1. Introduction
The Blue Nile River is critical to the energy, water, and food security aspirations of Ethiopia, Sudan, and Egypt. In the Ethiopian portion of the Blue Nile basin, a largely subsistence agriculture population depends on seasonal rains for crops and pastures, while at a national level the river is viewed as an opportunity for large-scale hydropower development. In Sudan, the waters of the Blue Nile drive hydroelectric power generation and feed major irrigation schemes. In Egypt, water from the Blue Nile accounts for approximately 60% of the annual average 84 Gm3 of the Nile River at the Aswan High Dam (Mohamed et al. 2005).
Flows in the Blue Nile, however, are far from consistent. The river is fed almost exclusively by precipitation in the Ethiopian highlands (Conway 1997, 2000; Beyene et al. 2010), which is concentrated in a 4-month rainy season [June–September (JJAS)] (Conway 2000; Beyene et al. 2010) and is subject to significant intraseasonal, interannual, and interdecadal variability (e.g., Conway and Hulme 1993; Camberlin 1995; Seleshi and Demaree 1995; Conway 1997, 2000; Segele and Lamb 2005; Abtew et al. 2009; Jury 2010; among many others). Improved understanding of the drivers of precipitation variability is required in order to enhance subseasonal and seasonal precipitation forecasts under current climate conditions and to provide a sound basis for projecting potential shifts in precipitation under climate change.
a. Precipitation in the Blue Nile headwaters
The proximal driver of rainfall in the headwaters of the Blue Nile is the northward and southward movement of the intertropical convergence zone (ITCZ). The ITCZ’s migration to the north in the summer months brings convective activity to the basin, with moist air masses from the south and the west (Conway 2000) and potentially the north and the east as well (Viste and Sorteberg 2011), fueling rainfall events (Fig. 1). In this paper, we use the term ITCZ to refer to the zone where the trade winds from both hemispheres converge.
Air masses fueling rainfall in the Blue Nile basin. The box shows the approximate geographic location of the Blue Nile basin. The contours show average summer precipitation (mm) from the Climatic Research Unit (CRU). Arrows represent the approximate direction of air inflow from 1) the Indian Ocean, 2) the Arabian Sea, 3) the Mediterranean and Red Seas, 4) the Sahel region, and 5) the Congo and Gulf of Guinea.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
To the east and southeast, southwest monsoon winds over the Arabian Sea, a strong cross-equatorial flow along the East African coast and over the adjacent ocean, and southeasterly trade winds in the Southern Hemisphere dominate lower atmosphere circulations (Gissila et al. 2004), while to the west and southwest, moist air from the equatorial Atlantic Ocean and the Gulf of Guinea is advected across the Sahel and the Congo into the Blue Nile basin by low-level westerly winds (Segele et al. 2009; Diro et al. 2011). From October through May, the ITCZ shifts southward and dry conditions prevail.
Notably, though precipitation generally peaks in July and August, its interannual variability is greatest at the beginning and end of the rainy season, in June and September (Fig. 2), reflecting variations in the onset and cessation of summer rains. Variability early and late in the rainy season is particularly important to rainfed agriculture in the region, as delayed onset and/or early cessation can result in crop failure (Segele and Lamb 2005).
The fraction of the summer season precipitation contributed by each month (lines) and the coefficient of variation of precipitation in each month (bars) in the Ethiopian Blue Nile from 1951 to 1997 using CRU and GPCC data.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
Spatial variability in precipitation within the Blue Nile headwaters region is also a topic of considerable interest. There is a general pattern of humid conditions in the southern highlands of the basin grading to drier conditions in the north and west, but the dissected topography of the highlands also leads to strong local climate contrasts that include diverse patterns of total rainfall and rainfall variability (Block and Rajagopalan 2007; Krauer 1988). In the present study, we focus on basinwide precipitation, which is the variable of greatest relevance to transboundary water resources and to regional agriculture on the whole, but the results obtained at basin scale will not necessarily hold at all locations within the basin.
b. Drivers of climate variability
Because of the magnitude and importance of precipitation variability in the Blue Nile, there have been a number of studies of the statistics, mechanisms, physical drivers, and predictability of annual precipitation in the region. In almost all cases these studies have treated the rainy season (JJAS) precipitation as a single variable and have examined interannual variability in the JJAS precipitation total (e.g., Gissila et al. 2004; Diro et al. 2011; Korecha and Barnston 2007). In this context, authors have long noted the association between annual variability in Blue Nile flows and precipitation and the Indian monsoon. This association was first noted by Sir Gilbert Walker (Walker 1910; Walker and Bliss 1932), who identified positive correlations between precipitation over India and the Nile flood level in Egypt—which is driven largely by the magnitude of the annual Blue Nile flood. This connection has been explored in more detail by Camberlin (1997), who noted that strong Indian monsoon conditions lead to a pressure gradient (caused by pressure intensification in the west and lowering of pressure in the east) near the equator, which gives rise to abnormally strong westerly winds that advect moisture from the Congo Basin to Ethiopia. The importance of these westerly winds has been noted by other authors as well, though without reference to the Indian monsoon circulation.
Segele et al. (2009), for example, found that high JJAS rainfall in the Horn of Africa is associated with the enhanced westerly advection of water vapor from the Atlantic Ocean and Gulf of Guinea into the Ethiopian highlands. However, they stressed the role of a sea level pressure (SLP) gradient between the Arabian Peninsula and the Gulf of Guinea, caused by intensification of sea level pressure over the Gulf of Guinea and deepening of the monsoon trough across the Arabian Peninsula—defined as the region of lowest sea level pressure over the Arabian Peninsula—in driving these winds and associated rainfall. The importance of low-level westerly winds to precipitation in the Blue Nile and the East African highlands more generally has also been noted in other studies (Flohn 1987; Seleshi and Demaree 1995; Vizy and Cook 2003; Seleshi and Zanke 2004; Mohamed et al. 2005; Korecha and Barnston 2007; Levin et al. 2009; Diro et al. 2011), though not always with reference to a driving mechanism.
The intensity and spatial extent of the St. Helena high, which is centered over the subtropical southern Atlantic Ocean, and the Mascarene high, centered in the southwest Indian Ocean, are associated with variability in Ethiopian summer rainfall (Conway 2000; Gissila et al. 2004; Seleshi and Zanke 2004; Korecha and Barnston 2007; Segele et al. 2009). The St. Helena high modulates the strength of the westerlies through West Africa and the southwesterlies that advect moisture to the Blue Nile from the Gulf of Guinea and the Congo Basin, while the Mascarene high affects the strength of the Somali low-level jet (SLLJ), which influences precipitation in the second half of the rainy season.
Perhaps the most widely analyzed driver of Blue Nile precipitation variability is the El Niño–Southern Oscillation (ENSO) and its associated indices. The link between ENSO variability and annual precipitation in the Blue Nile (El Niño is associated with dry conditions and La Niña with high rainfall) has been noted in many studies (e.g., Tadesse 1994; Camberlin 1997; Conway 2000; Gissila et al. 2004; Segele and Lamb 2005; Block and Rajagopalan 2007; Segele et al. 2009). A number of mechanisms have been invoked to explain this relationship, including ENSO influence on the strength of the southeasterly flow from the Indian Ocean, which feeds precipitation events in Ethiopia from the southeast; the intensity of westerlies from the Atlantic Ocean, which bring moisture into Ethiopia from the west; the position of the African easterly jet (AEJ), which has dynamical effects that influence convection; the intensity of the Mascarene high, which can also affect moisture transport from the Indian Ocean; and anomalies in the North African–Asian Jet, which affects the strength of the Tibetan upper-level anticyclone and can, therefore, potentially influence African precipitation through upper-level Rossby waves that may weaken divergence at the exit of the tropical easterly jet (TEJ) (Diro et al. 2011; Segele et al. 2009; Shaman and Tziperman 2007).
In addition, the northward migration of the ITCZ in Asia is proportional to the magnitude of the easterly vertical shear of the zonal winds over the Indian monsoon region (Jiang et al. 2004), which is modulated by the tropospheric temperature gradient between Asia and the equatorial Indian Ocean (Goswami and Xavier 2005). During summertime El Niño events, cold tropospheric temperature anomalies are experienced over much of Asia, and warm anomalies occur over the Indian Ocean (Goswami and Xavier 2005). This meridional temperature gradient anomaly decreases the easterly vertical shear of the zonal winds by altering the thermal wind balance and as a result slows the northward migration of the ITCZ (Shaman and Tziperman 2007). The influence that this ENSO-related influence on the ITCZ has on Indian monsoon precipitation has been noted in previous studies (Goswami and Xavier 2005), and the mechanism could potentially affect ITCZ migration in East Africa as well. None of these hypothesized ENSO-related mechanisms are mutually exclusive, but given the fact that such a diversity of independent mechanisms has been proposed—and the fact that most have been studied at seasonal time scales—ENSO influences in this region require further study.
c. Application to seasonal forecast
The range of perspectives on drivers of precipitation in the Blue Nile, as described above, have given rise to a range of proposals on the most effective way to predict Blue Nile precipitation at the seasonal scale. The Ethiopian National Meteorological Agency (NMA) supports the leading operational seasonal forecast system, which predicts Ethiopia-wide precipitation using SST as a predictor (Diro et al. 2011).
This general approach—predicting total rainy season precipitation as a function of global sea surface temperature (SST) anomalies—has been the focus of numerous research studies by the NMA and other research groups. Researchers have differed, however, in which predictors they identify as most promising. Camberlin (1997), for example, documented that Ethiopian precipitation is more strongly teleconnected with Indian monsoon precipitation than with ENSO, and he suggested that predictors of the Indian monsoon precipitation can be used as predictors for summer precipitation in East Africa. Gissila et al. (2004) divided Ethiopia into clusters on the basis of gauged precipitation patterns and then constructed multivariate regression forecast models for each region. These models made use of SST anomalies of the western Indian Ocean, the tropical eastern Indian Ocean, and the Niño-3.4 region. Korecha and Barnston (2007) found that the most important governing factor for Ethiopian summer rainfall, excluding the southern and southeastern lowlands, is ENSO. Block and Rajagopalan (2007) implemented a nonparametric local polynomial regression technique to predict upper Blue Nile summertime precipitation as a function of multiple large-scale predictor variables, including SST, sea level pressure, air temperature, 500-hPa geopotential height anomalies, and the Palmer drought severity index. Diro et al. (2011) regionalized the country into zones of homogeneous summer rainfall and then generated multivariate regression model forecasts that made use of different oceanic regions and lead times for different regions.
Importantly, virtually all analyses of Blue Nile precipitation have focused on seasonally averaged precipitation (June–September) as the primary predictand of interest, though Block and Rajagopalan (2007) did describe a probability-based method to disaggregate seasonal forecasts to monthly predictions on the basis of historical correlations between total seasonal precipitation and precipitation in each month. Given the range of large-scale drivers invoked to explain and predict precipitation in this region, a more detailed temporal analysis is warranted.
Here, we consider teleconnections and potential drivers of precipitation variability for each month—June, July, August, and September—in order to characterize the general evolution of the teleconnections over the course of the rainy season. In this context, calendar months are used as a convenience, though it is recognized that teleconnections evolve continuously over the season. We present these analyses in order to explore the reasons for the seemingly contradictory seasonal-scale findings of previous studies, to describe the general mechanisms through which diverse large-scale patterns of variability influence Blue Nile precipitation, and to motivate further work on prediction systems that take into account systematic differences in precipitation drivers between the early, mid-, and late rainy season. In addition, we include multiple precipitation and atmospheric reanalysis datasets in the analysis to assess the robustness of identified drivers and proposed mechanisms. The paper is organized as follows: section 2 describes data and methods, followed by results and discussion in section 3. Finally, summary and conclusions are offered in section 4.
2. Data and methods
For all gridded analyses in this study we treat the Blue Nile basin (8.25°–12.75°N, 34.25°–39.75°E) as a single region. This is roughly consistent with many regional-scale analyses of East African precipitation (e.g., Camberlin 1995, 1997; Abtew et al. 2009; Segele et al. 2009; Jury 2010), but we note that a number of researchers have attempted more detailed regionalizations of the Ethiopian precipitation that divides the Blue Nile headwaters region into several distinct subregions. This is the case for regionalizations by Gissila et al. (2004), who divided Ethiopia into clusters on the basis of 42 gauged precipitation patterns and by Diro et al. (2011), who regionalized the country into zones of homogeneous summer rainfall climate using 45 gauged precipitation patterns but presented different regions from that of Gissila et al. (2004). Our study area spans four regions in both the Gissila et al. (2004) and Diro et al. (2011) regionalizations, suggesting that further regionalization of the basin—supported with a sufficient number of station records—could offer more detailed understanding of intraseasonal and interannual precipitation variability.
a. Precipitation and climate data
Precipitation: Gridded monthly precipitation data were drawn from the Climatic Research Unit (CRU) and Global Precipitation Climatology Centre (GPCC) precipitation analyses. The GPCC dataset, operated by the National Meteorological Service of Germany, is gridded at 0.5° latitude and longitude resolution for Earth’s land surface and is generated using meteorological station data (Schneider et al. 2011). The CRU dataset, which was developed at the University of East Anglia using data from meteorological stations all over the globe, contains historical monthly precipitation values for global land areas, also gridded at 0.5° latitude and longitude resolution (Harris et al. 2013). This project uses CRU Time Series, version 3.10.01 (TS 3.10.01) analysis (Harris et al. 2013). In the construction of this dataset, monthly station observations are converted to anomalies by subtracting the 1961–90 normal and values more than 4.0 standard deviations from the normal are deemed outliers and excluded (Harris et al. 2013). Following that, the monthly station anomalies are gridded using triangulated linear interpolation (Harris et al. 2013). We consider the period from 1951 to 1997 because after 1997 the number of stations in the basin CRU TS 3.10.01 uses for interpolation drastically falls (see the appendix, Fig. A1). For our study area, the number of observation stations used for the interpolation ranges from 9 to 19 over the period of analysis; moreover, more than 88% of the months in the study period have 13 or more observation stations, 41 months have 18 observations stations, and 13 months have 19 observation stations, while only 6 months have 9 observation stations. In spite of the paucity of station records employed in CRU TS 3.10.01 in the period since 1997, we did repeat all analyses for the entire 1950–2009 period and the results were broadly consistent with the 1951–97 analysis. Consistencies and discrepancies between the 1951–97 and 1950–2009 analyses are noted as appropriate in the results section. For detailed information about CRU TS 3.10.01 product, the reader is referred to Harris et al. (2013). While analyses presented in this paper focus on these gridded precipitation datasets, monthly point precipitation records were also extracted for four in situ meteorological stations in Gonder, Bahrdar, Nekemt, and Mehal Meda maintained by NMA. These stations have data records ranging from 29 to 44 years in length and are distributed across the Blue Nile.
Wind speed, SLP, and SST: The monthly zonal and meridional wind speed and SLP data were drawn from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis 1 (NCEP-R1) and the 40-yr European Center for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40). NCEP–NCAR reanalysis data were obtained from the National Oceanic and Atmospheric Administration (NOAA) Cooperative Institute for Research in Environmental Sciences (CIRES) Climate Diagnostics Center (Boulder, Colorado). It is gridded at 2.5° × 2.5° latitude–longitude resolution and spans the period from 1948 to the present. ERA-40 is a reanalysis of meteorological observations from September 1957 to August 2002 produced by the European Centre for Medium-Range Weather Forecasts using different data sources (Uppala et al. 2005). It is gridded at 2.5° by 2.5° latitude–longitude resolution. SST fields were drawn from the Met Office Hadley Centre global SST dataset (Rayner et al. 2003).
b. Climate indices
A wide range of climate indices and atmospheric fields was considered in this study. Our selection of indices was guided by previous studies of precipitation variability in the East African highlands and was refined through the application of multivariate regression analysis, as described below. For ENSO, we include multiple indices, as this climate feature has received extensive attention in previous studies, and it is possible that correlations with Blue Nile precipitation are different for different ENSO indices. Indices considered in the analysis include:
1) Multivariate ENSO index (MEI): calculated as the first unrotated principal component (PC) of the following observations combined after normalization: surface air temperature, sea level pressure, SST, zonal and meridional components of the surface wind, and total cloudiness fraction of the sky over the tropical Pacific. The MEI values are computed separately for each of the 12 sliding bimonthly seasons and are standardized with respect to each season and to the 1950–93 reference period (Wolter and Timlin 1993). The MEI is derived from long-term marine records from the Comprehensive Ocean–Atmosphere Data Set.
2) Niño-3.4: defined as the SST in the east central tropical Pacific from 5°S to 5°N and from 120° to 170°W. It is calculated using the NOAA optimum interpolation (OI) SST, version 2 (OISSTV2).
3) Southern Oscillation index (SOI): An index defined by a standardized monthly-mean sea level pressure difference between Tahiti and Darwin.
4) Bombay monthly SLP: an indicator of monsoon activity. It is defined as the SLP anomaly at 19°N, 72.8°E, located on the western coast of India.
5) Arabian Peninsula SLP: area average SLP over the Arabian Peninsula over the region 15°–28°N, 40°–57°E.
6) Equatorial Atlantic SLP: area average SLP in the equatorial Atlantic Ocean over the region 3°–20°N, 38°–20°W.
7) Pacific decadal oscillation (PDO) index: represents low-frequency changes in the SST patterns of the Pacific Ocean with centers of action in the northwest Pacific and eastern equatorial Pacific (Mantua et al. 1997). PDO index is the leading PC of monthly SST anomalies in the North Pacific Ocean poleward of 20°N. The index is calculated employing three different datasets: Met Office (UKMO) historical SST dataset for the period from 1900 to 1981, Reynold’s optimally interpolated SST, version 1 (V1), from January 1982 to December 2001, and OISSTV2 since January 2002.
8) The Atlantic meridional mode (AMM): represented as the leading maximum covariance analysis (MCA) mode in the tropical Atlantic basin (Chiang and Vimont 2004) and is used to characterize anomalous meridional gradient of SST between the tropical North and South Atlantic.
9) Tropical Southern Atlantic index (TSA): anomaly of the average of the monthly SST from the equator to 20°S and 10°E–30°W. It is created using Global Sea Ice and Sea Surface Temperature dataset (GISST) and the NOAA optimum interpolation SST.
10) St. Helena high SLP: area average SLP over the South Atlantic high covering 27°–20°S, 25°–10°W.
11) Tropical easterly jet index: area average of the zonal wind greater than 25 m s−1 (Ugt25 in Table 2) in the TEJ region and longitude of the TEJ maximum (longmax in Table 2) are used.
12) African easterly jet index: calculated as the zonal wind averaged from 700 to 600 hPa over 5°–15°N and 10°W–20°E.
13) West African westerly jet (WWJ) index: the area average of the 925-hPa zonal wind speed for 8.4°–10.6°N, 15°–25°W. This averaging region captures the maximum westerly wind (Pu and Cook 2012).
14) Indian summer monsoon index (ISMI): the difference of zonal wind at 850 hPa over the regions 5°–15°N, 40°–80°E and 20°–30°N, 70°–90°E (Wang and Fan 1999).
Indices 3–6 and 10–14 are calculated using NCEP-R1 data.
15) Sahel rainfall index (SRI): standardized rainfall in the region 20°–8°N, 20°W–10°E, calculated using station data obtained from the National Center for Atmospheric Research World Monthly Surface Station Climatology (WMSSC). The averaging region is based on rotated principal component analysis of average June–September African rainfall presented in Janowiak (1988). The index record was obtained from the University of Washington (http://jisao.washington.edu/data/sahel/).
16) Madden–Julian Oscillation (MJO) indices: calculated by applying an extended empirical orthogonal function (EEOF) analysis to pentad velocity potential at 200 hPa for ENSO-neutral and weak ENSO winters (November–April) during 1979–2000, then constructing 10 MJO indices by regressing the daily data onto the 10 patterns of the first EEOF. The 10 centers of enhanced convection (20°, 70°, 80°, 100°, 120°, 140°, and 160°E, and 120°, 40°, and 10°W) for the 10 indices are determined from the 10 time-lagged patterns of the first EEOF of the 200-hPa velocity potential.
17) Quasi-biennial oscillation (QBO) index: zonal average of the 30-mb wind speed at the equator from NCEP–NCAR reanalysis.
MEI, SOI, Niño-3.4, PDO, MJO, AMM, QBO, and TSA indices were obtained from NOAA/Climate Prediction Center.
18) Equatorial planetary wave index (EPWI): calculated by projecting the anomalous eddy zonal wind field at 150 hPa onto the seasonally varying climatological-mean 150-hPa eddy component of the zonal wind field over the domain 20°N–20°S. This index has been available since 1979 [see Grise and Thompson (2012) for details].
c. Data analysis
To explore the range of regional and global variability indices listed above, and also to include the possibility of lagged effects on precipitation in each of the 4 months of the rainy season, it is necessary to perform some form of variable selection. Here, we employ a generalized linear model (GLM) to identify significant explanatory variables in each month. In a GLM, a linear predictor relates (by a link function) the linear predictor to a function of the predictor variables specifying the conditional mean (Cameron and Trivedi 1998). The link function transforms the expectation of the linear predictor. A normal identity link function 
Of the indices listed in section 2b, only those that represent potential large-scale drivers were considered as predictors: Arabian Peninsula SLP, Bombay SLP, equatorial Atlantic SLP, Niño-3.4, SOI, MEI, and PDO. Monthly averages of each predictor were introduced to the GLM, both for the concurrent month and for each leading month from April onward. Both CRU and GPCC precipitation were used as predictands. The collinearity of predictors in each GLM was checked using a variable inflation factor (VIF) measure, and covariates with VIF values greater than 10 were removed. Two different methods were used to compare fitted models for the dataset: the Akaike information criteria (AIC) and a likelihood ratio test, which is used when the covariates in one model are a subset of the covariates in another model.
In this application, GLM analysis was used to focus our study on variables of demonstrated relevance to Blue Nile precipitation. The goal was not to eliminate nonsignificant variables from consideration in a definitive way, nor was it to develop a formal statistical predictive model; the models simply serve as a guide for further examination of mechanisms, which was performed using a combination of composites and linear correlations. For all analyses, interannual autocorrelation is found to be insignificant in all months. For analyses that involve wind speed or distributed sea level pressure fields, NCEP-R1 and ERA-40 were both employed to confirm the robustness of findings. Unless otherwise noted, both datasets provided similar results, and NCEP-R1 is used in tables and figures.
3. Results and discussion
The rationale for applying a subseasonal approach to study Blue Nile precipitation variability is clear from both qualitative and quantitative analysis. Qualitatively, patterns of correlation between Blue Nile precipitation and concurrent monthly averaged SST anomalies differ distinctly between early and late months of the rainy season. For example, monthly composites of precipitation in the basin based on the difference between strong El Niño years and the climatological average (El Niño − climatology) and the difference between strong La Niña years and the climatological average (La Niña − climatology) show significant differences only in September (Figs. 3d,h). We also found that June precipitation is negatively associated with specific humidity in the lower midtroposphere (700–600 hPa) averaged over the study area, while September precipitation is positively associated with specific humidity at that level. These associations are significant at the 85% level—as evaluated using composite analysis statistics described by Terray et al. (2003)—suggesting that regional-scale processes associated with high or low precipitation also differ between the beginning and end of the rainy season.
Composites of precipitation (mm month−1) based on (top) strong El Niño years − climatology and (bottom) strong La Niña years − climatology in (a),(e) June, (b),(f) July, (c),(g) August, and (d),(h) September. Shading shows results significant at the 90% confidence level. Strong El Niño years were 1957, 1965, 1972, 1982, 1991, and 1997, and strong La Niño years were 1955, 1973, 1975, and 1988.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
While the seasonally averaged (June–September) pattern of correlation reflects the frequently noted connection between ENSO-like SST variability and Blue Nile precipitation (Fig. 4a), early season precipitation shows extremely weak correlations with tropical Pacific SST (Figs. 4b,c). The seasonally averaged pattern is almost exclusively the result of correlations found in the second half of the rainy season, particularly in September (Figs. 4d,e). The correlation of detrended SST with Blue Nile precipitation gives similar results (not shown) indicating that the correlation pattern is not caused by the SST trend. Moreover, composites of SST based on wet–dry Blue Nile conditions give significant results in the eastern equatorial Pacific Ocean only in September and when the rainy season is considered as a single variable (not shown). We also considered impacts of outliers by removing values greater than or less than two standard deviations. As shown in Table 3, the correlations after removing extremes have a similar pattern to the correlations shown in Table 2, which shows that the intraseasonal correlation of the basin precipitation with ENSO indices is stable. These correlation results are similar when the period 1950–2009 is considered.
Correlation of Blue Nile CRU precipitation with concurrent SST from 1951 to 1997 for (a) the full rainy season (June–September) and each month within the rainy season: (b) June, (b) July, (d) August, and (e) September. Only correlation coefficients exceeding the 90% confidence levels are shown.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
Correlations between Blue Nile precipitation and global SLP fields show a similar pattern (Fig. 5). The seasonal correlation between Blue Nile precipitation and high pressure (La Niña–like conditions) over the equatorial and southeast Pacific Ocean and low pressure centered over the Indian Ocean (indicative of strong Indian monsoon conditions) derives almost entirely from correlations in August and September. In composite analysis, the ENSO-like SLP signal is significant only in September.
Correlation of Blue Nile CRU precipitation with NCEP-R1 SLP from 1951 to 1997. (a) Summer precipitation with summer SLP, and monthly correlation of precipitation with concurrent SLP for (b) June, (c) July, (d) August, and (e) September. Only correlation coefficients exceeding the 90% confidence levels are shown.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
Correlations between precipitation in the Blue Nile and concurrent SLP anomalies over the Arabian Peninsula are pronounced in July, August, and September (Figs. 5c–e). There is some significant correlation in June as well (Fig. 5b); however, the center of the correlation in this month is in the eastern Mediterranean Sea and surrounding areas. Arabian Peninsula SLP can influence Blue Nile precipitation through at least two mechanisms. First, Arabian Peninsula SLP shows significant negative correlation with water vapor transport at 850 hPa through West Africa in all months (June–September), with a maximum in July and August (Fig. 6). This suggests that low SLP in the Arabian Peninsula is associated with the enhanced transport of wet air into the Ethiopian highlands from the west. Second, Arabian Peninsula SLP is significantly negatively correlated with the 850-hPa water vapor transport from the western Indian Ocean and East African coast, with maximum correlation found in September (Fig. 6d). This water vapor convergence over the Arabian Peninsula—caused by the deepening of the monsoon trough over the Arabian Peninsula that strengthens the south-to-north pressure gradient—promotes the development of more intense convective systems over the Yemen highlands, which propagate westward and produce wetter conditions over Ethiopia (Segele et al. 2009). In June, there is significant correlation with SLP in the tropical Atlantic Ocean, where low pressure is associated with wet conditions in the Blue Nile (Fig. 5b). Correlations of Blue Nile precipitation with SLP, and Arabian Peninsula SLP with water vapor transport at 850 hPa are broadly consistent in the periods 1950–2009 (not shown) and 1951–97.
Correlation of Arabian Peninsula SLP with concurrent water vapor transport (horizontal wind speed × specific humidity) at 850 hPa for the period 1951–97: (a) June, (b) July, (c) August, and (d) September. Only correlation coefficients exceeding the 90% confidence levels are shown. The box indicates the approximate geographic location of the Blue Nile basin. The arrows are the mean climatology of wind.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
This general subseasonal pattern of varying influences is captured by multivariate regression analysis as well: predictors associated with ENSO, as well as with variability in the Indian monsoon zone, are most significant in August and September (Table 1). In June and July, the only significant predictors are concurrent equatorial Atlantic Ocean and Arabian Peninsula pressure anomalies, respectively. While this analysis was not optimized for predictions—it is likely that regionalization of the Blue Nile and an exclusive focus on leading, rather than concurrent, indicators would provide improved predictive skill—the clear implication is that ENSO- and Indian monsoon–derived prediction models draw their skill from late season variability and that the prediction of early season precipitation requires a different approach. Indeed, time-lagged correlations between MEI and Blue Nile precipitation, for example, indicate that the predictive potential is weak at the beginning of the rainy season and progressively strengthens toward the end of the season (Fig. 7; Table 2), while concurrent correlations with precipitation over India are similarly limited to September (Fig. 8; Table 2).
Major drivers of precipitation in the Blue Nile basin in each month as found from GLMs.
Correlations between monthly CRU precipitation and MEI for the period 1951–97. For each month, correlations are shown for CRU in that month and MEI in the leading month that provides the highest average correlation in the Blue Nile analysis region. (a) June precipitation with March–April MEI, (b) July precipitation with May–June MEI, (c) August precipitation with June–July MEI, and (d) September precipitation with June–July MEI. Only correlation coefficients exceeding the 90% confidence levels are shown. The box indicates the approximate geographic location of the Blue Nile basin.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
Monthly correlation of Blue Nile precipitation from 1951 to 1997 with different climatic indicators. Correlations in boldface are statistically significant at the 99% confidence level. Correlation values of 0.29 and above are significant at the 95% confidence level. The indices 1–10 are potential drivers, 11–14 are mechanisms, and 15 and 16 are precipitation indices. The 〈c〉 denotes concurrent month.
Correlation of precipitation with the concurrent southwest Indian monsoon rainfall using CRU, for the period 1951–97 in (a) June, (b) July, (c) August, and (d) September. Only correlation coefficients exceeding the 90% confidence levels are shown. The box indicates the approximate geographic location of the Blue Nile basin.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
Significant correlations between Blue Nile precipitation and indices of the MJO are found in all months of the rainy season, but as monthly scale analysis of MJO indices is subject to temporal aliasing, we defer analysis of this correlation for a future study at submonthly time scales. The QBO indices show no significant correlations with Blue Nile basin precipitation in this study. Results are similar for CRU and GPCC precipitation datasets. Correlations with NMA meteorological station precipitation records were, on average, consistent with the gridded products, though the stations reveal intrabasin spatial variability. These patterns of intrabasin spatial variability of precipitation are the subject of ongoing study.
Table 2 offers a summary of all correlations found to have predictive or explanatory potential in this study, in the form of univariate linear correlations. Lead time correlations are shown only for statistically significant associations. The monthly evolution of teleconnection patterns found here strongly suggests that seasonally averaged analyses fail to capture the full character of remote drivers of Blue Nile precipitation variability. To further explore this observation, we now consider teleconnections and proposed mechanisms in detail for each month of the rainy season, beginning with September—the month that most closely matches mechanisms emphasized by the majority of previously published studies of Blue Nile precipitation variability—and working back through August, July, and June.
a. September
September precipitation variability is strongly correlated with climate indices of the tropical Pacific and Indian Oceans. The often noted ENSO connection with Blue Nile precipitation is strongest in this month, as evident in correlations with MEI, Niño-3.4, and SOI at lead times of up to 5 months and correlations with PDO with a lead of up to 2 months (Table 2). Both CRU and GPCC September precipitation show statistically significant correlations at the 99% confidence level with MEI, Niño-3.4, PDO, and SOI. For example, CRU and GPCC reveal, respectively, correlations of −0.68 and −0.80 with July Niño-3.4. Both precipitation datasets show significant correlations, at the 99% confidence level, with MEI starting in April–May, with Niño-3.4 starting in April, and with SOI starting in May.
A range of mechanisms have been proposed to explain the link between ENSO and East African precipitation, including ENSO-modulated variability in the SLLJ, TEJ, and westerlies and southwesterlies from the Atlantic Ocean. Correlation analysis reveals that there is significant correlation between ENSO indices and these atmospheric features, which exhibit marked weakening during El Niño and strengthening during La Niña (Table 2). In addition, Goswami and Xavier (2005) have suggested that ENSO influence on the tropospheric meridional temperature gradient between Asia and the Indian Ocean modulates the northward migration of the ITCZ, with significant influence on Asian monsoon precipitation. Such an influence on the ITCZ might be expected to influence rainfall in East Africa as well, and we do find that there is a significant correlation between Blue Nile (CRU/GPCC) precipitation and the meridional temperature gradient between boxes 10°–35°N and 15°S–10°N averaged from 30° to 110°E and from 700 to 200 hPa in September (0.53 CRU/0.60 GPCC). This correlation is found in July (0.43/0.50) and August (0.42/0.47) as well.
Some researchers have documented that the TEJ is one of the phenomena most likely to control Ethiopian highland precipitation in JJAS (Camberlin 1997; Grist and Nicholson 2001; Nicholson and Grist 2003; Segele and Lamb 2005; Segele et al. 2009; Diro et al. 2011). Two indices for TEJ are presented in Table 2. The east–west location of the jet core (longmax) shows significant correlations with Blue Nile precipitation in September only (−0.63 and −0.69 with CRU and GPCC data, respectively). As shown in Table 3, the correlation values are −0.68 and −0.76 for CRU and GPCC data, respectively, when values greater and less than two standard deviations are removed. Our analysis shows that when the jet is stronger, its core tends toward the Arabian Sea and the East African coast and is associated with more precipitation in the Blue Nile, and when it is weaker its core is located in the Bay of Bengal and Blue Nile precipitation decreases. When the TEJ weakens, divergence at the jet exit (and hence upward vertical motion) decreases, which results in reduced convection in Ethiopia (Nicholson and Grist 2003; Segele et al. 2009; Diro et al. 2011). The other index is the area average of TEJ wind speed (Ugt25), which is taken as zonal wind speed in the TEJ region greater than 25 m s−1. This index shows strong associations with Blue Nile precipitation in the second half of the rainy season and is strongest in September. Composites of vertical motion based on strong–weak TEJ employing the TEJ indices (longmax and Ugt25) are shown in Fig. 9; for both indices the composites of vertical motion are strongest in September (Figs. 9d,h). The correlations between the two TEJ indices and Blue Nile precipitation are similar in the 1950–2009 and 1951–97 time periods. For example, in September, correlations between CRU/GPCC precipitation and concurrent longmax are −0.63/−0.69 and −0.59/−0.65 for the 1950–2009 and 1951–97 time periods, respectively.
Monthly correlation with Blue Nile precipitation, from 1951 to 1997, after removing values greater than and less than two standard deviations. Correlations in boldface are statistically significant at the 99% confidence level. The 〈c〉 denotes concurrent month.
Composites of vertical velocity (Pa s−1), averaged from 8.25° to 12.75°N, based on (top) strong–weak TEJ longmax and (bottom) strong–weak TEJ Ugt25: (a),(e) June, (b),(f) July, (c),(g) August, and (d),(h) September. Shading shows results significant at the 90% confidence level.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
Camberlin (1997) suggested that the major trigger for East African precipitation variability is monsoon activity over India and that predictors of Indian monsoon precipitation could potentially be used as precipitation predictors for East Africa. The correlation between Indian and Blue Nile precipitation is clear in September (Fig. 8) but not in any other month. We note that even this very strong correlation between Blue Nile and Indian monsoon precipitation in September can be explained in large part by the fact that precipitation in both regions is modulated by ENSO variability in this month. This can be demonstrated by partial correlation analysis: in September, partial correlation of the Blue Nile and Indian precipitation when the Niño-3.4 index is kept fixed results in a significant decrease in the calculated correlation between Blue Nile and Indian monsoon precipitation. In this month, the correlation values with the Indian summer monsoon index are 0.62 and 0.63 for CRU and GPCC, respectively (Table 2). If Niño-3.4 is held fixed, the partial correlation values drop to 0.33 and 0.29 for CRU and GPCC, respectively. These values are still significant at the 95% confidence level, suggesting that the linkage between the monsoon and East Africa is not solely due to a common response to ENSO, but the drop does suggest that ENSO variability plays a large role in observed correlations. The fact that the Indian monsoon is weaker during summertime El Niño events has been known for some time (Shaman and Tziperman 2007). In September, correlations of Blue Nile precipitation with precipitation over India and the Indian summer monsoon index over the period 1950–2009 show similar correlation and partial correlation values to those over the 1951–97 time period shown in Fig. 8 and Table 2.
As September is the month of Indian monsoon cessation and the last month of the rainy season in the Blue Nile, it may be more useful to explore processes that link the cessation of rainfall in these two regions than to apply Indian monsoon predictors to East Africa in a general fashion. No significant time-lagged correlations between the two regions were found for CRU or GPCC.
Figure 10 shows one potential mechanism linking Indian and Blue Nile precipitation: in September (and, to a lesser extent, August as well), wet conditions in the Blue Nile—and in India (e.g., Joseph and Srinivasan 1999; Pai et al. 2011)—are associated with strengthened low-level flow in the western Indian Ocean. This strengthening originates in the Southern Hemisphere, as the southern trade winds are strengthened when low pressure in the northern Indian Ocean enhances the pressure gradient between the Mascarene high and the monsoon low over the Indian subcontinent. These wind anomalies manifest as stronger southeasterlies in the Horn of Africa, bringing wet air into the Blue Nile highlands from the Indian Ocean. The same wind anomaly is evident in strengthened westerlies across the Arabian Sea, bringing moisture into the Indian monsoon zone (Fig. 10d). Figure 10d also shows that Blue Nile precipitation is strongly associated with the Pacific Walker circulation. When this circulation cell is strong, precipitation in the basin increases. Figure 10 was generated using CRU precipitation and NCEP-R1 wind fields, but the same patterns are evident when GPCC precipitation data and ERA-40 wind fields are used.
Correlation of Blue Nile precipitation with the (left) concurrent zonal and (right) meridional wind speed at 850 hPa for the period 1951–97: (a),(e) June, (b),(f) July, (c),(g) August, and (d),(h) September. Only correlation coefficients exceeding the 90% confidence levels are shown. The arrows are the mean climatology of the wind.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
Vizy and Cook (2003) used a regional climate model to investigate links between summer precipitation anomalies in East Africa and India by perturbing SST in the Arabian Sea and found that rainfall increases over the Blue Nile basin when the monsoon trough deepens. They associated this increase with moisture-laden cross-equatorial flow penetration into southern Ethiopia through the Turkana jet increasing moisture convergence and rainfall rates in the basin. Figure 4e shows that in September SST in the Arabian Sea has a negative correlation with precipitation in the Blue Nile basin, which is in agreement with the summer-scale model results by Vizy and Cook (2003). But this association is evident only in September, which again indicates that mechanisms of precipitation variability evolve over the course of the summer season. The influences of the Turkana jet, through moisture transport and divergence, on Blue Nile precipitation at subseasonal scales is a subject of an ongoing study.
b. August
Examining the mechanisms that are responsible for precipitation development in the basin in August, we see that some are common with September, some are common with July, and others are found exclusively in August. Commonalities with September include correlations with ENSO indices, winds of the Pacific Walker circulation (Fig. 10c) and, to some extent, the strength of the TEJ and phase of the PDO (Table 2), though all of these correlations are weaker in August than in September. As shown in Table 2, in August the correlations between the Indian summer monsoon index and the concurrent CRU/GPCC Blue Nile precipitation are present, but they are much weaker than the correlation values in September. The result is still statistically significant at the 95% confidence level (Table 2), but if we include the entire 1950–2009 period in the analysis, the correlation drops to below 95% significance. Composites of zonal wind speed based on wet–dry Blue Nile precipitation reveal an enhanced Pacific Walker cell in August and September only (not shown). In August and September, concurrent partial correlations of Blue Nile precipitation and PDO keeping Niño-3.4 fixed are insignificant. For example, in September the correlation values are −0.52 and −0.58 (Table 2); while the partial correlation values are −0.19 and −0.2, for CRU and GPCC, respectively. This suggests that the associations of Blue Nile precipitation and PDO are mainly caused by the PDO component associated with ENSO.
There is no significant association with the SST anomalies in the western Indian Ocean (Fig. 4d) that provide a hypothesized link between the Indian monsoon and Blue Nile precipitation, though Bombay SLP is one of the covariates of the August GLM (Table 1), suggesting that there is some link between the regions. In contrast to September, when the most evident correlations and proposed mechanisms for precipitation variability all came from eastern influence—primarily the Indian and Pacific Oceans—August Blue Nile precipitation exhibits strong correlations with processes to the west as well. Western influences include SLP anomalies across West Africa (Fig. 5d), low-level winds across the Sahel region (Fig. 10c), and the WWJ (Table 2), which brings Atlantic Ocean moisture to the Sahel. A strong cross-equatorial pressure gradient intensifies the low-level westerlies via an inertial instability mechanism, which in turn is associated with wet conditions over the Sahel (Nicholson 2009). All of these anomalies have the effect of enhancing moisture transport into the African continent from the west. In August—and only in August—there is also a strong correlation between Blue Nile precipitation and the anticyclonic surface winds in the Southern Atlantic associated with SLP in the region of the St. Helena high (Table 2). This result holds for both NCEP-R1 and ERA-40 (not shown). Mechanistically, a strong St. Helena high pressure center is associated with strengthened anticyclonic winds in the South Atlantic, leading to intensified westerlies through the Sahel and southerly winds entering the Gulf of Guinea that feed into the low-level Congo air stream. An anomalous meridional gradient of SST between the tropical North and South Atlantic Ocean is also correlated with Blue Nile precipitation in August only, as revealed by strong correlations with AMM (Table 2). An anomalously cold tropical South Atlantic Ocean SST (TSA) favors more precipitation in the Blue Nile basin (Table 2), though this association is not significant for the tropical North Atlantic Ocean SST anomaly.
Finally, there is a significant correlation between August precipitation and the AEJ (Table 2). Yeshanew and Jury (2007) showed that a strong AEJ suppresses vertical upward motion in the Sahel regions and as far east as the Blue Nile basin. The connection between Blue Nile and western Africa precipitation processes in August is evident in the strong correlation between Blue Nile and Sahel rainfall in August, which is significant for both CRU and GPCC data (Table 2). The correlations between Blue Nile precipitation and Sahel rainfall, AEJ, and WWJ hold up at the 99% confidence level when extremes are removed from the dataset (Table 3) and are generally the same when the analysis is extended to the full 1950–2009 period.
c. July
The contrast between early and late rainy season teleconnections affecting Blue Nile precipitation is quite striking (e.g., Figs. 3, 4, 5, and 8). Where August and September show clear, significant correlations with SST, SLP, and wind features associated with major modes of climate variability (ENSO, PDO, and the Indian monsoon), it is much more difficult to identify significant teleconnections affecting precipitation variability either in July or in June. Nevertheless, these months account for approximately 50% of total interannual precipitation variability in the Blue Nile, so it is well worth exploring any remote drivers that can be identified in order to understand the nature of this variability and, potentially, enhance prediction systems.
In July, precipitation variability correlates most strongly with activity of the AEJ and WWJ. Correlations with the TEJ are significant with Ugt25 but not the longmax index (Table 2). As noted for August, above, correlations between precipitation and the AEJ may be a function of the jet’s suppression of vertical upward motion across the Sahel and Blue Nile regions (Yeshanew and Jury 2007). However, the dynamical mechanisms linking the AEJ and Blue Nile precipitation require further study. Variability in July Blue Nile precipitation is also similar to August precipitation variability in that it significantly correlates with Sahel rainfall, with correlation values of 0.44 and 0.55 for CRU and GPCC, respectively, and is similar to August and September rainfall in its significant correlation with Arabian Peninsula SLP: low Arabian SLP is associated with wet conditions in the Blue Nile. This association was also found for NCEP-R1 with GPCC, but it is weaker for ERA-40 for both CRU and GPCC (not shown). The significance level of the correlations of Blue Nile precipitation in this month with concurrent Sahel rainfall and WWJ decreases from 99% to 95% from the period 1951–97 to 1950–2009. GPCC shows correlations significant at the 99% confidence level with both indices over both time periods.
ERA-40 does indicate that there is some correlation with eastern Pacific Ocean SST in July, which is not seen in NCEP-R1 (Fig. 4), but the strength of this association is much weaker in July than in August and September. There is some evidence as well for correlation with SLP in the Maritime Continent and parts of the southern Pacific as well—for both CRU and GPCC with ERA-40 and NCEP-R1—reflecting the development of ENSO-like correlation patterns that emerge in August and September (Fig. 5).
The relationship between low pressure over the Arabian Peninsula and wet conditions in the Blue Nile in July is consistent with a mechanism of enhanced low-level westerlies across Africa that bring moist air into the East African highlands. At the monthly scale we find a statistically significant relationship between the observed Arabian SLP anomaly and the strength of westerly winds affecting vapor transport to the Blue Nile basin (Fig. 6b). The relatively weak association between ENSO and July Blue Nile precipitation compared to the stronger relationship with the AEJ and WWJ suggests that predictive models that take into account AEJ and WWJ predictors have the potential to provide more skill than ENSO-based models in this month.
July precipitation also shows significant correlation, +0.65 with CRU and +0.50 with GPCC, with the concurrent equatorial planetary wave index. Though the EPWI used in this analysis covers 19 years only, as the index is restricted to the period since the satellite era, the correlation values are significant at the 99% and 97% confidence levels for CRU and GPCC, respectively. When the data over the period 1979–2009 are considered, July EPWI gives significant correlations at the 99% confidence level, with both concurrent CRU and GPCC (not shown). Correlations with this relatively new index are intriguing and may warrant further investigation.
d. June
Of the 4 months of the rainy season, June shows the least evidence of correlation with traditionally used predictors (Table 2) and very limited association with large-scale SST and SLP anomalies in general (Figs. 4b, 5b). This month does show correlations with SLP in the equatorial Atlantic Ocean within about 20°S–20°N (Fig. 5b). This pattern is observed for ERA-40 as well.
While SLP in the equatorial Atlantic has been noted in earlier studies of East African precipitation, the proposed mechanism of association—that SLP anomalies over the equatorial Atlantic lead to enhanced westerlies across western and central Africa (Segele et al. 2009)—requires a positive correlation between SLP in this region and Blue Nile precipitation, since it is high pressure over the eastern Atlantic that would lead to an intensified SLP gradient between the Gulf of Guinea and the Arabian Peninsula monsoon trough. Here, we find that it is reduced SLP in the equatorial Atlantic Ocean that strengthens precipitation in the basin (Fig. 5b). This association was also found for GPCC precipitation (Table 2). The negative correlation is evident in all datasets considered in this study. It is somewhat stronger than shown in Fig. 5b when CRU or GPCC precipitation data are used with ERA-40 SLP, and it is weaker when GPCC data are used with NCEP-R1 SLP (not shown). One potential mechanism that is consistent with this correlation pattern is via equatorial Atlantic SLP associations with easterly and northeasterly winds and water vapor transport across North Africa and the Mediterranean and Red Seas. These associations are statistically significant in June, with low SLP in the equatorial Atlantic Ocean associated with strengthened easterly and northeasterly winds and water vapor transport. This circulation pattern includes northerly to northeasterly flow that enters northeast Africa from the eastern Mediterranean Sea and Red Sea. These winds may provide significant vapor flux into the Blue Nile basin (Viste and Sorteberg 2011) or simply contribute to enhanced convergence in the region. We note that differences between our results and those of Segele et al. (2009) might be attributed to different time scales of analysis—they considered the entire summer season in pentad composites, while we look at the monthly scale—different extent of study region, and differences in data sources and the time period of analysis.
The paucity of large-scale associations for precipitation in June relative to other months suggests that precipitation variability in this critical planting month could be particularly difficult to predict seasonally or to project for future climate conditions using statistical methods that employ standard SLP and SST indices. Interestingly, however, June precipitation shows very strong correlations with geopotential height anomalies across the tropical troposphere and lower stratosphere (Fig. 11). This correlation pattern is observed for both ERA-40 and NCEP-R1 when CRU precipitation data are used. The pattern is not as clear between NCEP-R1 and GPCC precipitation, though it is robust for ERA-40 and GPCC. The fact that the upper-level geopotential height anomalies also correlate with tropical Atlantic SLP and SST anomalies suggest that there is a connection between the pantropical geopotential height correlations and the tropical Atlantic influence on Blue Nile precipitation. This could be an instance of an atmospheric bridge teleconnection linking the Atlantic to the ENSO development zone, but the full mechanisms of the association require further investigation. The connection may be particularly relevant to predictions of early season Blue Nile precipitation, as correlations between upper-tropospheric geopotential height anomalies and June precipitation are highly significant with up to a 3-month lead time (March geopotential height and June precipitation).
Correlations of Blue Nile CRU June precipitation with NCEP-R1 geopotential height at various pressure levels for the period 1951–97. Only correlation coefficients exceeding the 90% confidence levels are shown.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
e. Nonstationarity
In 1976, the tropical climate system experienced a shift that included significant changes in the structure and evolution of ENSO (Graham 1994; Wang 1995) and climate regime changes in the Indian Ocean (Clark et al. 2000). After 1976, more El Niño and fewer La Niña events occurred than in previous years (Trenberth and Hoar 1996). The connection between Blue Nile precipitation and ENSO appears to be similarly nonstationary. As Fig. 12 shows, the link between tropical Pacific SST and Blue Nile precipitation diminished after 1976. This is also true for the link with Indian monsoon precipitation (Torrence and Webster 1999). The waxing and waning of the associations of the basin’s precipitation with different climatic indicators indicate that care must be taken in statistical analysis and prediction of rainfall in the basin. The correlations with detrended SST give similar results (Figs. 12b,d), which shows that the correlation shift is not an artifact of the secular warming trend. When data from 1977 to 2009 are employed, the correlation results are generally similar to Figs. 12c and 12d. In addition to the shifts in ENSO and Indian monsoon influence, we found differences between the 1951–75 period and the 1977–97 period in a number of the correlations identified in this paper. Associations between equatorial Atlantic SLP and June precipitation, for example, appear to strengthen in the later period, as did correlations between August precipitation and St. Helena SLP, the AMM, and the WWJ. Some differences are also observed in the significance of correlation patterns with Arabian SLP, the AEJ, and the EPWI between 1951–75 and 1977–97, though these differences are not as dramatic. The 1976 shift also has clear implications for the statistical predictability of precipitation. While ENSO associations have weakened, other correlations with significant lead time appear to have strengthened: for example, August precipitation shows significant correlation with the preceding September–November SST in the tropical Atlantic and northwestern Indian Ocean from 1977 to 1997, but not from 1951 to 1975.
Correlation of September Blue Nile CRU precipitation with concurrent SST (a),(b) from 1951 to 1975 and (c),(d) from 1977 to 1997. Panels (b) and (d) are with detrended SST. Only correlation coefficients exceeding the 90% confidence levels are shown.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
The 1976 transition does present a challenge for the analyses presented in this paper and for any study that employs statistical analysis of the historical observational record as a basis for exploring predictability and mechanisms of variability in tropical precipitation. On the one hand, use of the 1951–97 baseline results in a statistical analysis that is informed by both pre-1976 and post-1976 conditions and that may not be fully representative of present, post-1976 conditions. On the other, analyses limited to 1977–present are often too short to capture multiannual- to decadal-scale variability patterns. Also, insomuch as the 1976 shift may be a product of a multidecadal oscillation, an analysis that ignores the pre-1976 period is an incomplete representation of tropical variability. It is clear, however, that studies of past precipitation variability in the Blue Nile, and probably in surrounding regions as well, must consider the 1976 shift when interpreting results and applying statistical correlations to seasonal prediction or climate projection.
4. Conclusions
Many studies have documented that Ethiopian highland boreal summer precipitation is associated with ENSO, while others have noted significant correlations with the Indian summer monsoon. Still other studies have emphasized the role of the TEJ and AEJ or have focused on correlations with the sea level pressure gradient between the Arabian Peninsula and the Gulf of Guinea and associated westerlies and south westerlies that advect moisture to the Ethiopian highlands from the Atlantic Ocean and the Congo Basin.
This study examined the associations of monthly precipitation in the Blue Nile with large-scale patterns of SST, SLP, and winds, and with a range of climate indices that have been examined at seasonal scale in previous studies. We found that associations between large-scale atmospheric and SST fields and precipitation in the Blue Nile are different in each month of the rainy season. The frequently reported correlations with ENSO are most prominent in the second half of the rainy season, and associations with the Indian monsoon are significant at the cessation of rains. While September shows the strongest associations with eastern influences—ENSO and the Indian monsoon—in August there is evidence of an Atlantic influence as well, including strong associations with the St. Helena high and low-level westerlies and south westerlies from the Atlantic Ocean. July precipitation, in contrast, shows limited Pacific and Indian Ocean influence and instead correlates most strongly with the AEJ, WWJ, and SLP over the Arabian Peninsula. June precipitation also exhibits only weak association with ENSO. Instead, there is evidence of strong correlation with SLP patterns in the tropical Atlantic Ocean and pantropical geopotential height anomalies in the upper troposphere.
June and September contribute about 20% each of the summer rainfall in the Blue Nile basin while August and July account for about 30% each, but June and September have higher coefficients of variation than July and August. Droughts in the basin are usually caused by delayed onset and/or early cessation of rainfall. Analyses described in this paper indicate that the mechanisms that govern precipitation variability in the months of precipitation onset (June) and cessation (September) are largely distinct from each other. This strongly suggests that subseasonal and seasonal precipitation forecasts for the beginning, middle, and end of the summer rainy season in the Blue Nile require a different model structure. The results also suggest that projections of Blue Nile precipitation in a changing climate need to consider multiple mechanisms associated with Pacific, Indian, and Atlantic Ocean variability and change. Efforts to apply this information to forecasts and future climate projections will be of value for agricultural outlooks as well as for basin-scale hydrologic analysis in support of projects such as the Ethiopian Grand Renaissance Dam.
Acknowledgments
The authors thank Dr. Anand Gnanadesikan for useful discussion. A portion of this study was supported by NASA Applied Sciences Grant NNX09AT61G.
APPENDIX
Number of Stations CRU TS 3.10.01 Uses for Interpolation in Each Month in the Blue Nile Basin
Number of stations CRU TS 3.10.01 uses for interpolation in each month from 1950 to 2009 in the Blue Nile basin.
Citation: Journal of Climate 27, 1; 10.1175/JCLI-D-13-00094.1
REFERENCES
Abtew, W., A. M. Melesse, and T. Dessalegne, 2009: El Niño Southern Oscillation link to the Blue Nile River basin hydrology. Hydrol. Processes,23, 3653–3660, doi:10.1002/hyp.7367.
Beyene, T., D. P. Lettenmaier, and P. Kabat, 2010: Hydrologic impacts of climate change on the Nile River basin: Implications of the 2007 IPCC scenarios. Climatic Change, 100, 433–461.
Block, P., and B. Rajagopalan, 2007: Interannual variability and ensemble forecast of upper Blue Nile basin Kiremt season precipitation. J. Hydrometeor., 8, 327–343.
Camberlin, P., 1995: June–September rainfall in north-eastern Africa and atmospheric signals over the tropics: A zonal perspective. Int. J. Climatol., 15, 773–783.
Camberlin, P., 1997: Rainfall anomalies in the source region of the Nile and their connection with the Indian summer monsoon. J. Climate, 10, 1380–1392.
Cameron, A., and P. Trivedi, 1998: Regression Analysis of Count Data. Cambridge University Press, 436 pp.
Chiang, J. C. H., and D. J. Vimont, 2004: Analogous Pacific and Atlantic meridional modes of tropical atmosphere–ocean variability. J. Climate, 17, 4143–4158.
Clark, C. O., J. E. Cole, and P. J. Webster, 2000: Indian Ocean SST and Indian summer rainfall: Predictive relationships and their decadal variability. J. Climate, 13, 2503–2519.
Conway, D., 1997: A water balance model of the upper Blue Nile in Ethiopia. Hydrol. Sci. J., 42, 265–286.
Conway, D., 2000: The climate and hydrology of the upper Blue Nile River. Geogr. J., 166, 49–62.
Conway, D., and M. Hulme, 1993: Recent fluctuations in precipitation and runoff over the Nile sub-basins and their impact on main Nile discharge. Climatic Change, 25, 127–151.
Diro, G. T., D. I. F. Grimes, and E. Black, 2011: Teleconnections between Ethiopian summer rainfall and sea surface temperature: Part I. Observation and modelling. Climate Dyn., 37, 103–119.
Flohn, H., 1987: Rainfall teleconnections in northern and northeastern Africa. Theor. Appl. Climatol., 38, 191–197.
Gissila, T., E. Black, D. I. F. Grimes, and J. M. Slingo, 2004: Seasonal forecasting of the Ethiopian summer rains. Int. J. Climatol., 24, 1345–1358.
Goswami, B. N., and P. K. Xavier, 2005: ENSO control on the south Asian monsoon through the length of the rainy season. J. Geophys. Lett., 32, L18717, doi:10.1029/2005GL023216.
Graham, N. E., 1994: Decadal-scale climate variability in the tropical and North Pacific during the 1970s and 1980s: Observations and model results. Climate Dyn., 10, 135–162.
Grise, K. M., and D. W. J. Thompson, 2012: Equatorial planetary waves and their signature in atmospheric variability. J. Atmos. Sci., 69, 857–874.
Grist, J., and S. Nicholson, 2001: A study of the dynamic factors influencing the rainfall variability in the West African Sahel. J. Climate, 14, 1337–1359.
Harris, I., P. D. Jones, T. J. Osborn, and D. H. Lister, 2013: Updated higher resolution grids of monthly climatic observations—The CRU TS3.10 dataset. Int. J. Climatol., doi:10.1002/joc.3711, in press.
Janowiak, J. E., 1988: An investigation of interannual rainfall variability in Africa. J. Climate, 1, 240–255.
Jiang, X., T. Li, and B. Wang, 2004: Structures and mechanisms of the northward propagating boreal summer intraseasonal oscillation. J. Climate, 17, 1022–1039.
Joseph, P. V., and J. Srinivasan, 1999: Rossby waves in May and the Indian summer monsoon rainfall. Tellus, 51A, 854–864.
Jury, M. R., 2010: Ethiopian decadal climate variability. Theor. Appl. Climatol., 101, 29–40.
Korecha, D., and A. G. Barnston, 2007: Predictability of June–September rainfall in Ethiopia. Mon. Wea. Rev., 135, 628–650.
Krauer, J., 1988: Rainfall, Erosivity, and Isoerodent Map of Ethiopia. University of Berne, 132 pp.
Levin, N. E., E. J. Zipser, and T. E. Cerling, 2009: Isotopic composition of waters from Ethiopia and Kenya: Insights into moisture sources for eastern Africa. J. Geophys. Res., 114, D23306, doi:10.1029/2009JD012166.
Mantua, N. J., S. R. Hare, Y. Zhang, J. M. Wallace, and R. C. Francis, 1997: A Pacific interdecadal climate oscillation with impacts on salmon production. Bull. Amer. Meteor. Soc., 78, 1069–1080.
Mohamed, Y. A., B. Van Den Hurk, H. H. G. Savenije, and W. G. M. Bastiaanssen, 2005: Hydroclimatology of the Nile: Results from a regional climate model. Hydrol. Earth Syst. Sci. Discuss., 2, 319–364.
Nicholson, S. E., 2009: A revised picture of the structure of the monsoon and land ITCZ over West Africa. Climate Dyn., 32, 1155–1171.
Nicholson, S. E., and J. Grist, 2003: The seasonal evolution of the atmospheric circulation over West Africa and equatorial Africa. J. Climate, 16, 1013–1030.
Pai, D. S., J. Bhate, O. P. Sreejith, and H. R. Hatwar, 2011: Impact of MJO on the intraseasonal variation of summer monsoon rainfall over India. Climate Dyn.,36, 41–55.
Pu, B., and K. H. Cook, 2012: Role of the West African westerly jet in Sahel rainfall variations. J. Climate, 25, 2880–2896.
Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res., 108, 4407, doi:10.1029/2002JD002670.
Schneider, U., A. Becker, A. Meyer-Christoffer, M. Ziese, and B. Rudolf, 2011: Global precipitation analysis products of the GPCC. Global Precipitation Climatology Centre/DWD Rep., 1–13. [Available online at ftp://ftp.dwd.de/pub/data/gpcc/PDF/GPCC_intro_products_v2011.pdf.]
Segele, Z. T., and P. J. Lamb, 2005: Characterization and variability of Kiremt rainy season over Ethiopia. Meteor. Atmos. Phys., 89, 153–180.
Segele, Z. T., P. J. Lamb, and L. M. Leslie, 2009: Large scale atmospheric circulation and global sea surface temperature associations with Horn of Africa June–September rainfall. Int. J. Climatol., 29, 1075–1100.
Seleshi, Y., and G. Demaree, 1995: Rainfall variability in the Ethiopian and Eritrean highlands and its link with the Southern Oscillation index. J. Biogeogr., 22, 945–952.
Seleshi, Y., and U. Zanke, 2004: Recent changes in rainfall and rainy days in Ethiopia. Int. J. Climatol., 24, 973–983.
Shaman, J., and E. Tziperman, 2007: Summer time ENSO–North African–Asian jet teleconnection and implications for the Indian monsoons. Geophys. Res. Lett., 34, L11702, doi:10.1029/2006GL029143.
Tadesse, T., 1994: Summer monsoon seasonal rainfall of Ethiopia in ENSO episodic years. Proc. WMO/TOGA Int. Conf. on Monsoon Variability and Prediction, Trieste, Italy, International Centre for Theoretical Physics/World Meteorological Organization, 48–55.
Terray, P., P. Delecluse, S. Labattu, and L. Terray, 2003: Sea surface temperature associations with the late Indian summer monsoon. Climate Dyn., 21, 593–618.
Torrence, C., and P. J. Webster, 1999: Interdecadal changes in the ENSO–monsoon system. J. Climate, 12, 2679–2690.
Trenberth, K. E., and T. J. Hoar, 1996: The 1990–1995 El Niño–Southern Oscillation event: Longest on record. Geophys. Res. Lett., 23, 57–60.
Uppala, S. M., and Coauthors, 2005: The ERA-40 Reanalysis. Quart. J. Roy. Meteor. Soc., 131, 2961–3012.
Viste, E., and A. Sorteberg, 2011: Moisture transport into the Ethiopian highlands. Int. J. Climatol., 33, 973–983.
Vizy, E. K., and K. H. Cook, 2003: Connections between the summer east African and Indian rainfall regimes. J. Geophys. Res., 108, 4510, doi:10.1029/2003JD003452.
Walker, G. T., 1910: Correlation in seasonal variations of weather, II. Indian Meteor. Mem., 21, 1–21.
Walker, G. T., and E. W. Bliss, 1932: World weather V. Mem. Roy. Meteor. Soc., 4, 53–84.
Wang, B., 1995: Interdecadal changes in El Niño onset in the last four decades. J. Climate, 8, 267–285.
Wang, B., and Z. Fan, 1999: Choice of South Asian summer monsoon indices. Bull. Amer. Meteor. Soc., 80, 629–638.
Wolter, K., and M. S. Timlin, 1993: Monitoring ENSO in COADS with a seasonally adjusted principal component index. Proc. 17th Climatic Diagnostics Workshop, Norman, OK, NOAA/NMC/CAC, 52–57.
Yeshanew, A., and M. R. Jury, 2007: North African climate variability. Part 2: Tropical circulation systems. Theor. Appl. Climatol., 89, 37–49.
