Understanding, predicting and mitigating low frequency variations and trends in climate is a challenge facing Africa. While seasonal climate forecasting has seen advances in development and application (Moura and Sarachik, 1997; Stockdale et al., 1998; NOAA, 1999; Washington and Downing, 1999; Jury, 2002a), the multi-scale factors underlying climate change at country level are beginning to be explored (Xue and Shukla, 1998). Our study of trends in climate over northeastern Africa is set in the context of an emerging understanding of how greenhouse gases, aerosols (Santer et al., 1996; Schimel et al., 1996; Haywood et al., 1997; Mitchell and Johns, 1997; Penner et al., 1998; Mitchell and Hulme, 1999; Giorgi and Francisco, 2000) and land cover (Charney, 1975; Cunnington and Rowntree, 1986; Xue et al., 1990; Xue, 1997; Zheng and Eltahir, 1997; Zeng and Eltahir, 1998; Zeng et al., 1999; Wang and Eltahir, 2000) alter the radiation budget. Because of its agricultural basis, Africa is more vulnerable to climate change (Dixon et al., 1996; Benson and Clay, 1998). There economic growth is coupled to environmental conditions (Jury, 2002b), so it is necessary to quantify the background trends.
Model simulations of climate change over Africa conducted before 1998 suggested widespread warming of ∼1 °C (Hulme, 1994a, 1994b; Hernes et al., 1995; Conway et al., 1996; Ringius et al., 1996) and a drying trend in many places (Hudson, 1997; Joubert and Hewitson, 1997; IPCC, 1998; Feddema, 1999) over the period 2000–2050. However that generation of coupled general circulation models (cGCM) had difficulty representing the near-equatorial rainfall maxima and inter-decadal climate variability in the subtropics (Hulme, 1998; Kittel et al., 1998; Boer et al., 2000; Jury et al., 2007). Comparisons between observed- and model-simulated rainfall yielded low correlation (r < 0.4) (Funk and Brown, 2009). Advances in modelling technology mean that the cGCM projections from the IPCC 4th assessment better represent Africa's mean climate and its variability, and orographic forcing over continental highlands (Meehl et al., 2007). Thus projected trends may be used with some confidence, particularly where the projected patterns and rate-of-change are in reasonable agreement with past observations.
Ethiopia is a highland country with a vegetation index > 0.5, surrounded by lowland semi-desert (Figure 1(a)). Nocturnal maximum temperatures estimated by MODIS range from 15 °C over the mountains to 35 °C in the surrounding lowlands (Figure 1(b)), and give the impression of a ‘cool island’. The moist, elevated zones are critical to Ethiopia's agriculture. Cool temperatures slow soil evaporation and crop development; many high yielding varieties of maize, teff, barley and sorghum require ∼180 d to maturation. These ‘long cycle’ crops provide nearly 50% of Ethiopia's grain production. Our research follows on from Hulme (1996a, 1993b) and Hulme et al. (2001) and is motivated by declining rainfall over the southern half of Ethiopia (Seleshi and Zanke, 2004; Seleshi and Camberlin, 2006) and a tendency for rain deficits in both planting (March to May) and summer season (June to September) (Funk et al., 2005; Verdin et al., 2005). While surface water run-off exceeds 30 km3 year−1 over the western highlands according to Princeton hydrological (VIC) reanalysis (Figure 1(c)), the demand for water-fed resources is related to population density (Figure 1(d)) which is highest north of the Rift Valley that cuts across the country. Despite locally heavy rainfall, evaluations of water availability and use indicate that hydrologic resources may soon constrain development. As population increases and agricultural expansion stalls, Ethiopia may face increasing food shortages (Funk and Brown, 2009).
Focusing on Ethiopia, this study evaluates regional climatic trends and large scale circulation features in cGCM and reanalysis records. In Section '2. Data and methods', the data and methods are given, while Section '3. Results' covers the results in terms of observed and projected trends at local and hemispheric scale. In Section '4. Conclusions', we describe some of the drivers of Ethiopian climate change and discuss potential impacts.
2. Data and methods
To study the observed climatic trends in northeastern Africa we employ a dense set of monthly temperature and rain gauge data that combines measurements from the Global Historical Climate Network (Peterson and Easterling, 1994), the Food and Agriculture Organization and the African National Meteorology Services (Figure 2). These data are interpolated to 0.1° resolution using a kriging technique and termed the ‘famine early warning system network’ (FEWS) (Funk et al., 2005; Verdin et al., 2005; Funk and Verdin 2009). We also consider gridded monthly data sets of the Climate Research Unit (CRU) of the University of East Anglia (Mitchell and Jones, 2005) and the Global Precipitation Climatology Center (Rudolf and Schneider, 2005). These and many other data were extracted from the International Research Institute Climate Library website (IRI, 2000), as interpolated from the station network (Hulme et al., 2001). The density for CRU temperature and rainfall observations is given in Figure 2(a) and (b); and for the FEWS network in Figure 2(c) and (d). Surface measurements are relatively scarce before World War 2 (Conway et al., 2004), so our main analysis commences thereafter. To define the spatial pattern of trends in the atmosphere, National Center for Environmental Prediction (NCEP) meteorological reanalysis products at 2° resolution in the period 1948–2006 (Kalnay et al., 1996) are analysed as maps and vertical sections for temperature, specific humidity, zonal and meridional wind and vertical motion. Local hydrological trends are studied using the Princeton VIC model reanalysis (Sheffield et al., 2006). Trends of surface temperature and rainfall over Ethiopia may be affected by global ocean–atmosphere coupling (Lamb, 1978; Folland et al., 1986, 1991; Lough, 1986; Palmer, 1986; Glantz et al., 1991; Camberlin, 1995, 1997; Semazzi et al., 1996; Nicholson, 1998; Shanko and Camberlin, 1998; Birkett et al., 1999; Tourre et al., 1999; Webster et al., 1999; Nicholson and Grist, 2001; Xie et al., 2002; Yeshanew and Jury, 2007; Segele et al., 2009a). Thus a wider context is sought through analysis of Global Precipitation Climatology Project (GPCP) rainfall (Adler et al., 2003) and NCEP surface air temperature and wind fields. The evolving nature of climatic trends is analysed via hovmoller plots of 5-year smoothed rainfall and temperature anomalies averaged over the 5–15N latitude band across Africa from 1948 to 2006 (observed) and 2001 to 2050 (projected). Earlier work linking convection over the Indian Ocean with African moisture transports (Funk et al., 2008) is considered in respect of changes in atmospheric circulation and land surface fluxes.
Searching for long-term trends has certain pitfalls. The observational density changes, satellite data are assimilated after 1979, and the time series may contain multi-decadal oscillations that yield second order trends. To establish the rate and pattern of linear trends, we average the monthly data into annual blocks. A linear trend line is fitted over the entire period, and its slope or rate-of-change is calculated at each grid point. The slope is then mapped using contouring software. For winds, the trends of U, V, W components are recombined into vectors. An annual time step is used for spatial maps; a 5-year running mean is used for temporal analysis over the period 1948–2006. Projected trends are provided by cGCM results from the IPCC 4th assessment, using a slow rise of CO2 (A1B scenario; Leggett et al., 1992; Carter et al., 1999; SRES, 2000) in the period 2001–2050. Although a number of cGCM outputs are available, it is the Geophysical Fluid Dynamics Lab model (GFDL; Delworth, 2006; Gnanadesikan et al., 2006; Griffies et al., 2006) that exhibits a pattern consistent with observed trends across northeast Africa. To place the observed and projected trends in context, Baro River catchment averaged time series are presented which cover a longer 200 year period 1900–2100 using FEWS, NCEP (Whitaker et al., 2004; Smith et al., 2008) and IPCC model data that includes the Community Climate System Model (CCSM3; Kiehl et al., 1998) for temperature. Seasonal evaluations are briefly considered in Section '3.3. Synthesis and discussion', but inter-annual variability (i.e. ENSO) is outside the scope of this study.
3.1. Observed and projected trends over Ethiopia
Surface air temperatures have risen ∼0.03 °C year−1 in the observational period since 1948 across most of Ethiopia, with exception of the eastern Rift Valley and lowlands (Figure 3(a)). Warming has been greatest in the Nile Valley of Sudan. The temperature increase over the 1948–2006 period amounts to ∼1.8 °C along the northwestern border of Ethiopia. Rainfall trends are mapped in Figure 3(b), and these indicate weak rising trends in the arid lowlands of southeastern Ethiopia. Over the western highlands and Sudan border, a downward trend was observed. It reaches − 0.4 mm month−1 year−1, yielding a desiccation of ∼− 23 mm month−1 in the observed period. Vertical N–S sections of NCEP data are analysed for trend in Figure 3(c) and (d). The layer of greatest warming extends from 10 to 20N up to 600 hPa and exhibits a northward tilt. Above 600 hPa the trends are weak, suggesting a bottom-up effect. Similarly the specific humidity exhibits a drying trend in the same subtropical zone and near-surface layer; the drier conditions extend upward in the equatorial zone. Funk et al. (2008) suggested that ridging over equatorial Africa could switch cool-moist transports from the Indian Ocean to hot-dry air drawn from the Sahara—this appears consistent with the observed trend towards aridity. A further impact of this switch, however, may be that ridging over equatorial Africa also disrupts moisture transports from the Congo basin into southern Ethiopia. Trends in run-off are analysed across Ethiopia for both observed and projected eras (Figure 3(e) and (f)). There is a declining trend for observed run-off as expected from the rainfall pattern that extends across the Blue Nile catchment, according to VIC reanalysis. In the GFDL A1B simulation, the downward trend for run-off continues at about the same rate and tends to focus in the Baro catchment (6–10N, 34–38E).
Circulation trends over Ethiopia are characterized in Figure 4. The observed era has a trend towards low level northeast wind and upper level westerlies, indicating that changes in the zonal Walker circulation have occurred. The northerly low level winds originating from the desert are drier and hotter than the climatogical conditions, which blow from the southwest—advecting cooler moist air from the Congo. The tendency towards upper level westerlies denotes a weakening Tropical Easterly Jet (TEJ). A weakening TEJ is associated with less upper level divergence, increased subsidence across Ethiopia, and decreases in precipitation (Segele et al., 2009b). The GFDL-projected trends, on the other hand, exhibit low level northerlies and upper level southerlies, suggesting a gradual strengthening of the northern Hadley circulation. Both observed and projected patterns of 500 hPa vertical motion are consistent with the rainfall pattern, with a trend of sinking motion across western Ethiopia and southern Sudan (Figure 4(c)), that is consistent with declining run-off in the Baro River catchment of western Ethiopia (cf Figure 3(e)).
To analyse how the trends evolve across Africa, we construct hovmoller plots of 5-year smoothed rainfall and temperature anomalies averaged over the 5–15N latitude band in two periods: 1948–2006 and 2001–2050. The FEWS observations clearly distinguish a multi-decadal wet period up to 1970 (Figure 5), followed by alternating periods of drought through the 1980s. A recovery of rainfall in the late 1990s is carried forward into the GFDL projection, in agreement with recent flood spells (Jury, 2010). In the A1B scenario, dessication is projected to set in quickly and is only interrupted by a wet spell in the 2020 decade. Conditions in the 2040 decade are projected to be drier than the 1980s all across the Sahel. The temperature anomaly hovmoller (Figure 5) reveals sharper observed signals over Sudan (20–30E) than elsewhere across Africa. A cool era prevails until 1980 and is followed by significant warming west of the highlands through the 1980s and 1990s. The GFDL projection does not accentuate the signal from 20 to 30E, rather it shows a gradual warming with time that intensifies in two steps after 2020 and 2040.
3.2. Trends at hemispheric and centennial scales
A wider context is needed to understand the factors underlying the climatic trends in Ethiopia. The Walker and Hadley circulations are analysed in Figure 6(a) and (b). In the zonal dimension, a trend towards sinking easterly flow in the 700–500 hPa layer prevails in both observed and projected fields (1948–2006 and 2001–2050, respectively). In the observed structure, an upper westerly/lower easterly Walker circulation trend is found in agreement with Hastenrath (2000). Descent over the eastern Sahel is compensated by weak rising motions over 30 W and 80E, an undulating pattern of wavelength 110° longitude that reflects an atmospheric Rossby wave train and zonal overturning cell over the Atlantic Ocean. Ascending air and increased warming over the Indian Ocean (80E) will also be associated with a westward Rossby wave response contributing to subsidence across the eastern Sahel and western Ethiopia (Funk et al., 2005, 2008). The GFDL projection indicates a trend towards rising motion over the West Indian Ocean (60E) and a continuation of sinking easterly flow over the western Sahel, with stronger rising motions over the east Atlantic (30 W). Thus the undulating pattern is maintained with wavelength closer to 90° longitude that intensifies subsidence over Sudan and Ethiopia. In the meridional dimension, a trend of sinking motions prevails in the 5–15N band above 600 hPa that was stronger in the past than is projected for 2001–2050. The sinking motion is connected with trends in the southern Hadley circulation in the observed era, with trends of lower northerly and upper southerly flow over tropical southern Africa (15S–0). In the GFDL A1B simulation, the southern Hadley trend is disorganized, while the northern Hadley cell exhibits a trend of upper southerlies and lower northerlies that favour uplift (descent) around 15–20N (35–40N). Thus the observed trends in the Hadley circulation contribute to desiccation, but the projected Hadley trends are indeterminant. The Walker circulation, on the other hand, is consistent from observed to projected eras. Its westward shift could desiccate the Sahel.
Considering the hemispheric scale trend patterns for rainfall, surface air temperature and 700 hPa wind circulation (Figure 7), the observed drying trend over Ethiopia coincides with increasing rainfall over the West Indian Ocean (55–90E, 5S–20N) in the satellite era. In the GFDL projection, the trend of oceanic rainfall accelerates (> + 0.4 mm month−1 year−1) and shifts southwestward. The dry area spreads from northeast Africa across the Sahel. Surface air temperatures in NCEP reanalysis warmed rapidly over Sudan and the tropical Indian Ocean (> + 0.03 °C year−1). In the GFDL projection, the zone of warming spreads across the Sahel along 15N and reaches + 0.04 °C year−1. Some of these teleconnections may be explained by the Gill (1982) model of atmospheric convection adjustment to a tropical (Indian) ocean heat source. Compensating subsidence occurs (over Africa) and forms an atmospheric Rossby wave train. Analysing the observed trends of 700 hPa winds, twin anticyclonic gyres are seen over Africa on either side of the equator: easterly trends sweep from Ethiopia past the Congo. This twin anticyclonic circulation trend is maintained and shifts westward to the Gulf of Guinea in the GFDL projection. Funk et al. (2008) describe a model simulation that accounts for the observed desiccation of northeast Africa via changes in circulation forced by a warmer West Indian Ocean.
It is necessary to examine centennial time scales to determine whether trends may be related to multi-decadal oscillations. For this we extend the earlier analysis across the 20th and 21st centuries (Figure 8), using an area-average for the Baro River catchment (7–10N, 34–37E). Both the FEWS and NCEP surface air temperatures indicate a second order upward trend with warming since 1960. The two IPCC A1B model projections exhibit linear upward trends, with the rate of GFDL around + 0.03 °C year−1 and the CCSM3 ∼ + 0.02 °C year−1. Both observed and projected rainfall contain decadal variability as described in Jury (2010), yet exhibit opposing second order trends. The observed record is relatively wet in mid-20th century, whereas the projected rainfall declines in mid-21st century. The feature underlying this multi-decadal climate oscillation is a global SST dipole in northern and southern mid-latitudes (Enfield et al., 2001) that influences the subtropical gyres, Hadley circulation, and rainfall over tropical Africa (Fontaine et al., 1998). Although the drying trend found in the 1948–2050 period is projected to reverse in the second half of the 21st century, drought may affect Ethiopia from 2020 to 2050 (Figure 8(b)) and spread across the Sahel.
3.3. Synthesis and discussion
Over the past 25 years the population of northeast Africa has doubled while per capita cropped area declined 33% and undernourishment increased 80% (Funk et al., 2008). Low per-capita agricultural production limits progress towards better health and rural economic prosperity. The observed and projected decline in rainfall and increase of temperature over Ethiopia may limit growth and cause food insecurity. The desiccation of Africa appears linked with increased rainfall over the tropical oceans. The rate of regional warming is greatest over Sudan and across the Indian Ocean: ∼1 °C since 1960. Many of the models employed in the Coupled Model Intercomparison Project (CMIP3) simulate this warming and implicate greenhouse gas emissions and the global ocean circulation as key drivers.
Model-projected increases in tropical Indian Ocean temperatures appear driven by trends in equatorial Pacific zonal wind stress and shifts in the position of subtropical anticyclones (Jury and Huang, 2004; Funk et al., 2008). While warmer SSTs stimulate marine convection (cf Figure 7(a)), the additional heat released to the atmosphere creates a Rossby wave train and altered zonal circulation. Moisture convergence over Africa is reduced and the continental airmass becomes more stable. Bands of rising and sinking motion are embedded in the zonal circulation (cf Figure 6(a)), enhancing marine rainfall at the expense of Africa. While prior research has linked warming of the adjacent tropical oceans with dryness across the Sahel (Giannini et al., 2003; Held et al., 2005), our analysis suggests that this extends into southwestern Ethiopia in accord with Seleshi and Zanke (2004). Given the observed and projected decline in rainfall and increase of temperature, a decline in crop yields is possible that would put the population of southwestern Ethiopia (32 million in 2000) at risk. Assuming the prevailing growth rate continues, the population may reach 88 million in 2050. Some of these impacts could be mitigated by sustainable development strategies (Downing et al., 1997). A spatially detailed understanding of climatic trends can help guide this process. While the decline in rainfall is significant (−10%) across the various data sets, inter-annual fluctuations of rainfall and river run-off are even larger. Yet, the geographic diversity of Ethiopia is an advantage: intensification of agriculture in the highlands, combined with improved transportation infrastructure, fair land tenure practices and equitable market access could help offset productivity deficits in the lowlands.
Although climate change affects resources in a cumulative manner, it is useful to determine whether certain seasons are more susceptible. Inspection of seasonal rainfall in the FEWS dataset reveals a small reduction in March to May rains across the southern Rift Valley (3.5–8N, 38–42E) and a steady decline of June to September rains across western Ethiopia, consistent with Funk et al. (2005). These areas are densely populated and critical for local agriculture production. The Baro River catchment March to September rainfall has declined since the 1950s (cf Figure 8(b)), with implications for food security and competition for water resources. It has been suggested that anthropogenic warming may accelerate desiccation over Sahelian countries (Held et al., 2005) whose economies depend on foreign aid. Our analysis (cf Figure 7(c)) highlights that twin anticyclonic circulations inhibit moisture inflows from the Gulf of Guinea and Congo, currently affecting Ethiopia and projected to spread across the lowlands: Sudan, Chad, Niger and Kenya's Rift Valley. Satellite estimated surface temperature over the 2000–2009 era shows rises up to + 1.5 °C across the eastern Sahel that contribute to soil water stress and desiccation.
This study has investigated climatic trends over Ethiopia, through analysis of temperature and rainfall trends and related meteorological fields. The observed datasets include gridded station records and reanalysis products in the period 1948–2006, while projected trends were analysed mainly from the GFDL cGCM A1B simulation 2001–2050. Upward trends in air temperature of +.03 °C year−1 and downward trends in rainfall of − 0.4 mm month−1 year−1 were found along the Sudan border, 1948–2006. These trends are projected to continue to 2050 in the GFDL simulation assuming a gradual doubling of CO2. Over higher elevations the trends are weaker than elsewhere. Large scale forcing derives from warming of SST and increased convection over the West Indian Ocean. This contributes to stronger anticyclonic circulations over northern and southern Africa that inhibit the eastward penetration of moisture from the Gulf of Guinea and Congo. Changes in the regional Walker and Hadley circulations modulate climatic trends over North Africa. In particular, the projected westward shift of subsidence associated with the Walker cell could make recent conditions over Ethiopia, the future scenario for Sahelian West Africa. Although Ethiopia may experience some desiccation, orographic effects on rainfall and temperature make the highlands potentially less vulnerable to the impacts of climate change than the arid lowland countries which surround it. A doubling of agricultural productivity could be achieved with modest investment (Sanchez et al., 2009); Ethiopia could serve as a regional ‘breadbasket’ given socio-economic stability.
The lead author thanks Arba Minch University, Ethiopia, for stimulating this research. The second author received support from the US Agency for International Development's Famine Early Warning System Network and NASA Precipitation science grant NNH06ZDA001N-PMM and thanks Tufa Dinku and Mellese Lemma for their guidance.