, the 1920s to the 1950s (as station density rises), and they may also incorporate the effects of land cover most records show a rapid (step like) decay roughly change. around 1980, at which point data from the 1990s would

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC 52 JOURNAL OF HYDROMETEOROLOGY VOLUME 10

TABLE 3. Breakpoint years identiﬁed using four statistical tests on rainfall (top) and river ﬂow series (bottom). Here, A 5 test is accepted, the series are ‘‘stationary’’; R 5 test is rejected, the series are not ‘‘stationary’’; ‘‘—’’ 5 no break detected; and NA 5 test is not applied (series nonnormal or gap/missing data).

Region and river Station Buishand Pettitt Lee and Heghinian Hubert West All stations show similar results R R 1967 1967 1967 1970 1971 1972 Central Tests are accepted for most of the A A — no common date — stations, except: Oubangui Bangui A R 1970 1970 1935, 1938, 1969 East Tests are accepted for most of the A A — no common date — station, except: Sobat Hillet Doleib R R 1967 1978 1960, 1961, 1978 Kagera Nyaka Ferry R R 1929 1929 1929, 1997 Southern All stations show different results Zambezi Victoria Falls A A — 1901 — Okavango Mohembo A A — 1978 — Region–River/Lake Station Buishand Pettitt Lee and Heghinian Hubert West All stations show similar results, except: R R 1967 1967 1967 1971 1971 1971 Niger Makurdi R R 1976 1971 1981 1988 Chari Ndjamena R R 1971 1971 1964, 1982 Central All stations show similar results, except: R R 1970 1969 1969 1971 1971 Congo Kinshasa R A 1922 1981 1959, 1969, 1981 Ogooue´ Fougamou A A — 1977 1977 East All stations show different results Kagera Nyaka Ferry R R 1961 1961 1961, 1964 Lake Victoria Owen Falls NA R 1961 NA 1961, 1970 Equatorial Lakes Owen Falls R R 1960 1919 1916, 1918, 1960 Sobat Hillet Doleib A A — 1981 — Blue Nile El Deim A A — 1913 1913, 1978, 1987 Atbara Kilo3 R R 1964 1964 1915, 1916, 1964 Tana Garissa R R 1955 1955 1960, 1968 Southern All stations show different results Zambezi Victoria Falls R R 1945 1945 1945, 1980 Okavango Mohembo R R 1986 1992 1960, 1969, 1992

just begin to feed into the relationships while data from Similar observations hold for the Bangui, where the 1970 are removed. This behavior most likely reflects absolute low numbers of rainfall stations and their the dramatic decline in rainfall stations after 1990, with change over time are likely to account for shifts in the numbers in most instances dropping from more than nature of their rainfall–runoff relationships. 50 to fewer than 10 between the late 1980s and the early East Africa has had better rainfall station coverage 1990s. than central Africa, although not as good as West Af- The time series of rainfall stations in central Africa rica. Figures 8 and 9 show that the Blue Nile has had a (Fig. 8; Congo and Bangui) shows their extreme scar- remarkably stable and robust relationship over time, city for this region during most of the last century. The which is somewhat surprising given the low number of Congo shows a rainfall break point in 1980 (1981 river rainfall stations contributing to the rainfall series and flows) that is associated with a modest difference in the the complex large basin (Conway 2000). The Atbara, slope and a marked reduction in the strength of the just to the north and with a similar number of contrib- rainfall–runoff relationship. The running correlation is uting stations, has a weaker and much less stable rela- highly unstable until 1940 because no rainfall stations tionship and a breakpoint in river flows that occurred in contributed to the basin until 1930; from then on, only 1964 when runoff decreased. This phenomenon may be five rainfall stations contributed to this vast basin. Con- related to the construction of the Khasm el Girba res- sidering the paucity of data, the relationship with river ervoir upstream from the river gauge in the early 1960s. flow is remarkably strong from the 1950s to the 1980s. The Sobat and Kagera both show very weak rainfall–

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC FEBRUARY 2009 C O N W A Y E T A L . 53

FIG. 8. Rainfall–runoff relationships before and after breakpoint years identiﬁed from rainfall series for West Africa, Niger (Koulikoro and Douna); Central Africa, Congo and Bangui; East Africa: Lake Victoria, Atbara, and Blue Nile; and Southern Africa, Zambezi. runoff relationships probably because of the combined Equatorial Lakes and Kagera River series) associated effects of low rainfall station density and substantial with the dramatic rise in lake level between 1961 and internal storage in wet years because of their complex 1964. wetland hydrology. The relationship between Lake The previous section highlighted the weak relation- Victoria outﬂows and rainfall (Figs. 8 and 9) clearly ships between rainfall and runoff in both of our south- shows the nonstationary behavior (also present in the ern African basins, where rainfall station coverage is

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC 54 JOURNAL OF HYDROMETEOROLOGY VOLUME 10

FIG. 9. Twenty-year running correlation between rainfall and river flow (black line) and the total number of rainfall stations (gray line) within range of grid boxes in the basin for West Africa, Niger (Koulikoro and Douna); central Africa, Congo and Bangui; East Africa, Atbara and Blue Nile; and Southern Africa, Zambezi and Oka- vango. poor and running correlations between rainfall and exists before and after the 1945 breakpoint in river river flows are generally weak (Fig. 9). For the Zam- flow, with substantially higher runoff after this point bezi, before ;1960 the correlation is very low (with a being possibly related to the integrity of the river flow short peak caused by a couple of years ‘‘in phase’’ dur- series (Fig. 9). The early part of the Okavango record ing the 1930s) and after ;1960 it slowly increases. A produces reasonable correlations but from the 1950s, quite marked shift in the rainfall–runoff relationship these decrease to random behavior. Neither series

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC FEBRUARY 2009 C O N W A Y E T A L . 55 shows a clear relationship with station density, a result the combination of data coverage and quality, possibly that is somewhat surprising and difficult to explain. exacerbated by local physical conditions beyond the scope of this basin-scale analysis, for example, the im- 6. Conclusions pact of geology on the regime of the right-bank tributaries of the Congo River (Laraque et al. 2001) and the a. Hydrometeorological data in Africa Niger River (Mahe´ et al. 2000). It is not easy to explain The collection, quality control, and regular update of why some basins show robust stable relationships, with datasets used in this analysis results from the long-term rainfall series composed of relatively few gauges (e.g., collective efforts of many individuals and funding the Blue Nile). sources. The compilation and quality assurance of the Overall, the best period for analysis is broadly 1961– hydrological records have been time consuming and 90, with the strongest relationships occurring in West even with extensive contacts across Africa, we have Africa, reasonable relationships in central (even though not been able to update many records. Our choice of station densities are very low), highly variable relation- river basins and hydrological series represents most of ships in East Africa because of the Rift Valley’s com- the key extensive, high-quality records for the large plex hydrology, and very weak relationships in south- international rivers in sub-Saharan Africa. These se- ern Africa. These variations deserve further study (see ries are a valuable scientific resource and should be below); however, one important implication is that for recognized as benchmark stations for studying envi- many of the basins analyzed here macroscale modeling of ronmental change. The decline in the overall number variability using these data will be limited in accuracy. of sites, their frequency of reporting, and—in some c. Land use and land cover change cases—quality of measurements are major concerns for the understanding of environmental change in The high variability and weak rainfall–runoff rela- sub-Saharan Africa. Although this is part of a global tionships means that it is difficult to identify and attrib- phenomenon with hydrological data (Vo¨ ro¨ smarty et al. ute changes in runoff to particular causes, such as cli- 2001), the situation is particularly bad in sub-Saharan mate change, or land use or land cover change (LUCC). Africa. The massive decline in rainfall stations used Recent work on global precipitation variability has in CRU TS 2.1 from the 1980s onward severely identified a climate change signal that includes Sahel constrains efforts to accurately monitor climate vari- drying (Zhang et al. 2007). However, no other clear ability and confidently model biophysical systems. This patterns have emerged across Africa. Gedney et al. is part of Africa’s wider financial, political, and insti- (2006) identify a direct effect of CO2 on transpiration tutional challenges to undertaking climate research and global runoff patterns that they postulate has con- (Washington et al. 2006). tributed to a recent increase in global runoff. Our results identify marked changes in runoff ratios (espe- b. Rainfall–runoff relationships cially in the Sahel after 1970) but these integrate Our results show a complex pattern of behavior that changes in data, nonlinearities in the runoff response, includes strong but nonstationary relationships, with and the effects of LUCC. We find that in sub-Saharan most examples in West Africa; a large group from Africa, robust identification and attribution of hydro- across Africa with marked variations in strength, often logical change is, therefore, severely limited by conflict- but not always, showing the influence of rainfall station ing behavior across basins/regions, low signal-to-noise density; weak almost random behavior (particularly in ratio, sometimes weak rainfall–runoff relationships, southern Africa but examples occur in all other re- and limited assessment of the magnitude and potential gions); and very few examples of strong, temporally effects of LUCC or other anthropogenic influences. An stable relationships. important area deserving further research that we have For some basins, limited spatial coverage of rainfall not looked at is the role of evaporation in rainfall– stations leads to weak rainfall–runoff relationships. In runoff analyses. many cases, this limits the ability to establish robust d. Future climate change relationships very far back in time (generally prior to the 1950s). However, there are situations in which weak The high levels of variability found in the historical relationships exist throughout the period of analysis, records provide excellent opportunities to better under- even with reasonable station coverage. There are no stand their societal effects and adaptive responses obvious reasons for this, particularly because some ba- (Glantz 1992; Adger et al. 2003). The areas with rea- sins produce good results with relatively few rainfall sonable climate model convergence show runoff in- stations. We surmise that the most likely reasons are creases in East Africa and reductions in southern

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC 56 JOURNAL OF HYDROMETEOROLOGY VOLUME 10

Africa but no clear signal for the Sahel and central d Station densities in CRU TS 2.1 (and other similar Africa (Milly et al. 2005; Christensen et al. 2007). The products) for Africa before ;1930 and after ;1980 main and most understood climate drivers of interan- are very low, and some basins have very low densities nual and decadal rainfall variability are Atlantic (and throughout their records (e.g., Congo: maximum of 6 2 other) Ocean SST patterns (West Africa and the Sahel), five gauges across 3.5 3 10 km ). Care is required in ENSO behavior (West, southern and East Africa), and the interpretation of time series for these periods and Indian Ocean dynamics (East Africa and southern Af- regions. rica). However, the underlying drivers of variability in d Trends in rainfall and river flows have been large these factors and their African teleconnections are not during the twentieth century. Rainfall (river flows) well captured by climate models, and model simula- have displayed 20-yr moving trends of up to 63% tions of future climate do not show clear tendencies in (215%/111%) of annual means per year. Changes in their behavior (e.g., ENSO; Merryfield 2006; Indian rainfall are magnified in the runoff response. This Ocean; Conway et al. 2007). Improvements in physical level of variability presents significant challenges to understanding and modeling capability will hopefully water resources management. improve confidence and lead time in seasonal forecasts d On decadal time scales, sub-Saharan Africa is char- and climate projections, although nonstationary behav- acterized by drying across the Sahel after the early ior in teleconnections may affect progress toward these 1970s, relative stability punctuated by extreme wet goals (Richard et al. 2000). years in East Africa (sometimes spreading into the Congo basin, e.g., 1961), and periodic behavior un- e. Overall conclusions derlying high interannual variability in southern Af- The conclusions are based around our aims, which rica. Central Africa shows very modest decadal vari- for sub-Saharan Africa were to characterize the spatial ability, with some similarities to the Sahel in adjoin- and temporal dimensions of rainfall and river flow vari- ing basins. A subregion of East Africa, the Horn, ability. The region is possibly unique in its possession of shows drying in the 1970s and 1980s similar to the so many large, relatively undisturbed river basins in Sahel but has recovered substantially during the which to study long-term hydrometeorological behav- 1990s. ior. Rainfall records from a global high-resolution d Runoff coefficients tend to increase with increasing product and extensive river flow records from nine ma- annual rainfall; they show a widespread decrease in jor international river basins provide the basis for the West Africa after the 1970s drought but no consistent analysis. The early and latter decades of last century patterns elsewhere. generally show very sparse coverage of rainfall stations. d Overall, the best period for robust rainfall–runoff re- In most cases, we are confident the hydrological series lationships analysis is broadly 1961–90. The strongest are reliable and possess thorough supporting informa- relationships occur in West Africa, reasonable relation, although the Zambezi record is an exception to tionships in central (even though station densities are this. very low), highly variable relationships in East Africa Our findings confirm that rainfall variability in the because of the Rift Valley’s complex hydrology, and region is high, but also that rainfall provides the dom- very weak relationships in southern Africa. During inant control along with river basin physiography and the period 1961–90 (1931–60), rainfall explains 60%– human interventions on interannual and interdecadal 80% (40%–60%) of the variability in river flows in variability in river flows and hence surface water West Africa. Equivalent approximations for other availability. River flows in major basins show clear ex- regions are 40%–70% (insufficient data) in central amples of significant variability that challenge the ef- Africa, 5%–65% (5%–65%) in East Africa, and only fective management of water resources and result in 5%–20% (5%–20%) in southern Africa. huge socioeconomic costs. Although this work has con- d For some basins, limited spatial coverage of rainfall centrated on biophysical variability, we recognize an stations leads to weak rainfall–runoff relationships. important need to link this understanding with the in- However, there are examples in which weak relation- stitutional and policy context of water resources man- ships exist throughout the period of analysis, even agement in Africa. More effective management of vari- with reasonable station coverage and examples of ability (the foundation for adaptation) is contingent robust stable relationships with rainfall series com- upon operational capacity, which is weak in many parts prising relatively few gauges. In basins with weak of Africa. relationships, macroscale modeling using these data The main findings of this analysis are presented be- will be of limited success without considering data low. and subbasin scale conditions.

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC FEBRUARY 2009 C O N W A Y E T A L . 57 d Our results identify marked changes in runoff ratios African Great Lakes: Limnology, Palaeolimnology and and nonstationary rainfall–runoff relationships (espe- Biodiversity, E. O. Odada and D. O. Olago, Eds., Advances in cially in the Sahel after 1970), which integrate Global Change Research Series, Vol. 12, Kluwer, 63–92. changes in data, nonlinearities in the runoff response, ——, 2005: From headwater tributaries to international river: Ob- serving and adapting to climate variability and change in the and the effects of LUCC. We conclude that for sub- Nile basin. Global Environ. Change, 15A, 99–114. Saharan Africa, robust identification and attribution ——, and M. Hulme, 1993: Recent fluctuations in precipitation and of hydrological change is severely limited by data runoff over the Nile subbasins and their impact on main Nile limitations, conflicting behavior across basins/re- discharge. Climatic Change, 25, 127–151. gions, low signal-to-noise ratios, sometimes weak ——, C. E. Hanson, R. Doherty, and A. Persechino, 2007: GCM simulations of the Indian Ocean dipole influence on East rainfall–runoff relationships, and limited assessment African rainfall: Present and future. Geophys. Res. Lett., 34, of the magnitude and potential effects of LUCC or L03705, doi:10.1029/2006GL027597. other anthropogenic influences. Christensen, J. H., and Coauthors, 2007: Regional climate projections. Climate Change 2007: The Physical Science Basis, S. Solomon et al., Eds., Cambridge University Press, 849–940. Acknowledgments. The authors wish to acknowledge Christie, F., and J. Hanlon, 2001: Mozambique and the Great John Sutcliffe, Dick Grove, and Jacques Sircoulon for Flood of 2000. International African Institute and James Cur- their pioneering work on this topic. We also thank the rey, 176 pp. many African hydrological agencies for the information Dai, A. D., P. J. Lamb, K. E. Trenberth, M. Hulme, P. D. Jones, and P. Xie, 2004: The recent Sahel drought is real. Int. J. and data that supported this analysis. The United King- Climatol., 24, 1323–1331. dom–France collaboration was facilitated by a British Gedney, N., P. M. Cox, R. A. Betts, O. Boucher, C. Huntingford, Council Alliance/EGIDE award (2004–06). Helpful and P. Stott, 2006: Detection of a direct carbon dioxide effect comments from three reviewers improved this manu- in continental river runoff records. Nature, 439, 835–838, script. doi:10.1038/nature04504. Glantz, M. H., 1992: Global warming and environmental change in sub-Saharan Africa. Global Environ. Change, 2, 183–204. Goulden, M., 2006: Livelihood diversification, social capital and REFERENCES resilience to climate variability amongst natural resource de- Adger, W. N., S. Huq, K. Brown, D. Conway, and M. Hulme, pendent societies in Uganda. Ph.D. thesis, School of Envi- 2003: Adaptation to climate change in the developing world. ronmental Sciences, University of East Anglia, 421 pp. Prog. Dev. Stud., 3, 179–195. Grove, A. T., 1996: African river discharges and lake levels in the Ardoin-Bardin, S., 2004: Hydroclimate variability and impacts on twentieth century. The Limnology, Climatology and Paleocli- water resources of large hydrological catchments in the matology of the East African Lakes, T. C Johnson and E. O. Sudanese–Sahelian area (in French). Ph.D. thesis, University Odada, Eds., Gordon and Breach, 95–100. of Montpellier II, 437 pp. Hamandawana, H., 2007: The desiccation of southern Africa’s Ba, M. B., and S. E. Nicholson, 1998: Analysis of convective ac- Okavango Delta: Periodic fluctuation or long-term trend? tivity and its relationship to the rainfall over the Rift Valley Past Global Changes, 15, 12–13. lakes of East Africa during 1983–90 using the Meteosat in- ——, F. Eckardt, and R. Chanda, 2005: Linking archival and re- frared channel. J. Appl. Meteor., 37, 1250–1264. motely sensed data for long-term environmental monitoring. Bader, J. C., 1990: Homogenization and completion of hydromet- Int. J. Appl. Earth Obser. Geoinf., 7, 284–298. ric data in Senegal catchment upstream of Bakel (in French). ——, R. Chanda, and F. Eckardt, 2007: The role of human factors ORSTOM Rep., 18 pp. in the degradation of natural resources in and around the ——, 1992: Study of Manantali’s dam impacts on hydrological Okavango Delta. Int. J. Environ. Stud., 64, 589–605. regime of the Senegal river at Bakel (in French). ORSTOM Hulme, M., 1992: Rainfall changes in Africa: 1931–1960 to 1961– Rep., 56 pp. 1990. Int. J. Climatol., 12, 685–699. Birkett, C. M., 2000: Synergistic remote sensing of Lake Chad: ——, 1996: Climate change within the period of meteorological Variability of basin inundation. Remote Sens. Environ., 72, records. The Physical Geography of Africa, W. M. Adams, 218–236. A. S. Goudie, and A. R. Orme, Eds., Oxford University Boyer, J. F., C. Dieulin, N. Rouche, A. Cres, E. Servat, J. E. Pa- Press, 88–102. turel, and G. Mahe´, 2006: SIEREM: An environmental in- ——, R. Doherty, T. Ngara, M. New, and D. Lister, 2001: African formation system for water resources. Climate Variability and climate change: 1900–2100. Climate Res., 17, 145–168. Change—Hydrological Impacts, S. Demuth et al., Eds., IAHS Hurst, H. E., 1933: The Nile Basin. Vol. IV and Supplements 1–13, Publication 308, IAHS/AISH, 19–25. Ten-Day Mean and Monthly Mean Discharges of the Nile and Bricquet, J. P., F. Bamba, G. Mahe´, M. Toure, and J. C. Olivry, its Tributaries, Government Press, 291 pp. 1997: Water resource variations of the Atlantic river basins of Janicot, S., 1992: Spatiotemporal variability of West African rain- Africa: The long-term effects of rain shortage (in French). fall. Part I: Regionalization and typings. J. Climate, 5, 489– Revue Sci. Eau, 3, 321–337. 497. Conway, D., 2000: The climate and hydrology of the Upper Blue Lamb, P., 1982: Persistance of subsaharan drought. Nature, 299, Nile River. Geogr. J., 166, 49–62. 46–47. ——, 2002: Extreme rainfall events and lake level changes in East Lankford, B. A., and T. Beale, 2007: Equilibrium and non- Africa: Recent events and historical precedents. The East equilibrium theories of sustainable water resources management:

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC 58 JOURNAL OF HYDROMETEOROLOGY VOLUME 10

Dynamic river basin and irrigation behaviour in Tanzania. Mitchell, T. D., and P. D. Jones, 2005: An improved method of Global Environ. Change, 17, 168–180. constructing a database of monthly climate observations and Laraque, A., G. Mahe´, D. Orange, and B. Marieu, 2001: Spa- associated high-resolution grids. Int. J. Climatol., 25, 693– tiotemporal variations in hydrological regimes within Central 712. Africa during the XXth century. J. Hydrol., 245, 104–117. Moniod, F., B. Pouyuad, and P. Sechet, Eds, 1977: The Catchment Le Barbe´, L., and T. Lebel, 1997: Rainfall climatology of the of the Volta River (in French). Hydrological Monogr., No. 5, HAPEX-Sahel region during the years 1950–1990. J. Hydrol., ORSTOM, 513 pp. 188–189, 43–73. Nemec, J., and J. Schaake, 1982: Sensitivity of water resource Leblanc, M., G. Favreau, S. Massuel, S. Tweed, M. Loireau, and B. systems to climate variation. Hydrol. Sci. J., 27, 327–343. Cappelaere, 2008: Land clearance and hydrological change in New, M., M. C. Todd, M. Hulme, and P. Jones, 2001: Precipitation the Sahel: SW Niger. Global Planet. Change, 61, 135–150. measurements and trends in the twentieth century. Int. J. Leduc, C., G. Favreau, and P. Schroeter, 2001: Long-term rise in Climatol., 21, 1889–1922. a Sahelian water-table: The continental terminal in south- Nicholson, S. E., 1983: Sub-Saharan rainfall in the years 1976–80: west Niger. J. Hydrol., 243, 43–54. Evidence of continued drought. Mon. Wea. Rev., 111, 1646– Lemoalle, J., 2004: Lake Chad: A changing environment. Dying 1654. and Dead Seas: Climatic Versus Anthropic Causes, J. C. J. ——, 1996: Environmental change within the historical period. Nihoul, P. O. Zavialov, and P. P. Micklin, Eds., NATO Sci- The Physical Geography of Africa, W. M. Adams, A. S. ence Series, Vol. 36, Kluwer, 321–340. Goudie, and A. R. Orme, Eds., Oxford University Press, 60– L’Hoˆ te, Y., G. Mahe´, B. Some, and J. P. Triboulet, 2002: Analysis 87. of a Sahelian annual rainfall index updated from 1896 to ——, 1998: Historical fluctuations of Lake Victoria and other lakes 2000: The drought continues. Hydrol. Sci. J., 47, 563–572. in the northern Rift Valley of East Africa. Environmental Li, K. Y., M. T. Coe, and N. Ramankutty, 2005: An investigation Change and Response in East African Lakes, J. T. Lehman, of hydrological variability in West Africa using land surface Ed., Kluwer, 7–35. models. J. Climate, 18, 3173–3188. ——, 1999: Historical and modern fluctuations of lakes Tan- ——, ——,——, and R. De Jong, 2007: Modeling the hydrological ganyika and Rukwa and their relationship to rainfall variabil- impact of land-use change in West Africa. J. Hydrol., 337, ity. Climatic Change, 41, 53–71. 258–268. ——, 2005: On the question of the ‘‘recovery’’ of the rains in the Lørup, J. K., J. C. Refsgaard, and D. Mazvimavi, 1998: Assessing West African Sahel. J. Arid Environ., 63, 615–641. the effect of land use change on catchment runoff by com- Ogutunde, P. G., J. Friesen, N. van de Giesen, and H. H. G. bined use of statistical tests and hydrological modelling: Case Savenije, 2006: Hydroclimatology of the Volta River Basin in studies from Zimbabwe. J. Hydrol., 205, 147–163. West Africa: Trends and variability from 1901 to 2002. Phys. Lube`s-Niel, H., J. E. Paturel, J. F. Boyer, and E. Servat, 1998a: Chem. Earth, 31, 1180–1188. Khronostat: Statistical analyses software of chronological se- Olivry, J. C., J. P. Bricquet, and G. Mahe´, 1998: Water Resources ries (in French), version 1.0. IRD-ORSTOM. Variability in Africa during the XXth Century. IASH Publi- ——, J. M. Masson, J. E. Paturel, and E. Servat, 1998b: Variablite cation 252, IAHS/AISH, 189–197. climatique et statistique. Etude par simulation de la puis- Orange, D., A. Wesselink, G. Mahe´, and C. Feizoure, 1997: The sance et de la robustesse de quelques tests utilises pour effects of climate changes on river baseflow and aquifer stor- verifier l’homogeneı¨te de chroniques. Revue Sci. Eau, 11, age in Central Africa. Sustainability of Water Resources under 383–408. Increasing Uncertainty, E. Servat et al., Eds., IAHS Publica- Magistro, J., and M. Lo, 2001: Historical and human dimensions of tion 240, IAHS, 113–123. climate variability and water resource constraint in the Sene- Paturel, J. E., E. Servat, B. Kouame, H. Lubes, M. Ouedraogo, gal River Valley. Climate Res., 19, 133–147. and J. M. Masson, 1997: Climatic variability in humid Africa Mahe´, G., and J. C. Olivry, 1999: Assessment of freshwater yields along the Gulf of Guinea. Part II: An integrated regional to the ocean along the intertropical Atlantic coast of Africa. approach. J. Hydrol., 191, 16–36. C. R. Acad. Sci., 328, 621–626. ——, ——, M. O. Delattre, and H. Lubes-Niel, 1998: Analysis of ——, ——, R. Dessouassi, D. Orange, F. Bamba, and E. Servat, rainfall long series in non-Sahelian West and Central Africa 2000: Surface water and groundwater relationships in a trop- within the context of climate variability (in French). Hydrol. ical river of Mali (in French). C. R. Acad. Sci., 330, 689–692. Sci. J., 43, 937–946. ——, Y. L’Hoˆ te, J. C. Olivry, and G. Wotling, 2001: Trends and Peel, M. C., T. A. McMahon, B. L. Finlayson, and F. G. R. Watson, discontinuities in regional rainfall of West and Central Af- 2001: Identification and explanation of continental dif- rica: 1951–1989. Hydrol. Sci. J., 46, 211–226. ferences in the variability of annual runoff. J. Hydrol., 250, ——, J. E. Paturel, E. Servat, D. Conway, and A. Dezetter, 2005: 224–240. Impact of land use change on soil water holding capacity and ——, ——, and ——, 2004: Continental differences in the vari- river modelling of the Nakambe River in Burkina-Faso. J. ability of annual runoff—Update and reassessment. J. Hy- Hydrol., 300, 33–43. drol., 295, 185–197. McMahon, T. A., M. C. Peel, R. M. Vogel, and G. G. S. Pegram, Piper, B. S., D. T. Plinston, and J. V. Sutcliffe, 1986: The water 2007: Global streamflows—Part 3: Country and climate zone balance of Lake Victoria. Hydrol. Sci. J., 31, 25–37. characteristics. J. Hydrol., 347, 272–291. Richard, Y., S. Trzaska, P. Roucou, and M. Rouault, 2000: Modi- Merryfield, W., 2006: Changes to ENSO under CO2 doubling in a fication of the southern African rainfall variability/ENSO re- multimodel ensemble. J. Climate, 19, 4009–4027. lationship since the late 1960s. Climate Dyn., 16, 883–895. Milly, P. C. D., K. A. Dune, and A. V. Vecchia, 2005: Global pattern Saji, N. H., B. N. Boswami, P. N. Vinayachandran, and T. Yama- of trends in streamflow and water availability in a changing gata, 1999: A dipole made in the tropical Indian Ocean. Na- climate. Nature, 438, 347–350, doi:10.1038/nature04312. ture, 401, 360–363.

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC FEBRUARY 2009 C O N W A Y E T A L . 59

Sarch, M.-T., and E. H. Allison, 2000: Fluctuating fisheries in Af- Troy, B., C. Sarron, J. M. Fritch, and D. Rollin, 2007: Assessment rica’s inland waters: Well-adapted livelihoods, maladapted of the impacts of land use changes on the hydrological regime management. Proc. 10th Int. Conf. of the Institute of Fisheries of a small rural catchment in South Africa. Phys. Chem. Economics and Trade, Corvallis, OR, U.S. National Marine Earth, 32, 984–994. Fisheries Service. [Available online at http://osu.orst.edu/ UNESCO, 1995: Discharge of Selected Rivers of Africa (in dept/IIFET/2000/papers/sarch.pdf.] French). Studies and Reports in Hydrology, Vol. 52, Savenije, H. H. G., 1996: Does moisture feedback affect rainfall UNESCO, 166 pp. significantly? Phys. Chem. Earth, 20, 507–513. Vo¨ ro¨ smarty, C. J., and D. Sahagian, 2000: Anthropogenic distur- Sene, K. J., and D. T. Plinston, 1994: A review and update of the bance of the terrestrial water cycle. Bioscience, 50, 753–765. hydrology of Lake Victoria. Hydrol. Sci. J., 39, 47–63. ——, and Coauthors, 2001: Global water data: A newly endan- Sircoulon, J., 1976: Hydropluviometric data for the recent drought gered species. Eos, Trans. Amer. Geophys. Union, 82, 54. in intertropical Africa. Comparison with the droughts of 1914 ——, E. M. Douglas, P. A. Green, and C. Revenga, 2005: Geo- and 1940 (in French). Cah. ORSTOM, Ser. Hydrol., 13, 75– spatial indicators of emerging water stress: An application to 174. Africa. Ambio, 34, 230–236. ——, 1985: The drought in West Africa. Comparison between the Washington, R., and Coauthors, 2006: African climate change: years 1982–1984 and the years 1972–1973 (in French). Cah. Taking the shorter route. Bull. Amer. Meteor. Soc., 87, 1355– ORSTOM, Ser. Hydrol., 21, 75–86. 1366. Sutcliffe, J. V., and D. G. Knott, 1987: Historical variations in Webster, P. J., A. M. Moore, J. P. Loschnigg, and R. R. Lebden, African water resources. The Influence of Climate Change 1999: Coupled ocean–atmosphere dynamics in the Indian and Climatic Variability on the Hydrologic Regime and Water Ocean during 1997–98. Nature, 401, 356–360. Resources, S. I. Solomon, M. Beran and W. Hogg, Eds., Woyessa, Y. E., E. Pretorius, P. S. van Heerden, M. Hensley, and IAHS Publication 168, IAHS/AISH, 463–476. L. D. van Rensburg, 2006: Impact of land use on river basin ——, and Y. P. Parks, 1999: The Hydrology of the Nile. IAHS water balance: A case study of the Modder River Basin, Special Publication, Vol. 5, IAHS, 179 pp. South Africa. Comprehensive Assessment of Water Manage- ——, and G. Petersen, 2007: Lake Victoria: Derivation of a cor- ment in Agriculture Research Rep. 12, 37 pp. [Available on- rected natural water level series. Hydrol. Sci. J., 52, 1316– line at http://www.iwmi.cgiar.org/Assessment/files_new/ 1321. publications/CA%20Research%20Reports/CARR12- Tarhule, A., 2005: Damaging rainfall and flooding: The other Sa- low.pdf.] hel hazards. Climatic Change, 72, 355–377. Zhang, X., F. W. Zwiers, G. C. Hegerl, F. H. Lambert, N. P. Tate, E., J. Sutcliffe, D. Conway, and F. Farquharson, 2004: Water Gillett, S. Solomon, P. A. Stott, and T. Nozawa, 2007: Detec- balance of Lake Victoria: Update to 2000 and climate change tion of human influence on twentieth-century precipitation modelling to 2100. Hydrol. Sci. J., 49, 563–574. trends. Nature, 448, 461–465, doi:10.1038/nature06025.

Unauthenticated | Downloaded 10/01/21 03:31 AM UTC