ENSO AND INTERANNUAL RAINFALL VARIABILITY IN UGANDA: IMPLICATIONS FOR AGRICULTURAL MANAGEMENT

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 20: (2000) ENSO AND INTERANNUAL RAINFALL VARIABILITY IN UGANDA: IMPLICATIONS FOR AGRICULTURAL MANAGEMENT JENNIFER PHILLIPS a, * and BEVERLY McINTYRE b a Center for Climate Systems Research, Columbia Uni ersity, Columbia, NY, USA b Rockefeller Foundation, Kawanda Agricultural Research Institute, Kampala, Uganda Recei ed 12 No ember 1998 Re ised 8 June 1999 Accepted 11 June 1999 ABSTRACT The El Niño Southern Oscillation (ENSO) is correlated with the short rainy season (September December) in East Africa. Seasonal climate forecasts made on this basis are being disseminated in the hope that this information will be useful in regional or even local planning and resource management. In order to improve the likelihood of success in using regional forecasts in Uganda, particularly in the agricultural sector, climate analysis was performed at the sub-regional level with distinctions being made between unimodal (short season peak in August) and bimodal (short season peak in November) rainfall zones. Monthly climate data from 1931 to 1960 were available for 33 sites. Averaging across all stations, it is shown that Pacific Ocean NINO3 region sea-surface temperatures (SSTs) from July to September (JAS) are significantly correlated with both the concurrent August September (AS) rainfall (r= 0.75) and the following November December (ND) rainfall (r=0.57), but with opposite signs. When station data are separated into uni- and bimodal sites, it becomes clear that the importance of ENSO events is different in the two zones. In the unimodal zone, El Niño events are associated with a depression of the August peak in rainfall, but a lengthening of the season, potentially providing an opportunity for growing later-maturing crops. At bimodal sites, there is very little impact in August but November rainfall is enhanced in El Niño years and depressed in La Niña years. Given a forecast of ENSO, the primary strategies that will be useful in farm management will differ by rainfall zone and will revolve around the choice of crop or cultivar and the timing of planting in order to make optimal use of the growing period. Copyright 2000 Royal Meteorological Society. KEY WORDS: East Africa; Uganda; forecasts; rainfall variability; precipitation; agricultural applications 1. INTRODUCTION The period from mid-1997 to early 1998 marked the most extreme El Niño event of the century and wide media attention raised awareness of this Pacific Ocean phenomenon and its potential influence on the global climate. For decades, climate scientists have been studying the relationships between ocean-surface temperatures and large-scale climate events, and are now at the point of issuing seasonal climate forecasts for a number of regions around the world based in large part on trends in the sea-surface temperature (SST) ( asmt/). Efforts to disseminate these seasonal forecasts are founded on the assumption that they can be useful at the regional level for food security and water resources planning, and at the individual farm level for planning agronomic activities (Phillips et al., 1998). Within the East Africa region, however, El Niño Southern Oscillation (ENSO) responses vary from one location to another and, in many cases, details about the implications for sub-regional climate patterns are lacking. Hence, planners face some uncertainty when using information that is presented on a regional scale. This paper focuses on the sub-regional significance of ENSO events in Uganda, distinguishing between unimodal and bimodal rainfall zones. The authors attempt to clarify the implications of the existing literature and new data analysis with regard to the potential influence of ENSO * Correspondence to: NASA GISS, 2880 Broadway, New York, NY 10025, USA. Copyright 2000 Royal Meteorological Society

2 172 J. PHILLIPS AND B. McINTYRE events on seasonal rainfall in these zones, in order to enhance knowledge that is critical to agricultural and food security planning in Uganda El Niño Southern Oscillation (ENSO) ENSO refers to both SST and sea-level pressure fluctuations in the Pacific Ocean. Oscillations in atmospheric pressure between Darwin, Australia, and Tahiti were first documented by Sir Gilbert Walker (Walker and Bliss, 1932) during his investigation into the failure of the 1898 Indian Monsoons, a weather event that led to massive crop failures and famine. Research over the last 70 years has resulted in a much greater understanding of the relationship between ocean temperature and global atmospheric circulation. In the late 1980s, global maps were produced documenting the relationship between Pacific SSTs and both rainfall and temperature patterns (Ropelewski and Halpert, 1987; Kiladis and Diaz, 1989). Concurrent advances in the ability to predict ocean temperatures (Cane et al., 1986; Zebiak and Cane, 1987) mean that forecasts of SSTs can now potentially be translated into climate forecasts (Latif et al., 1994). El Niño events, in which SSTs in the eastern equatorial Pacific Ocean are anomalously warm, occur on average one to two times a decade, although the 1990s have seen an unusual string of events which some suggest is related to climate change (Trenberth and Hoar, 1996; see also Rajagopalan et al., 1997). When SSTs are anomalously cool, the event is called a La Niña. The Southern Oscillation Index (SOI), a measure of sea-level pressure gradients, is inversely related to the temperature anomalies in the equatorial Pacific. Global responses to ENSO events vary, depending on the location, and can involve either more or less precipitation, and/or warmer or cooler temperatures. After the seasonal cycle, ENSO-related climate variation is now considered the largest global periodic climate signal (Hastenrath, 1995). Its influence is particularly strong in Australia, India, the west coasts of North and South America, coastal Brazil, and southern Africa. Influences on climate have also been discerned in Mexico, the American Midand Southwest, and East Africa ENSO and East African climate The dominant rainfall pattern over East Africa is related to the movement of the Inter-tropical Convergence Zone (ITCZ), which brings rainfall approximately 1 month after the sun s path and the plane of the equator coincide. Close to the equator, with the sun passing overhead twice yearly, the result is a bimodal pattern in rainfall, with the primary rainy season occurring in March May (MAM), and a shorter secondary rainy season occurring in October December (OND). This moisture generally comes with the northeasterly winds originating in the Indian Ocean (Ogallo, 1988; Mutai et al., 1998). Further north, including the northern edge of Uganda, the short rains tend to peak earlier, in August. Sites with August rains receive this moisture from the Congo/Zaire Basin to the southwest (Ogallo, 1988). Jameson and McCallum (1970, p. 14) note that, in northern Uganda, the period between the end of the first rains and the beginning of the second rains is shorter than would be expected from the zenith of the sun. Hence, the two rains are sufficiently close together to constitute a single rainy season for all practical purposes. Some sites are truly unimodal in rainfall distribution, with little precipitation during the rest of the year. Extremes of both rainfall types and an intermediate site in Uganda are illustrated in Figure 1. Furthermore, both the complex topography and the presence of large bodies of water (such as Lake Victoria) tend to influence rainfall patterns, resulting in a high degree of spatial heterogeneity in regional climate (Ogallo, 1989). In the work of Indeje et al. (2000), three of the eight homogeneous rainfall zones found for East Africa through combined Empirical Orthogonal Function (EOF) and simple correlation analysis fall over Uganda: Zone IV to the northeast, Zone VII to the southwest, and Zone VI which is dominated by Lake Victoria. The majority of the work relating ocean temperature and sea-level pressure to East African climate has focused on the areas with a November peak in the short rainy season (Farmer, 1988; Hutchinson, 1992; Mutai et al., 1998). There is agreement in the literature that there is no significant relationship between ENSO and the MAM rains, which generally show low interannual variability (Mutai et al., 1998).

3 ENSO AND UGANDAN RAINFALL 173 However, a number of studies have demonstrated a significant relationship between ENSO and OND rainfall. Ogallo (1988) found peak lag zero correlation coefficients, r, of 0.6 between the SOI and OND rainfall along the coast of southern Kenya and northern Tanzania. Similar values relate SOI and SOND (September December) precipitation along the Kenyan coast (Farmer, 1988), whereas higher values (r= 0.8) describe the relationship of SOI to the Der rainfall season (OND) in Somalia (Hutchinson, 1992). Beltrando and Camberlin (1993) analysed the Indian Ocean temperature and pressure (which vary with Pacific Ocean values) and rainfall in three zones along the east coast of Africa. The strongest relationships were found between September and October northern Indian Ocean temperatures and OND rainfall in southern Somalia, and were significant at the 95% level of confidence. Recent work by Mutai Figure 1. Monthly rainfall distribution for three representative sites in Uganda ( ). (a) Kitgum, strongly unimodal; (b) Namasagali, intermediate; (c) Busheyni, bimodal

4 174 J. PHILLIPS AND B. McINTYRE et al. (1998), using SSTs from the Pacific, Indian and Atlantic Oceans in a multiple linear regression prediction scheme for the east African short rains, found that variability in equatorial Pacific SSTs is the lead predictor, with equatorial Indian Ocean and southern Atlantic Ocean surface temperatures playing a smaller, although significant, role. Multiple r values ranged from 0.56 to 0.78, depending on the time period tested, in regressions with JAS SSTs predicting OND rainfall. Analyses of SST rainfall relationships as a function of rainfall modality (rather than large-scale regional grouping) have resulted in significant relationships for sites with an August peak in rainfall. Ogallo (1988) found that the relationships between SOI and JAS rainfall were strongest over Uganda and western Tanzania, and were opposite in sign from that relating SOI to regional OND rainfall. Beltrando and Camberlin (1993) analysed data from a bimodal rainfall site (Baidoa, southern Somalia) and a unimodal site with an August peak further north (Addis Ababa, Ethiopia). They found the short season rainfall in Baidoa (October) to be positively correlated with Indian Ocean temperatures (r=0.51), while in Addis Ababa, September rainfall is negatively correlated with SST in the northern reaches of the Indian Ocean at a zero lag (r= 0.44). Based on these analyses, we hypothesized that the influence of ENSO on rainfall patterns in Uganda would differ significantly between the north, which experiences an August maximum in rainfall, and the south, where rainfall is more bimodally distributed and the secondary rainy season peaks in November Common cropping patterns in uni- and bimodal rainfall zones of Uganda Because rainfall patterns differ from the north to the south in Uganda, so do dominant crops and cropping systems. In the north, where rainfall is almost unimodal, annual crops such as millet and other grains are prominent. Late-maturing varieties of sorghum, sesame and pulses are also planted, as are cassava, legumes and other vegetables. Grains are usually planted first in March and then again in August or September, so that physiological maturity will occur in the dry season. Perennial crops, such as banana and coffee, dominate most of the cropping systems of the bimodal rainfall zone, although locally distinct cropping systems (e.g. grain in southwestern and eastern highlands) do exist. Early-maturing grain crops (e.g. 120 days to maize dry harvest) and pulses are grown in both rainy seasons. Other commonly planted crops include sweet potato, potato, cassava and legumes. Planting dates for annuals vary with the onset of the rains. Depending on soil moisture, maize may be planted from mid-august to mid-september in the first season and mid-february to mid-march in the second season. Beans are often planted well into April and October. Given the dependence of planting time and crop choice on rainfall distribution, there could be potential for utilizing forecasts of season arrival date and duration in crop management. 2. DATA AND METHODS Monthly mean rainfall records ( ) for a network of 33 stations in Uganda (East African Meteorology Department, 1965) were used in this study (Table I, Figure 2). This period was chosen because of the relative completeness of the data set. Fifteen of the 33 sites had at least 1 month, but no more than 8 months, of AS or ND rainfall data missing from the 30-year record. A total of 1.5% of all months used in the analysis was missing. For area averages, no attempt was made to fill in the missing data. The Pacific SST anomalies (SSTA) used were from the analysis of Kaplan et al. (1998), and limited to the NINO3 region of the Pacific (5.0 N 5.0 S, 150 W 90 W). The Kaplan et al. (1998) analysis extends from January 1856 to December Averages of the November February (NDJF) SSTA were used to reflect the standard peak period of ENSO activity. Averages of the JAS SSTA were also tested, because of evidence that this period is a strong predictor of rainfall in the region (Mutai et al., 1998) and because such predictions do not rely on forecasts of SST. ENSO events were defined as years in which the average NDJF anomaly was greater than 1 S.D. from the long-term mean. For the time period between 1931 and

5 ENSO AND UGANDAN RAINFALL 175 Table I. Location of stations used in study Station Number (map) Altitude above Latitude/longitude sea level (m) Arua E Bikira E Bubondo E Bukalasa E Bunyaruguru E Bushenyi E Butiaba E Buunga E Entebbe E Ft. Portal E Gulu E Hoima E Kabale E Kalangala E Katakwi E Katera E Katigondo E Kitgum E Kivuvu E Lira E Lwasamaire E Masaka E Masindi E Mbale E Mbarara E Moniko E Moroto E Moyo E Nagoje E Namasugali E Ngora E Serere E Tororo E 1960, 5 years are defined as El Niños (1940, 1941, 1951, 1953, 1957) and 7 years are defined as La Niñas (1933, 1938, 1942, 1949, 1950, 1954, 1955), based on the NDJF SST anomaly. Note that during this 30 year period La Niña events outnumbered El Niño events, which contrasts with the more recent period. 3. RESULTS 3.1. All-station a erage Separating out years in the time series by ENSO category, based on the NDJF SSTA, mean monthly precipitation for an all-station average was plotted and compared with the climatology of all years (Figure 3). Because this grouping combines both the uni- and the bimodal rainfall sites, the peak short season rainfall appears as strongly shifted from a November peak in El Niño years to an August peak in La Niña years, with little change in the total annual rainfall. The long rains in MAM are unaffected by ENSO phase. The negative relationship between ENSO and AS rainfall, and the positive relationship with ND rainfall are illustrated in a scatter plot of seasonal rainfall versus SST anomalies (Figure 4). Linear regressions confirm these relationships (Table II). Correlation coefficients from the test between NDJF SSTA and AS and ND rainfall are significant at the 99% level of confidence. With JAS SSTA as

6 176 J. PHILLIPS AND B. McINTYRE Figure 2. Map of Uganda with sites used in study marked (see Table I) the predictor of AS or ND rainfall, the relationships are stronger yet, and indicate that more than half of the interannual variability in AS rainfall in Uganda is explained by NINO3 SSTs. Testing the SOND rainfall mean, r drops to zero using NDJF SSTAs as the predictor, and 0.32 using JAS SSTAs, as a result of averaging across the two rainfall periods that respond inversely to each other to SST forcings. This highlights the importance of separating the uni- and bimodal rainfall zones in the analysis Regional impacts Observations of the apparent unimodality of the northern rains suggested that the temporal shift in peak rainfall season noted in Figure 3 could be a methodological artefact of averaging rainfall across sites with either uni- or bimodal rainfall. All 33 sites were divided into those with a typical bimodal rainfall distribution (OND/MAM) and those in which the short rains peak in August and tend toward a unimodal distribution (Figure 5). This regional separation follows closely that found by Indeje et al. (2000), in cluster analysis. Figure 6(a and b) represents the seasonal influence of ENSO in the two rainfall zones. The positive impact of El Niño events on ND rainfall at the bimodal rainfall sites can be seen in Figure 6(a). Tests of the linear correlation between NINO3 SSTAs and ND rainfall show the relationship to be significant (Table III). Although similar tests show SSTAs to be significantly correlated with AS rainfall as well, Figure 3. Monthly rainfall distribution by ENSO phase and for all years for 33 sites in Uganda ( ). See Data and Methods section for definition of ENSO events

7 ENSO AND UGANDAN RAINFALL 177 Figure 4. Mean NDJF NINO3 sea-surface temperature (SST) anomaly versus mean monthly precipitation in AS ( ; slope: 20.64, intercept: 115.1) and ND ( ; slope: 25.51, intercept: 88.0) for all sites in Uganda ( ) F-tests of the difference between AS rainfall in El Niño years and that in La Niña years indicate that changes in August rainfall are less distinct between ENSO phases than those in November (Table IV). In the zone with unimodal rainfall (Figure 6(b)), the strong peak in August rainfall seen in La Niña years is extended over a longer period in El Niño years, and substantial rainfall continues through November. Although the shift in peak rainfall seen in the all-station average monthly distribution is relevant for both zones, the implications and usefulness of forecast information depend on site location. 4. REGIONAL IMPLICATIONS FOR CROP MANAGEMENT 4.1. Bimodal zone In the bimodal zone, where the short season maximum rainfall generally occurs in November December (ND), El Niño events strengthen this pattern. In La Niña years, maximum rainfall is shifted up to August October and the total precipitation is less than normal. In an El Niño year, early (July) Table II. Correlations between monthly precipitation across all 33 sites in Uganda and SST anomalies in the NINO3 region at various lags ( ) Rainfall period NDJF JAS r r 2 r r 2 MAM SOND AS ** ** ND ** ** ** Significance at the 99% level.

8 178 J. PHILLIPS AND B. McINTYRE Figure 5. Map of Uganda with bimodal ( ) and unimodal ( ) sites indicated sowing of a short-duration crop, such as beans, and a late (September) sowing of late-maturing grains could be a potentially successful cropping strategy, given the abundance of soil moisture. Higher inputs, as well as increased plant density, would enable farmers to take advantage of the strong rains in order to enhance productivity. A La Niña event would indicate that farmers should shift the sowing of grains to earlier in August or perhaps to even dry-sow in July. Since less precipitation falls during the short season of a La Niña year, early-maturing varieties and shorter-duration crops would be more likely to succeed Unimodal zone In the northern areas of Uganda where rainfall peaks in August, the potential for a strong August maximum and a steep decline in precipitation to the January minimum are enhanced in La Niña years. In contrast, rainfall in El Niño years is depressed in August, although rain generally continues into November. The key importance of an ENSO forecast would be in crop selection, and timing of planting and fertilizer applications. For example, in an El Niño year, late-maturing varieties will have a strong advantage over other maturity classes. An August sowing of a late-maturing maize would ensure sufficient rainfall in October November to avoid peri-anthesis water stress and ensure that maturity would not occur prior to the subsidence of the rains in January. Forecasts would also enable farmers to select crops better adapted to the protracted rains. For example, maize would have a distinct advantage over legumes in terms of resistance to disease and enhanced productivity during longer rains. Farmers could take advantage of the greater productivity potential of late-maturing cultivars and longer-duration crops by increasing inputs. InaLaNiña year, early-maturing varieties sown in July will have a better chance of success. Fertilizer should be applied when soil moisture is sufficient to promote plant uptake of nutrients but is not so high as to result in leaching. Nitrogen applications should be minimal or avoided immediately prior to the expected August maximum in order to decrease losses to leaching and run-off. 5. DISCUSSION AND CONCLUSIONS In order for local level decision-makers to make use of seasonal ENSO forecasts, climate information should be tailored at least to the sub-regional scale. By separating out uni- and bimodal rainfall sites in Uganda, the implications of ENSO events in these two zones have been distinguished. For Uganda as a whole, Pacific SSTs are positively correlated with ND rainfall, as has been shown by other work, and

9 ENSO AND UGANDAN RAINFALL 179 negatively correlated with AS rainfall (Table II). However, the rain falling in AS is spatially, as well as functionally, separate from that falling in ND. In the northern, quasi-unimodal rainfall zone, where AS rainfall is relied on for a second crop, in El Niño events the usual August peak is depressed and rainfall extends into November. This shift in seasonal maximum might be used to extend the cropping season by planting late-maturing, potentially high-yielding crops. Where the November maximum in short season rainfall is used for the second crop in the distinctly bimodal zone, La Niña years tend to show depressed November rainfall and a shift of the peak to roughly 2 months earlier. Thus, forecasts of cold SSTs would indicate early planting of short season crops. Figure 6. Monthly rainfall distributions by rainfall zone and ENSO phase in Uganda: (a) bimodal sites; (b) unimodal sites

10 180 J. PHILLIPS AND B. McINTYRE Table III. Correlations between monthly precipitation averaged by rainfall zone across Uganda and SST anomalies in the NINO3 region at various lags ( ) NDJF JAS r r 2 r r 2 Bimodal sites (n=22) AS ** ** ND * ** Unimodal sites (n=11) AS ** ** ND ** ** * Significance at the 95% level. ** Significance at the 99% level. Although the emphasis in each zone is different, there are similarities. In both cases, the general recommendation in El Niño years would be to plant later-maturing crops and possibly increase inputs to take advantage of higher (bimodal) or later (unimodal) rainfall. In La Niña years, lower rainfall (bimodal) or a shorter season (unimodal) both call for earlier than normal planting and short season varieties. The difference is that, in the northern unimodal zone, the recommended strategy for La Niña years is probably more closely aligned with management under normal conditions. Thus, forecast information is more likely to be of value in El Niño years, where a long season can be taken advantage of. In contrast, in the southern bimodal zone, it is more likely to be La Niña year forecasts that will be of value in management, encouraging early planting of short season varieties. Although the results of F-tests of rainfall by individual months (Table IVb, c) show a significant difference between El Niño and La Niña rainfall only for November at both the uni- and bimodal sites, this translates into a changed expectation for mid-season at the bimodal sites and late season for the unimodal sites. This analysis is based on rainfall records from 1931 to 1960, a period in which Pacific SSTs show somewhat lower variability than those in preceding or later years. It has been suggested that this time period is not necessarily valid as a basis for analysis of current climate patterns (Hulme, 1992). However, Table IV. Mean (S.D.) precipitation by ENSO phase for the critical months of August, September and November, and probability of significantly different values by phase resulting from an F-test: (a) all sites (n=33); (b) bimodal sites (n=21); and (c) unimodal sites (n=12) Month Precipitation (mm) El Niños La Niñas p value (a) All Uganda August 102 (15) 133 (19) 0.31 September 86 (9) 127 (17) 0.31 November 153 (38) 63 (9) 0.01** (b) Bimodal August 80 (13) 102 (20) 0.15 September 86 (29) 111 (18) 0.13 November (c) Unimodal August 162 (36) 140 (28) 77 (9) 188 (23) 0.01** 0.23 September 84 (23) 153 (21) 0.34 November 137 (46) 40 (19) 0.02* El Niño years: n=5; La Niña years: n=7.

11 ENSO AND UGANDAN RAINFALL 181 in tests comparing climate data from 1931 to 1960 and from 1961 to 1990 at a 5 grid over Africa, Hulme (1992) found that most of the differences observed between these two time-periods occur in Saharan and sub-tropical zones, with a minimum of change in interannual variability observed close to the equator. Some indication of a shift in climate was noted, however, by Farmer (1988) in his analysis of the short rainy season along the Kenyan coast which extended from 1901 to He found that the relationship between SOI and interannual rainfall variability was stronger for the period than for based on station data from four sites. In tests of correlations between NINO3 SSTA and AS or ND rainfall at three Ugandan sites for which a complete record from 1931 to 1995 was available (Entebbe, Kabale, and Mbarara), correlations for the full time series do not tend to improve or to worsen systematically when compared with tests restricted to At Entebbe, which is strongly influenced by lake effects from Lake Victoria, r increases from 0.13 to 0.18 when the full 65 years are included in the test, though neither result is significant. At Kabale, a highly significant result (r= 0.45) from the test using data drops to close to zero (r= 0.08) when using the data from all 65 years. The opposite result is found at Mbarara, where an r of 0.23 increases to 0.35, significant at the 95% level of confidence, by testing the complete series. Although these few stations are not enough to base a conclusion on regarding the validity of using the 30-year time series included in our analysis, the implication is that the period from 1931 to 1960 is not greatly different from the present. Furthermore, the general relationships found here between Pacific SSTAs and precipitation in Uganda show good agreement with other work based on long-term, or more recent records (Ogallo, 1988: ; Indeje et al., 2000: ). Testing the relationship at individual stations does point out the large-scale nature of the SST-climate forcing, and the limitations regarding the use of forecasts for agricultural management at the local level. When all stations are tested individually for correlations with Pacific SSTs, there is a wide variation in results (data not shown). The correlation coefficient, r, for the test of AS rainfall and JAS SSTA ranges from zero at Busheny a bimodal site in the south to 0.55 (p value 0.001) at Moyo the most northerly site in our set and strongly unimodal in rainfall distribution. The implications are twofold. First, because forecasts of both SSTs and precipitation are probabilistic, they can serve only to narrow the range of expectations for a season, and management decisions at the local or farm level still involve risk. Second, given the above caution, because there is coherence in climate patterns at a scale which is sub-regional, and for which there are significant relationships with ocean-surface temperatures (see Indeje et al., 2000), forecasts can be downscaled, either through statistical approaches or using mesoscale climate models. The effective use of forecasts would depend on the success of extension workers and non-government organizations in sensitizing farmers to the implications of an ENSO season. Also critical to the success of the implementation of the above mentioned cropping strategies would be the timely provision to farmers of inputs, such as fertilizers and seeds of the requisite maturity classes. A useful next step in applying ENSO-based forecasts to agricultural management would be to assess the criteria used in decision-making under uncertainty in a specific socioeconomic context, given the risks and benefits of using the forecasts. Until such criteria are defined, a note of caution should be included in disseminating forecasts for use at the local level. ACKNOWLEDGEMENTS Thanks are due to F.H.M. Semazzi and M.A. Cane for their useful comments on the manuscript. REFERENCES Beltrando, G. and Camberlin, P Interannual variability of rainfall in the eastern Horn of Africa and indicators of atmospheric circulation, Int. J. Climatol., 13, Cane, M.A., Zebiak, S.E. and Dolan, S Experimental forecasts of El Niño, Nature, 322, East African Meteorology Department Monthly and Annual Rainfall in Uganda during the 30 Years , East African Meteorlogical Department, Government of Uganda, January 1965.

12 182 J. PHILLIPS AND B. McINTYRE Farmer, G Seasonal forecasting of the Kenya coast short rains, , J. Climatol., 8, Hastenrath, S Recent advances in tropical climate prediction, J. Climate, 8, Hulme, M Rainfall changes in Africa: to , Int. J. Climatol., 12, Hutchinson, P The Southern Oscillation and prediction of the Der season rainfall in Somalia, J. Climate, 5, Indeje, M., Semazzi, F.H.M. and Ogallo, L.J ENSO signals in east African rainfall seasons, Int. J. Climatol., 20, Jameson, J.D. and McCallum, D Climate, in Jameson, J.D. (ed.), Agriculture in Uganda, 2nd edn, Oxford University Press, Oxford, pp Kaplan, A., Cane, M.A., Kushnir, Y., Clement, A., Blumenthal, M. and Rajagopalan, B Analyses of global sea-surface temperature , J. Geophys. Res., 103, Kiladis, G.N. and Diaz, H.F Global climate anomalies associated with extremes in the Southern Oscillation, J. Climate, 2, Latif, M., Barnett, T.P., Cane, M.A., Flügel, M., Graham, N.E., Von Storch, H., Xu, J.S. and Zebiak, S A review of ENSO prediction studies, Clim. Dyn., 9, Mutai, C.C., Ward, M.N. and Coleman, A.W Towards the prediction of the East Africa short rains based on sea-surface temperature atmosphere coupling, Int. J. Climatol., 18, Ogallo, L.J Relationships between seasonal rainfall in East Africa and the Southern Oscillation, J. Climatol., 8, Ogallo, L.J., The spatial and temporal patterns of the East African seasonal rainfall derived from principal component analysis, Int. J. Climatol., 9, Phillips, J.G., Cane, M.A. and Rosenzweig, C ENSO, seasonal rainfall patterns and simulated maize yield variability in Zimbabwe, Agric. For. Meteorol., 90, Rajagopalan, B., Lall, U. and Cane, M.A Anomalous ENSO occurrences: an alternate view, J. Climate, 10, Ropelewski, C.F. and Halpert, M.S Global and regional scale precipitation patterns associated with the El Niño Southern Oscillation, Mon. Weather Re., 115, Trenberth, K.E. and Hoar, T.J The El Niño Southern Oscillation event: longest on record, Geophys. Res. Lett., 23, Walker, G.T. and Bliss, E.W World weather V, Mem. R. Soc., 4, Zebiak, E.S. and Cane, M.A A model of El Niño Southern Oscillation, Mon. Weather Re., 115,