Water Resources Research

Temporal variability of hydroclimatic extremes in the Blue Nile basin



[1] This paper examines the long-term historical changes in frequency and amplitude of hydroclimatic extremes in the Blue Nile basin using data from the second half of 20th century. The temporal variability of basin-wide rainfall extremes and river flow extremes from four gauging stations was investigated under the hypothesis of no trend and no persistence in time. On the basis of a quantile anomaly analysis method, decadal variations in extreme daily, monthly, and annual quantiles were studied, and the periods of statistical significance were identified. The analysis showed that high and low river flows and rainfall depths do not vary in time in a fully random way but show a particular variation pattern. Their extremes show significant decadal variations. The 1980s had statistically significant negative anomalies in extremes in comparison with the long-term reference period of 1964–2009, while the 1960s–1970s and the 1990s–2000s had positive anomalies, although less significant. There is neither consistent increasing nor decreasing trend in rainfall and flow extremes of recent years. Therefore, anticipated trends due to global warming could not be identified. Conversely, low-flow extremes show an increasing trend during the last decade, which could be related to the effect of water regulation works at the outlet of Lake Tana. Moreover, similar patterns and statistically significant correlations were found between climatic indices representing the Pacific and Atlantic Oceans and the Blue Nile rainfall and flow extremes. Changes that occur on the Pacific Ocean appear to be a main driver for the decadal oscillations in climate and related high and low Blue Nile water availability for Ethiopia, Sudan, and Egypt.

1. Introduction

[2] The upper Blue Nile basin is one of the important river basins in Africa because of its large contribution to the Nile River. It contributes about 60% of the Nile's flow at Aswan, Egypt [Yates and Strzepek, 1998; Conway, 2005; Sutcliffe and Parks, 1999], even though the Blue Nile comprises only about 8% of the total Nile catchment area [Gebrehiwot et al., 2011]. The Nile River is the only significant water resource and its availability is a matter of survival for Egypt and Sudan. Equally, the upper Blue Nile basin is the largest and economically imperative water resource for Ethiopia. Ethiopia is planning irrigation and hydropower projects using the Blue Nile River [Tesemma et al., 2010] while all the other riparian countries are working on increasing their share of the water to boost their economic developments. Such developmental projects require equitable transboundary water management to avoid possible conflicts between the riparian countries. Since the Ethiopian Highlands are the major water contributors to the Blue Nile River Basin, reliable runoff information from this region is of great importance in the sustainable management of water resources [Kim and Kaluarachchi, 2008]. Extreme flow conditions from this region are also imperative for developmental projects within the basin and for the downstream users.

[3] The Blue Nile basin experienced recurrent extreme flow conditions, both floods and droughts, in the past few decades. For example, there was a devastating drought in the early 1980s in most parts of the Ethiopian highlands that consequently affected the Blue Nile flow; while Sudan experienced severe flooding during August–September 1988 [Sutcliffe et al., 1989]. Heavy rainfall events over Khartoum and Atbara were responsible for this kind of flooding in Sudan [Sutcliffe et al., 1989], whereas insufficient total rainfall amounts and/or long dry spells within a rainy season were responsible for the devastating consequences of droughts in Ethiopia [Seleshi and Camberlin, 2005], which were also observed as extreme low-flow events in the Blue Nile River. Most recent events in the Greater Horn of Africa include a prolonged drought that ended in 2005 and a severe flooding in August 2006 [Nawaz et al., 2010]. Climate variability triggered by large scale atmospheric circulations and/or by anthropogenic activities could be a major influencing factor in the recurrence of extreme conditions and as well contributes to changes that occur in the river's flow regime. Nevertheless, not only climatic factors but also other issues such as land use change, changes in water abstraction and watershed management practices alter the hydrological characteristics of a river basin and consequently the extreme flows. Melesse et al. [2010] states that the upper Blue Nile basin experienced environmental and natural resources degradation attributed to both anthropogenic and climate factors. To understand the influence of these factors in the river's extreme flows, it is important to look back to historical records and assess how variable the extreme flows were temporally. Studying temporal variability of extreme flows will also be useful to understand whether extreme events become more frequent or intense in the recent years.

[4] Climate variation can be evaluated from directly measured hydroclimatic data such as rainfall, streamflow, stream water level, groundwater level, and air and water temperature [Abtew et al., 2009]. Through statistical analysis of time series of such data, patterns manifested as trends and oscillation patterns can be detected and used to study the variability and availability of water. Among hydroclimatic records, analyzing river flow is useful as it gives the entire picture of a catchment's response. Previous studies on long-term trends of rainfall or flow in the upper Blue Nile basin concentrated on the general or mean state of the variables. To mention a few examples, Conway [2000] used linear regression to detect trends on annual and seasonal rainfall from 11 gauges within the Blue Nile basin as well as the Blue Nile flow. Tesemma et al. [2010] used Mann-Kendal and Sen's t tests to detect trends in both seasonal and annual runoff and rainfall. Bushara and Abdelrahim [2010] analyzed the absence of step trends on annual flow time series of four key stations in the Nile basin, one of which represents the upper Blue Nile basin, using the Pettit test.

[5] While most of these previous studies concentrated on total volumes, this paper focuses on hydrological extremes which are imperative in water management projects. In addition, usually applied trend detection methods were used by previous studies including Mann-Kendall test, Sen's t test, Pettit test, Spearman's rho test, linear regression test, and Student's t test. However, in this paper the temporal variability of extreme high and low flows and rainfall is studied using an empirical statistical analysis known as the quantile perturbation method (QPM). This method was proposed by Ntegeka and Willems [2008] to study trends and multidecadal oscillations in rainfall extremes. The method compounds two concepts, first, the frequency aspect that focuses on how often an event (a quantile) may occur and, second, the perturbation aspect that determines the relative magnitudes of events on the basis of a certain baseline, thereby providing possibility to examine changes in the extremes for a particular return period [Ntegeka and Willems, 2008]. The approach offers an insightful temporal evaluation of trends and oscillation patterns of extremes, where the pattern through time can be viewed using graphical plots. Ntegeka and Willems [2008] commented on the performance of the QPM method relative to other approaches. They stated that while previous studies that conducted Fourier and wavelet analysis on the same data set as they used did not find clear evidence of oscillating patterns on extreme rainfall, the QPM method was able to detect clustering of extremes in different historical subperiods. In light of this, the QPM method was chosen for this paper. Previous studies [e.g., Jury, 2011] used wavelet spectral analysis to investigate the nature of climatic oscillations in monthly flow and rainfall data in the Blue Nile basin but did not focus on the analysis of extremes.

[6] The study of temporal changes in extreme events identifies anomalies which can be attributed to different phenomena [Ntegeka and Willems, 2008]. Hence, observed changes in extreme flow variables might be explained with extreme rainfall events over the basin, which on its own is influenced by other factors such as large scale atmospheric circulations. It is therefore important to examine relationships between climate indices and their impacts on rainfall and flow extremes. This was achieved by selecting indices that represent changes at different oceanic regions associated with anomalous rainfall over the basin and investigating their influence on the hydroclimatic extremes.

2. Study Area and Data

2.1. Study Area

[7] The upper Blue Nile basin is located in the northwestern part of Ethiopia, with an approximate area of 176,000 km2. It is located between 7°45′ and 12°45′N latitude and between 34°30′ and 39°45′E longitude (Figure 1). It covers about 17% of the Ethiopian land area, 43% of the country's total average annual runoff and 25% of the population of Ethiopia [Zerfu and Moges, 2009]. Greater part of the basin lies in the highlands of Ethiopia and receives high rainfall amounts ranging between 800 and 2200 mm [Melesse et al., 2010]. The elevation of the basin varies from over 4000 m in the headwaters to about 500 m downstream (Figure 1). The Blue Nile leaves Lake Tana at Bahirdar and flows to the southeast through a series of cataracts (sections where the river tumbles over rocks). The river then enters a canyon and changes direction to the south, then to the west and finally to the northwest forming a large open loop. Along its 940 km journey from Bahirdar to El Diem, near the Ethiopian Sudanese border, the river is joined by several tributaries draining a large area of highlands in western Ethiopia [Elshamy et al., 2009]. Most of the streams that feed the Blue Nile are perennial [Tesemma et al., 2010].

Figure 1.

Location of Upper Blue Nile basin and rainfall and river flow gauging stations used in the study.

[8] The Blue Nile basin has three seasons, namely, the short rainy season (March to May, MAM), the main rainy season (June to September, JJAS) and the dry season (October to February, ONDJF). The rainfall within the basin shows high seasonality. This seasonal distribution of rainfall is controlled by the northward and southward movement of the Intertropical Convergence Zone (ITCZ) [Nawaz et al., 2010]. The rainfall peaks in July while the flow at El Diem reaches its maximum in August (Figure 2).

Figure 2.

Comparison between long-term mean monthly basin-wide rainfall (1964–2004) and runoff at El Diem (1964–2009) in the upper Blue Nile basin.

2.2. Data

2.2.1. Hydrological

[9] Despite the fact that the Blue Nile basin contributes the largest volume to the Nile River, it has limited hydroclimatic data that confines detailed analysis in the basin. River flow data are limited because of the remoteness of many of the catchments, the lack of economic resources and infrastructure to build and maintain monitoring sites [Conway, 2000]. The locations of the stations used in this study are as shown in Figure 1, and the availability of data for the flow gauging stations is given in Table 1.

Table 1. Data Availability of Four Flow Gauging Stations Used in the Study, Including Their Upstream Catchment Areas
StationCatchment Area (km2)Daily Data LengthMonthly Data LengthAnnual Data Length
El Diem176,0001964–20091964–20091964–2009
Gilgel Abay16641973–20051964–20051964–2005

[10] Among the four flow gauging stations, El Diem station which is located at the border between Ethiopia and Sudan has the most reliable and longer time series. Data from this station is reported as of good quality [Conway, 1997, 2000]. Some of the missing gaps found in the series were completed using a linear interpolation technique based on daily data available at Roseires (a station that is located around 100 kilometers downstream of El Diem). The flow at El Diem has an average runoff of 1500 m3 s−1 on the basis of the period 1964–2009 records. While the annual variability is around 20%, the seasonal variation is considerable (Figure 2).

[11] Streamflow measurement in Ethiopia spans the past five decades to the maximum length [United Nations Educational, Scientific, and Cultural Organization, 2004]. Hence, the data used in this paper are among the longest records. Discharge measurements are monitored by the hydrology department of the Ministry of Water and Energy, Ethiopia, on the basis of water level measurements (daily at 6:00 A.M. and 6:00 P.M.) using an occasionally updated rating curve of the stations [Uhlenbrook et al., 2010; Gebrehiwot et al., 2011]. According to the Ministry of Water and Energy, the measurements of river levels follow the guidelines of the World Metrological Organization (WMO). The monthly database contains maximum and minimum discharges of each month in addition to the monthly runoff, which assisted to construct the extremes time series. Therefore, the following hydrological variables were analyzed in this study: annual maximum (minimum) daily flow, monthly maximum (minimum) daily flow including annual, seasonal and monthly total flow.

2.2.2. Rainfall

[12] The areal rainfall was calculated using Thiessen polygon method applied to 11 stations with relatively long-term data (at least 30–40 years of data). Missing gaps in these stations were completed using inverse distance weighting method from neighboring stations. In order to relate extreme high-flow conditions with extreme rainfall events, peak over threshold (POT) values were extracted from the areal rainfall time series. Different thresholds were considered for the three seasons. The selection of these thresholds was based on average monthly precipitations for the period 1964–2004. The exceedance probability of precipitation values from each season was calculated and the value which is exceeded 5% of the time was considered as a threshold for that season. Similarly, in order to relate extreme low-flow conditions the number of dry days in each month was used.

2.2.3. Climate Indices

[13] Many studies attempted to explain climate variability over the East African region and also in relation to the Nile River. Studies that attempted to find a link between atmospheric circulations and the Nile river flow used the effect of Pacific Ocean's sea surface temperature (SST) changes. For instance, Eltahir [1996] studied the link between Nile annual natural flow and sea surface temperature of the Pacific Ocean (El Nino–Southern Oscillation (ENSO) index). The result reveals that 25% of the natural variability in the annual flow of the Nile is associated with the El Nino oscillations. Amarasekera et al. [1997] further extended the study and found that two major tributaries, Blue Nile and Atbara, show significant negative correlation with Pacific Ocean's SST and that the ENSO signature observed in the main Nile is entirely dependent on these rivers. Conway [2000] showed strong positive correlations between the Southern Oscillation Index (SOI) and annual rainfall totals estimated as basin-wide rainfall. Abtew et al. [2009] also stated that extreme dry and wet years are likely to correspond with ENSO events. Some additional studies that showed the influence of ENSO on Ethiopian rainfall include those of Beltrando and Camberlin [1993], Seleshi and Demarée [1995], Seleshi and Zanke [2004], and Diro et al. [2010].

[14] Other climate indices were also used by previous studies to explain rainfall variability over Ethiopia. For instance, Camberlin [1996], in his study concluded that monsoon activity over India is a major trigger for July–September rainfall variability in the East African highlands. Jury [2010] found that the rainfall mode in the northern part of Ethiopia that impacts Nile River flow is linked with the Atlantic zonal overturning circulation and exhibits 10–12 years cycles through much of the 20th century.

[15] From the climate indices that have been used to explain anomalous rainfall over Ethiopia, four indices were selected for this study and the data analyzed is for the period 1964–2009. These indices represent influences from three oceans: Pacific, Indian, and Atlantic. The indices are as follows.

[16] 1. For the Pacific Decadal Oscillation (PDO), the index is calculated by spatially averaging the monthly sea surface temperature (SST) of the Pacific Ocean north of 20°N [Mantua, 2001]. The data were downloaded from the database of the Joint Institute for the Study of the Atmosphere and Ocean (JISAO) (http://jisao.washington.edu/pdo/PDO.latest).

[17] 2. For the Southern Oscillation Index (SOI), The data were downloaded from the Climate Research Unit (CRU) database (http://www.cru.uea.ac.uk/cru/data/soi/). This data set is calculated on the basis of the method given by Ropelewski and Jones [1987]. The SOI is defined as the normalized atmospheric pressure difference between Tahiti and Darwin.

[18] 3. The Indian Ocean Dipole (IOD) is defined as the difference between the sea surface temperatures (SST) in the western (50°E to 70°E and 10°S to 10°N) and eastern (90°E to 110°E and 10°S to 0°S) equatorial Indian Ocean. IOD data were downloaded from the Japan Agency for Marine-Earth Science and Technology (JAMEST) (http://www.jamstec.go.jp/frsgc/research/d1/iod/).

[19] 4. For the Atlantic Multidecadal Oscillation (AMO), the index is defined as the SST averaged over 0°–60°N, 0°–80°W minus SST averaged over 60°S–60°N [van Oldenborgh et al., 2009]. The data were obtained from http://climexp.knmi.nl/data/iamo_hadsst2.dat.

3. Methodology

3.1. Temporal Variability Calculation

[20] Temporal variability of extreme flows and rainfall during the past few decades in the upper Blue Nile basin was studied using empirical-statistical analysis based on a quantile anomaly analysis method. The analysis is conducted starting from the prior hypothesis that there is no persistence (trend) in the temporal climate variation of extremes. The analysis attempts to find persistence patterns (significant trends in extremes) that might exist in time scale of several years (e.g., (multi)decadal climate oscillations). It makes use of the quantile mapping technique that is applied in some climate change impact studies. For instance, Harrold and Jones [2003], Harrold et al. [2005], Chiew [2006], and Willems and Vrac [2011] used the quantile mapping method to scale ranked historical daily rainfall data (rainfall quantile) and construct future climate change scenarios in a quantile-dependent method.

[21] In the current study, the quantile mapping method is applied only on historical data, following a similar approach to that of Ntegeka and Willems [2008], who called it the quantile perturbation method (QPM). It calculates the change factors (anomalies) in extreme value quantiles between a long-term baseline period and a selected subperiod in the past. The baseline period is the entire observed time series available for that specific station (in our case, the longest flow record was 46 years). The selected subperiod (hereafter called block period) is a subseries taken from the total time series representing the period of interest. For this particular study, block periods of 5, 7, and 10 years were used along with a sliding window of one year until the end of the time series. For each period, extremes were extracted from the time series. After sorting in a descending order, these extremes can be associated with empirical return periods or exceedance probabilities, hence considered as empirical quantiles. For given quantiles or corresponding empirical return periods, change factors (anomalies) can then be calculated as the ratio of the quantile in the block period to the corresponding quantile (same empirical return period) in the baseline period.

[22] Let X1≥ X2≥ X3≥ … ≥ Xn be the extreme values extracted from the baseline period, where n is the total number of extremes and Y1Y2≥ Y3≥ … ≥ Ym be the m extreme values in a block period time series. For any length of block period (b) the empirical return period (Rb) of extreme value Yk will be b/k. Similarly, for the total length of the baseline period (t), the empirical return period (Rt) of extreme value Xi will be t/i. Consequently, the extreme values in the block period correspond with quantiles Y(b/k) and for baseline time series the extreme values correspond with quantiles X(t/i). Here it is important to note that other formulations exist for defining empirical return period; for instance, Weibull or Cunnane formulas. However, analysis (not shown) has indicated that these other formulations did not lead to significant differences in the results.

[23] The next step requires comparison of similar quantiles for the two series. The return periods of the block series, b/k, are used to estimate the quantiles from the long-term series. In this way, quantiles with the same return periods can then be compared. It is possible to linearly interpolate for the return periods b/k from the return period-intensity relation for the long-term series. Afterward, the relative change is calculated as a ratio between the block period and the reference period extremes as inline image.

[24] This last step ensures the comparison of quantiles correspond to the same return period. For instance, see the explanation for the first quantile of a block period. Initially select a block period of interest, e.g., b = 10 years and a baseline period t = 50 years. The first quantile from the block period series will have an empirical return period of 10 years. Consequently, the relative change will be the ratio of the first quantile in the block period to the baseline extreme event with 10 years recurrence interval (obtained from the 50 years data). Similarly, within each block, the same procedure is conducted for all the quantiles beginning from the first empirical return period (e.g., 10 years) downward.

[25] Then the relative changes of the most extreme events have been averaged for each block period, and is hereinafter called “anomaly” in extreme quantiles. The criterion used to define the most extreme events is the one that selects at most three extreme measurements per year. For example, for a 10 year block period, 30 extremes would be selected. To illustrate the choice of most extreme events, monthly extreme high flows of El Diem station is used. Figure 3 shows the calculated anomaly for four different decades. If the anomaly values versus the return periods are plotted on the same graph for these four periods, it is possible to see a rather clear change of slope toward the higher anomaly factors (the circled points). The point where approximately the slope changes equals the maximum of 30 values which corresponds with the selected 3 values per year criteria.

Figure 3.

Illustration of change of slope on the anomaly values for four different periods.

[26] Thus, using the described method each block period will have one anomaly value representing the variation in extreme quantiles in the block period. Using the sliding window, it is possible to compute the anomalies for several block periods, which generate a temporal variation of the anomaly. These anomalies are transformed to percentage changes and plotted against the middle year of each block period for visualization (Figure 4). Although the sliding window approach introduces some autocorrelation and the anomalies are dependent, it is possible to investigate whether the same anomalous periods are persistent by using different window sizes when selecting the block periods. This analysis can be done in all time scales: daily, monthly and annual. In case of extreme low flows, the same procedure is used, but keeping in mind that the lowest flow has the highest return period.

Figure 4.

(a) Annual maxima and (b) wettest month flow oscillation patterns of Blue Nile flow at El Diem station for a block length of 5 years and the period 1964–2009.

3.2. Monte Carlo Confidence Interval

[27] Subsequent to the temporal variation computation, it was necessary to know the statistical significance of the variation. This was determined by using confidence intervals on the anomaly values. The confidence intervals were computed under the null hypothesis of no temporal dependency, thus under the assumption that the observed temporal variability was caused by only natural variability or randomness. The nonparametric bootstrapping method was applied to calculate these intervals. The set of extreme values from the baseline period was resampled to generate random samples for each block period. The procedure is as follows.

[28] 1. The original set of extreme values (full baseline period) are randomly shuffled to generate a new temporal sequence of extreme values (same time moments as in original series, but values interchanged).

[29] 2. The series is then subdivided into subseries, each referring a block period.

[30] 3. The QPM method is applied to the shuffled series which leads to a new temporal variation of anomalies.

[31] 4. Steps 1, 2 and 3 are then repeated 1000 times, which leads to 1000 anomaly factors for each subseries. This number of Monte Carlo runs was based on an evaluation that showed a higher number of runs did not lead to a different conclusion.

[32] 5. The 95% confidence interval is then estimated for each time moment by using the 975th and 25th ranked anomalies for each subseries.

[33] The estimated confidence intervals were superimposed on the same plot as the percentage change versus the middle year of each block period to identify periods that depict significant deviation from the zero anomaly value. The anomaly values which are outside the confidence interval, defined around the zero anomaly of the null hypothesis, were defined as statistically significant. This means that the likelihood that these anomalies are explained by randomness, or that there is no temporal dependency in the extremes, is very low.

[34] The anomaly calculation was done for the total volumes of flow (annual and seasonal) besides extremes in terms of annual and monthly maximum and minimum flows. A similar kind of analysis was performed for areal rainfall extremes. It was, therefore, possible to identify the temporal variation of different hydroclimatic variables with the method described above. Similarity and differences between hydroclimatic variables could be checked visually by overlaying one pattern over the other.

3.3. Correlation Analysis

[35] It is clear that the temporal variability in rainfall and river flows and related water availability has strong implications for water and agricultural management. Questions remain on what the driver(s) of the oscillations are or whether predictions can be made on oscillation highs and lows based on such drivers. To partially address this issue and identify the causes of variability in the extremes, the QPM analysis was performed on the selected climate indices and correlation analysis was conducted between the rainfall and flow anomaly factors and that of the climate indices. The statistical significance of the correlation was determined using the t test for significance levels of 1% and 5%. The assumption made here is that each nonoverlapping block delivers one independent anomaly value, to account for the dependency introduced while using the moving average method in the QPM analysis. The sample size then equals the total number of independent anomaly values (the nonoverlapping blocks).

4. Results

4.1. Temporal Variability of Extreme Flows

4.1.1. High Flows

[36] The temporal variability of extreme high flows in the Blue Nile River, measured at El Diem, shows an oscillating pattern. From the temporal variability calculated for a block period of 5 years and the period 1964–2009, it is apparent that there are four periods with high and low oscillations (Figure 4). For the purpose of identifying similarities and differences among oscillation patterns of different hydroclimatic extremes, these four periods are used.

[37] Primarily, the sensitivity of the results to block lengths of 5, 7 and 10 years for annual maxima and wettest month flow extremes was examined. It is noted that when the block length increases, the variability decreases. Consequently, only base periods 1 and 2 (as defined in Figure 4) are clearly visible for all the block lengths. While base periods 1 and 3 are those with noticeably positive anomalies in flow extremes; these anomalies are not significant. Base period 2, the 1980s, has significantly negative anomalies in comparison with the baseline period 1964–2009 (up to 15%) for each block length and in both annual maxima and wettest month flow anomalies. The latter anomalies are more significant for the higher block lengths.

[38] For the three seasons, the temporal variability of extremes was studied using daily extremes extracted from each month. Figure 5 shows the quantile changes of seasonal high flows for block length of 5 years. First, the long rainy season (JJAS) has much less variability than the two other seasons. Second, the oscillation pattern is very much similar to the previous patterns observed in the annual maxima and wettest month analyses. The confidence interval bounds are, however, wider and the anomalies insignificant. The negative anomaly of extreme daily flows in the 1980s, thus, become statistically insignificant when extreme daily flows of only the JJAS season are considered.

Figure 5.

Temporal variability of seasonal maximum flow of Blue Nile at El Diem station for three seasons and a block length of 5 years.

[39] The short rainy season (MAM) and the dry season (ONDJF) have high temporal variability, for the short rainy season after the 1980s there is a general increasing trend with anomalies of up to 50% in the 2000s. The anomalies in this recent period are (nearly) significant depending on the block length used. The increasing trend has a first peak on 1997/98, which is also visible in the dry season but without increasing trend. This same peak also explains the positive anomaly in base period 3 (Figure 4). It corresponds to what is referred to by Conway [2000] as the wettest on record of the dry season in 1997/98 which was responsible for flooding across Ethiopia and also parts of Somalia and Kenya. The short rainy season increasing trend was not visible during the long rainy season.

4.1.2. Low Flows

[40] Low-flow extremes were analyzed in the form of annual minima of daily flows and the driest month flows. These extremes have a significant increasing trend in the recent years (Figure 6). The pattern of the increasing trend is different from the patterns observed in the extreme high-flow analysis and the observed percentage change is also much higher. A closer look at the period before the mid-1990s reveals two low-oscillation periods and one high-oscillation period (Figure 6b). This pattern is somewhat similar to the pattern for the extreme high flows in the MAM and ONDJF seasons. The low oscillations are statistically significant, with anomalies up to 30%. The driest month flow shows a pattern similar to the wettest month and annual maximum flows. The base period 2 of the 1980s has a significant negative anomaly similar to the extreme high flows for all the three block sizes (not shown).

Figure 6.

Annual minima oscillation pattern of Blue Nile flow at El Diem station for a block length of 5 years and the periods (a) 1964–2009 and (b) 1964–1996.

[41] Seasonal analysis of extreme daily low flows does not show periods with statistically significant anomalies for the long rainy season (JJAS). However, the low flows in the recent years are slightly higher than in previous periods. Similar patterns are observed for the MAM and ONDJF seasons before the mid-1990s. There are two distinct low-oscillation periods and one high-oscillation period that are similar to the annual minima series before the mid-1990s (Figure 6b). The period after the mid-1990s shows increasing trends for both seasons. The change is statistically significant for the MAM season but insignificant for the ONDJF season (not shown).

4.2. Temporal Variability of Total Volumes

[42] Albeit the focus of the paper is on hydroclimatic extremes, it is worth checking the total volumes on annual and seasonal basis and to compare with the results obtained for the flow extremes. Similar oscillation patterns are found with the oscillations' highs and lows in the same four base periods. One similarity in these total volumes is observed in the 1980s where the anomaly is negatively significant. After the 1980s until the beginning of the 21st century, the anomaly factors showed an increasing trend; while in the most recent decade both decreasing and increasing patterns are observed in the total flow volumes. Greater pattern similarity is observed between the main rainy season (JJAS) flows and the annual flow. The JJAS flow season has indeed the largest contribution to the Blue Nile total flow. The strong similarity between the anomaly patterns of the flow extremes and annual volumes illustrates that patterns of flow extremes can also be indirectly derived from annual data. This is useful in cases where only total flow data, at annual and monthly scales, are available.

[43] The anomaly pattern in the El Diem flows were corroborated by other records of hydroclimatic variables and by other (more upstream) stations in the Blue Nile basin. For illustration, three additional stations upstream of El Diem were analyzed. Figure 7 shows the overlay of the temporal quantile anomaly variation of the results over that of El Diem station. Similar total flow anomaly patterns are found for the Blue Nile flow measurements at El Diem, Bahirdar and Kessie. This similarity gives more confidence on the pattern found for El Diem station, given the high uncertainties (wide confidence intervals) on the calculated quantile anomalies for one specific station. In terms of percentage change, the variability at Bahirdar and Kessie is higher than that at El Diem during the recent years. Tesemma et al. [2010] also noted that the annual discharge at Bahirdar and Kessie increased significantly over the period 1963–2003 as compared to that of El Diem's annual flow. They attributed the increasing trend in the upstream flows for the recent years to the erosion of hillside lands that used to store water but changed as contributors to direct runoff. The negative anomalies in the 1980s are statistically significant, while at the end of the 1990s and the beginning of the 21st century, the pattern shows significantly higher anomalies.

Figure 7.

Comparison between oscillation patterns of annual river flows in the Blue Nile basin for a block length of 5 years and the period 1964–2009.

4.3. Extreme Flow Comparison With Basin-Wide Rainfall and Dry Days

[44] In an attempt to find an explanation for the observed flow variability, basin-wide rainfall was investigated as it is the major driving factor for river flows. The river flow volumes and extremes at El Diem have similar oscillation patterns with that of rainfall in the region, which confirms that there is a positive relation between two variables (compare Figures 4 and 8a). The correlation coefficient between the anomalies for areal rainfall and extreme high flow at El Diem is 0.8, which shows high correlation. Hence, significant flow anomalies might be explained mainly by significant rainfall anomalies. Nonetheless, the temporal variability of the annual areal rainfall over the Blue Nile basin is less variable in terms of percentage change than what was observed for flow. During the 1980s the change in the annual rainfall was approximately −5% (Figure 8), while that of river flow at El Diem was approximately −17%. This indicates the catchment's sensitivity to changes in rainfall. Rainfall-runoff relations are indeed highly nonlinear; increases in rainfall will increase soil saturation levels, which in turn might strongly increase surface runoff coefficients.

Figure 8.

Temporal variability of (a) annual rainfall and (b) number of dry days per year in the Blue Nile basin for a block length of 5 years and the period 1964–2004.

[45] In terms of oscillation patterns, the base periods denoted as 1, 2, and 3 (Figure 4), on the basis of the extremes at El Diem, are also present in the annual rainfall pattern. However, the increasing trend in the extreme flows observed from the mid-1980s to the end of the 1990s is interrupted by an oscillation low period during the mid-1990s in the annual rainfall pattern. Since the available rainfall data was shorter by 5 years than that of the flow at El Diem, pattern comparison was not achievable for the very last 5 years of the analysis. Seasonal analysis using monthly total rainfalls for the three seasons does not show any period with statistically significant anomalies. Also Tesemma et al. [2010] did not find a significant trend in the basin-wide rainfall. They studied annual, dry season, and short and long rainy season rainfall of the Blue Nile Basin and tested the significance at the 5% significant level during the period 1963–2004.

[46] The temporal variability of the total number of dry days per year is shown in Figure 8b. This variability has (for the base period of 1 and 2) a somewhat inverse relationship with the annual minima flow at El Diem. The higher the number of dry days, the lower the extreme low flows, indicating the influence of rainfall on the region. When comparing the pattern between the annual rainfall and the total number of dry days per year, some periods show clear inverse relations; for example, the 1980s (base period 2) and the last few years of the analysis (base period 4). The end of the 1970s shows a significant decrease in the number of dry days, which shows the presence of wet years and corresponds to the positive anomaly for base period 1.

[47] Temporal variability of areal rainfall extremes for the three seasons shows oscillation patterns as in Figure 9. The extremes of the long rainy season in the 1990s have slightly higher anomalies than the 1960s. Hence, the extreme precipitation has become slightly intense during the most recent decade. The 1980s comparison show significantly lower anomalies in the JJAS season. Rainfall extremes during the dry season show significantly higher anomalies (up to 60%) during the late 1990s; which was also observed in the El Diem high flows. Generally, most wet anomalous years are situated during the 1960s–1970s and 1990s for rainfall extremes of the long and short rainy seasons, while the 1980s contain most of dry anomalous years. In order to relate the observed variability on low-flow extremes with rainfall indices, the total number of dry days for the ONDJF (dry) season was considered. Graphically, the temporal variability is as shown in Figure 9d. This pattern has inverse relation with the annual minimum flow change pattern. The variability observed in the hydroclimatic extremes is summarized in Tables 2 and 3 for high and low extremes, respectively. The most consistent pattern in both extremes is the one in 1980s with negative anomalies.

Figure 9.

Temporal variability of rainfall extremes in the Blue Nile basin for a block length of 5 years and (a) June–September, (b) March–May, and (c) October–February seasons and (d) the total number of dry days for the October–February season.

Table 2. Summary of Oscillation Pattern Similarity for El Diem High-Flow and Basin-Wide Rainfall Extremesa
VariablesOscillation High in 1960sOscillation Low in 1980sOscillation High in 1990sOscillation Low in the 21st Century
  • a

    P, shows presence; NC, not clear; JJAS, June–September; MAM, March–May; ONDJF, October–February.

Annual maxima flowPPPP
Wettest month flowNCPPP
Seasonal maxima flow    
Seasonal rainfall extremes    
Annual total flowPPPP
Annual total rainfallPPPNC
Table 3. Summary of Oscillation Pattern Similarity for El Diem Low-Flow Extremesa
VariablesOscillation High in Mid-1970sOscillation Low in Mid-1980sIncreasing Trend in the 1990s
  • a

    Abbreviations are as in Table 2.

Annual minima flowPPP
Driest month flowNCPP
Seasonal minima flow   

4.4. Influence of Climatic Indices on Hydroclimatic Extremes

[48] Investigating the causes of the variability in the hydroclimatic extremes with the use of statistical correlation analysis gave results as presented in Table 4. Table 4 includes the statistical significance of the correlation coefficients at 1% and 5% significance levels on the basis of the t test. Table 4 shows that rainfall and flow extremes are positively correlated with climate indices SOI and AMO, and negatively correlated with PDO. The relationship between PDO and the rainfall and flow extremes are strong and statistically significant at 1% significance level for both annual and long rainy season while SOI shows strong and significant correlation only at annual scale. AMO shows less strong but significant correlation with the long rainy season rainfall and flow variability. Conversely, AMO shows strong and significant correlation during the dry season. IOD has a weak correlation with the extremes in all the seasons and at annual scale. From this analysis it is concluded that the extreme rainfall and flow of the upper Blue Nile basin is influenced by changes that occur on both the Pacific and Atlantic Oceans.

Table 4. Annual and Seasonal Correlation Coefficients Between Rainfall and Flow Extremes and Climate Indicesa
  • a

    PDO, Pacific Decadal Oscillation; SOI, Southern Oscillation Index; IOD, Indian Ocean Dipole; AMO, Atlantic Multidecadal Oscillation.

  • b

    Statistically significant at 1% significance level.

  • c

    Statistically significant at 5% significance level.

Rainfall extremes−0.6b0.5b−0.10.3−0.5b0.1−0.20.4c−−0.20.4c0.30.8b
Flow extremes−0.8b0.5b0.00.3−0.5b0.3−0.30.4c0.0−0.1−0.10.8b−0.20.4c0.10.7b

5. Discussion

[49] The variability observed in the hydroclimatic extremes could be pure randomness or could follow a certain trend or temporal clustering similar to other influencing factors or phenomena (e.g., (multi)decadal oscillations related to large-scale atmospheric circulations). The estimated confidence interval bounds the range of expected natural variability in case of pure randomness. Any change that occurs outside the interval is thus considered statistically significant, where the hypothesis of the change caused by random deviation of a nonpersistent random signal could be rejected. Nevertheless, even within the confidence limits, if there are similar patterns for different hydroclimatic variables and at different locations, then the significance could be higher in comparison with what is concluded on the basis of only one variable and location.

[50] The hypothesis of no persistence in temporal climate variation in the past hydroclimatic extremes is rejected because of observed patterns. The uncertainty on the anomalies, observed as wide confidence intervals, raised the issue whether the variability is purely random and station specific. However, the results of the detailed analysis based on El Diem station were confirmed by comparisons performed with upstream flow records and areal rainfall. The similarity found among stations and rainfall variables increased the confidence that the observed variability is basin-wide and important to draw conclusions.

[51] The analysis unveils apparently various possible causes of variability, which can be separated into natural and man-made factors. The natural causes are (multi)decadal oscillations related to large scale atmospheric circulations. This was investigated using the four climate indices. It is shown that at annual scale, the higher runoff variability in the basin can be partly explained by the large-scale circulation and rainfall patterns associated with PDO and SOI, both are influences from the Pacific Ocean. The secondary oceanic influence comes from the Atlantic Ocean as it has a statistically significant correlation with the long rainy season extremes. These are, therefore, one of the important causes of temporal variability in the basin's hydroclimatic extremes. A recent study by Pui et al. [2011] showed that antecedent catchment conditions as approximated by antecedent precipitation index (API) influences multidecadal variability of design floods and it is modulated by Interdecadal Pacific Oscillation (IPO) climate index. Although short-term antecedent conditions are found to be important factors in design flood computation, in light of the method applied in this study (moving window of 5 years block period by 1 year time step) the influence of short-term antecedent conditions could not be detected.

[52] A further possible cause of variability in hydrological extremes of the upper Blue Nile is the one observed from the El Diem low-flow analysis, which was the strong increasing trend in the recent years. This might be related to the installation of flow regulation by means of a hydraulic structure (Chara-Chara structure) at the outlet of Lake Tana. The regulation aims to store water during the wet seasons and releases it during the dry season [Tesemma et al., 2010]. The Chara-Chara structure began its operation in 1996 and the outflow from Lake Tana is affected since then. If one compares the pattern of low flows before 1996, and the long-term period shown in Figure 6, the constant increasing trend was not present before the mid-1990s supporting the idea that the low flows are influenced by the structure in the recent years. Hence, land and water management causes also contributed to the variability observed in the Blue Nile flow.

[53] Another observation is the pattern found for Gilgel Abay station. It shows a slightly different variation when compared to El Diem station's variability. Gilgel Abay is a tributary river that flows into Lake Tana, the source of the Blue Nile River. The gauging station covers a smaller catchment area compared to the other stations considered. The slightly different temporal variation for this river illustrates that smaller catchments could be influenced by natural variability localized for that region and/or other anthropogenic influences such as land use change or changes in water management, e.g., Water abstraction.

[54] When the comparison is made for the high- and low-flow extremes, results at El Diem and Bahirdar are also similar, except that the anomaly values for Bahirdar are much higher than that at El Diem. In contrast, Gilgel Abay has an opposite pattern for extreme low flows where it shows a decreasing trend (not shown). The decrease in low-flow extremes at Gilgel Abay during recent years might be related to more abstraction of water in the dry season because of high population pressure and/or land cover change. This decreasing trend was also reported by Rientjes et al. [2011]. They quantified that the low flows in the Gilgel Abay basin decreased by 35% during the past 30 years and explained it by significant land cover changes in the basin. A study by Bewket and Sterk [2005] on the Chemoga watershed, a smaller catchment within the Blue Nile basin, showed a statistically significant decline on extreme low flows analyzed at monthly and daily scales. That study concluded that the observed adverse changes in the streamflow have partly resulted from changes in land cover and use and/or degradation of the watershed that involved destruction of natural vegetative covers, expansion of croplands, overgrazing and increased area under eucalypt plantations. In addition, increased dry-season water abstraction to be expected from the increased human and livestock populations in the area is mentioned as another contributing factor. This shows that hydrological changes within smaller catchments could be different than what is observed in the more downstream and larger main Blue Nile River.

[55] Generally, according to the patterns found for rainfall and flow extremes, the most consistent pattern is that of the 1980s, which is the low-oscillation period for both high and low extremes. Conversely, there is no consistent and significant increasing or decreasing trend during the recent years on both rainfall and flow extremes. This result is somewhat similar to the conclusion drawn by Seleshi and Camberlin [2005] on rainfall extremes over the Ethiopian highlands where they found investigated rainfall stations to be free of any trend for the long and short rainy seasons. It is therefore, difficult to spot signature of global warming on either rainfall or river flow extremes.

[56] The results found in this analysis are subject to different types of uncertainties. For instance, the influence of data scarcity is undeniable in computing areal rainfall as the number of stations used to obtain it is very limited with respect to the basin area. However, the fact that similar patterns are obtained for the different hydroclimatic extremes and the consistence of the results with other studies proves that the QPM method is applicable for such kind of analysis.

[57] The presence of (multi)decadal oscillations in rainfall and river flow extremes and volumes shows that one has to be careful with trend investigations based on short records or to associate recent trends with global warming. Trend analysis conclusions indeed depend on the period used for the analysis. In the case of our study area, studies that consider the period before the 1980s might conclude decreasing trend on high flows while studies after the 1980s until the end of the 1990s might conclude increasing trend. The quantile anomaly analysis method applied in this paper, however, showed its capacity to identify both trends and high- and low-oscillation periods and their significance graphically.

6. Conclusions and Recommendations

[58] Upper Blue Nile basin's hydroclimatic extremes variability has been investigated in this study using the QPM approach. From the results there is evidence of (multi)decadal oscillation on the extremes. The presence of statistically significant negative anomalies during the 1980s was consistent on all hydroclimatic variables that were explored. This period separates two oscillation high periods during the 1960s and 1990s. Although the 1990s showed slightly higher extremes in some of the analysis, in general there is no consistent increasing or decreasing trend during the recent years. Thus, it is intricate to conclude that there is an indication of climate change impact on hydroclimatic extremes of the basin.

[59] The cause of observed variability is two folds; primarily meteorological phenomena highly influence flow extremes in the basin as confirmed by the similar pattern observed between high flow extremes and rainfall extremes as well as a somewhat inverse relationship between the number of dry days and extreme low flows. On the other hand, land and water management practices have their own share. For instance, the influence of outflow regulation is undeniable in case of low-flow extremes of the main Blue Nile River. Similarly, the analysis based on the Gilgel Abay station showed that hydrological changes within smaller catchments could be different than what is observed in the larger main Blue Nile River. It is therefore important to conduct separate investigation on extremes in smaller catchments of the Blue Nile basin as the temporal variability might be unlike to that of the entire basin and might be influenced by local factors and catchment characteristics. This is also vital to assist water management practices of each catchment and the agricultural sector.

[60] The comparison made between climate indices and hydroclimatic extremes suggests that oceanic influences are part of the explanation for the variability observed in rainfall and flow extremes. Changes in the Pacific and Atlantic oceans are identified as the major natural causes for the observed (multi)decadal oscillations on the extremes. Understanding the influence of these climate oscillations is essential because of their considerable effect on the water and agriculture sector of the basin. The link between hydroclimatic extremes and climatic indices might also be important to extrapolate temporal anomalies in rainfall and flow for stations with short record lengths (given the large scale climate drivers such as the SOI have long historical records). It is also possible, using such a link, to separate natural variability (at (multi)decadal time scales) from long-term trends (e.g., due to global warming) or to predict upcoming periods with oscillation highs and lows.

[61] The QPM method used in this particular study showed its creditability to study extremes as the results found are consistent with previous studies which were conducted on seasonal totals. The method also showed the possibility of separating natural from long-term trends in extremes. It is apparent that such information would be very important for the water management, agricultural and other sectors, and for the better management of ecosystems.


[62] The research was carried out using data from the FRIEND and NILE projects of UNESCO and support from the Flanders in Trust Fund of the Flemish government of Belgium. The authors also acknowledge the Ministry of Water and Energy and the National Meteorological Agency in Ethiopia for the provision of data.