1. Introduction
Ethiopia is an agrarian nation that is heavily dependent on rain fed agriculture. More than 85% of its population is engaged in dominantly rain-fed agriculture. The irrigation infrastructure are lacking that small variabilities in rainfall have often been translated to droughts and subsequent famines [
1,
2]. As the most utilized river in the country providing water to most irrigated farms, agro industries, and urban centers, the Awash River is facing a number of challenges. In recent years, the basin is experiencing increasing variability in water availability, deteriorating water quality, and increasing incidence of extreme events- droughts and flooding [
3,
4,
5,
6]. Under such conditions, improved understanding of the impact of climate change on annual and seasonal rainfall is vital for policy makers and planners to prepare and plan against the prevailing changes.
According to the Intergovernmental Panel on Climate Change (IPCC) [
7], Climate change is long term deviations in mean values of climatic parameters. It is related to natural or anthropogenic activities, causing global warming directly or indirectly. Some of the major underlying human and natural drivers for the climate change are rapid industrialization, deforestation, agricultural activities, and volcanic eruptions. The activities such as fossil fuel burning driven by rapid population and economic growth increase emission of greenhouse gaseous (GHG) in to the atmosphere. The GHG have blanketing effect, allowing sun radiations in to earth’s surface but preventing reflected radiation from going back to the space. While activities such as deforestation release carbon dioxide in to the atmosphere and reduce forest cover, causing carbon dioxide sinking capacity of the earth to decline. Furthermore, the induced global warming itself causes melting of polar ice and this reduces reflective capacity of the earth, thereby increasing the proportion of solar radiations absorbed by the earth surface. This effect is termed as ice-albedo feedback effect. There are also other similar feedback mechanisms aggravating global warming, showing that global warming has ‘self-perpetuating’ nature. Overall, climate change is resulting from changes in atmospheric concertation of GHG and aerosols, in the amount of incoming solar radiation, and in the reflective properties of the earth surfaces. However, increasing amount of GHG in the atmosphere due to anthropogenic activities is the main cause of global warming in 20th century and the warming is responsible for overall climate change [
7,
8,
9,
10].
Climate change has been observed across the world, and is often related to global warming. According to the IPCC [
10], the global mean temperature has been rising since 1850, and has risen by 0.6 °C in the second-half of the twentieth century. The rising temperature is widespread across the globe; but, it is at a higher rate in northern latitudes and lower in some parts of the oceans. Yet, cooling also has been experienced in a few locations [
11]. The change is not limited to temperature. With the close link between the climate systems and the hydrological cycle and its components, alterations in precipitation characteristics, rising atmospheric humidity, increasing evaporation, and change in soil moisture and runoff have also occurred [
8]. Accordingly, global precipitation has been increasing over the 20th century. Generally, a very small upward trend of approximately 1.1 mm per decade (with uncertainty ±1.5 mm) has been estimated in the global annual mean precipitation over the period from 1901 to 2005. However, different trends have been observed in annual precipitation across regions and locations [
10,
12].
In Ethiopia, the national average annual rainfall remained almost constant at least for the last 50 years [
1,
13]. However, trends of annual and seasonal rainfall vary widely across locations. Seleshi and Zanke [
14] reported no significant trend in annual and seasonal rainfall over the central, northern, and the northwestern areas, but decreasing trends over the eastern and the southern areas of Ethiopia in the period from 1982 to 2002. Similarly, Cheung et al. [
1] found no significant trend in annual and autumn rainfall (February to May) in all individual watersheds in Ethiopia but a significant decreasing trend for summer rainfall for some catchments over the period from 1960 to 2002. While Viste et al. [
15] reported significantly decreasing trends in annual, spring (February to May), and summer (June to September) rainfall over the southern, eastern, and the southwestern areas but no particular trends for annual and seasonal rainfall in the central and north parts of Ethiopia. Teyso & Anjulo [
16] reported increasing trends over some locations and decreasing trends over other locations for annual and seasonal rainfall in the Gamo-Gofa zone in southern Ethiopia.
Particularly for the Awash River Basin, Cheung et al. [
1] and Viste et al. [
15] found no significant trend (α = 0.05) in annual, summer and spring rainfall for the entire Awash River Basin over different periods. Similarly, no significant trend in annual rainfall has been reported in a number of studies for some stations in the Awash River Basin [
14,
17,
18,
19]. On the other hand, Rosell [
20] found an increasing trend in annual rainfall in six out of the seven stations in the Awash River Basin; whereas, Bekele et al. [
18] and Asfaw et al. [
21] reported a significant decreasing trend in annual rainfall over some stations in the basin. There are clear contradictions on results for the Awash River basin among the existing publications both for annual and seasonal rainfall.
Trends can be detected using either parametric or non-parametric tests. The parametric test is based on an assumption that the sample data come from a population that follows a normal distribution. The parametric tests have higher efficiency and power than the non-parametric test for normally distributed data. However, they are rarely used for environmental data without adjustments for outlier and missing data. Further, additional uncertainties associated with using the model and difficulties in applying the methods make parametric tests less preferable in hydrological studies. The most widely used parametric method is linear regression [
22]. For linear regression, the data needs to be normally distributed and independent. Linear regression is severely affected by outliers, missing data, and the starting and ending values of the time series [
22,
23,
24]. However, linear regression based on ordinary least square (OLS)-based has been used for trend analysis in a number of studies [
1,
14,
15,
17,
19,
25,
26,
27,
28]. For the non-parametric tests, no assumption is made about the distribution of the population. They are simple to use and far less impacted by outlier and missing data than the parametric test. In addition, they represent a measure of monotonic dependence whether linear or not [
23]. The power of non-parametric tests rises with increasing sample size, and they can perform better than parametric tests when the data depart from normality.
The non-parametric tests have been favored in hydrological time series analysis due to simplicity and suitability for data with outliers. Even for normally distributed data, non-parametric tests are preferred and safer because they can be applied without prior assumption about the population distribution of the data [
22]. The most commonly used non-parametric statistical tests in trend analysis are Mann–Kendall’s for trend test and Sen’s Slope tests for slope estimation [
18,
29]. The Mann-Kendall (MK) test is a non-parametric approach for testing the significance of monotonic trends, linear or nonlinear, in time series data [
30]. The test is based on ranks of observations, not the actual values of the data series, making it uninfluenced by missing and outlier data. The MK test is strongly recommended by the World Meteorological Organization as a standard non-parametric procedure for testing trends. It is widely used in trend analysis of hydro-meteorological time series [
21,
23,
31,
32,
33]. Similarly, Sen’s Slope estimator has been widely used to estimate the slope of a linear trend for a time series [
18,
21,
23,
31].
Trend analysis involves different approaches of data analyses. The approaches can be combinations of different spatial and temporal extent of data. Trends of rainfall can be analyzed at different time scales using station-based rainfall time series [
14,
18] or area-averaged rainfall time series [
1,
15,
34]. Most of the decisions and actions in water resource management are at an areal scale, requiring results of the analyses at local or regional watershed levels. Furthermore, results of the analyses at one or more stations cannot directly represent a given area or watershed [
1,
24]. Therefore, analyses of trends of rainfall are practically more important at an areal spatial scale than at point scales. However, area- averaged rainfall would mask real variabilities, especially when station density varies in space [
14]. Cheung et al. [
1] have also shown the risk of averaging station data when there are an insufficient number of stations, resulting in unrepresentative averaging. This risk is intensified when averaging inconsistent station data for large watersheds. On the other hand, Cheung et al. [
1] noted that contradictions in results of the previous trend studies in Ethiopia are attributed to arbitrary division of the study area used in those studies, in addition to poor and inconsistent data quality. Thus, to overcome a similar problem, Cheung et al. [
1] have recommended the use of well-defined study areas, i.e., areas defined in an ‘objectively and geographically meaningful manner’ instead of using gauge locations, political boundaries, or any other arbitrary study area. In hydrology, such meaningful and objectively defined study areas are watersheds.
For a number of reasons, sufficient and reliable information regarding trends of annual and seasonal rainfall in the Awash River Basin is not yet available. There are few studies for the entire or parts of the basin. Some station-based climatic trend studies have included few locations from the Awash River Basin [
14,
17,
19,
20,
21]. While only a few of these studies have included significant parts of the basin or the entire basin [
1,
15,
18,
31]. The study by Bekele et al. [
18] is the only to exclusively investigate trends of rainfall and temperature over several locations within the Awash River Basins. However, the existing research has a number of limitations. Firstly, most of the previous studies used short record length and low data intensity. He and Gautam [
24] considered a data record length less than 50 years as short-record. Accordingly, most studies relevant to the Awash river Basin were based on short-data record length [
1,
15,
18].
However, Wagesho et al. [
31] carried out the analysis of trends based on a 50-year data record length at national level including the Awash River basin. As for the data intensity, some of the national studies were based on large numbers of data points [
1,
15,
31]. However, Bekele et al. [
18] used rainfall data from four stations to represent the entire basin. Obviously, this is very low data intensity to represent a 116,374 square-km basin with diverse climatological zones. Secondly, most of the studies are based on data analysis at point-scale or arbitrarily defined areas [
14,
17,
18,
19,
20,
31]. Cheung et al. [
1] also identified these problems over a number of publications. According to Cheung et al. [
1], use of such an arbitrary spatial scale of analysis may cause misleading results due to unrepresentative observations or subsequent unrepresentative averaging. Furthermore, the results would have less practical importance in water resource management, in which information at a watershed level is required for decision making and actions [
1,
24].
Thirdly, in most studies there are limitations related to data analysis techniques. Several studies for the Awash River Basin used linear regression to assess trends [
1,
14,
15]. He and Gautam [
24] noted that the results of trend analysis can be significantly affected by the use of linear regression. Additionally, most of the studies did not consider serial correlation in time series that would result in errors in trend detection [
1,
14,
15,
17,
18,
19,
20,
21]. Overall, as most of local studies citied above are carried out for areas outside the Awash River basin, it is clear that there are no studies covering significant portion of the Awash River basin with sufficient level of details. On the other hand, the existing local studies for the basin or elsewhere have limitations with respect to data quality, intensity, data record length, the approach of analysis, and the data analysis techniques.
Therefore, this study was carried out to investigate long-term trends and magnitude of trends for areal rainfall on an annual and seasonal basis over the Awash River Basin, in order to address the existing research gaps, and thereby contribute to a better understanding of the impact of climate change on trends of rainfall in the basin. To this end, firstly, long-term rainfall data from a gridded climate dataset were used for more consistent and reliable data, with long record length and high intensity over the study area. Secondly, the data analyses were carried out to a hydrologically meaningful spatial extent at the sub-basin and basin level. Lastly, the most suitable and widely acceptable techniques of trend analysis were applied.
4. Discussion
For annual rainfall, in agreement with the result, no significant trend was reported in a number of studies for the overall Awash River Basin or for some locations in the basin [
1,
14,
15,
17,
18,
19,
31,
54]. However, among the above studies, Bekele et al. [
18] reported a decreasing trend over two locations in the basin as well. Furthermore, Marshall et al. [
27] and Asfaw et al. [
21] reported a decreasing trend in annual rainfall over some locations in the basin. On the other hand, Rosell [
20] and Degefu and Bewket [
23] reported an increasing trend in annual rainfall over several locations in the basin. Outside the Awash River Basin, Tesemma et al. [
29] and Mengistu et al. [
28] found no significant trend over parts of the Blue Nile River Basin, whereas Addisu et al. [
38] found a declining trend in a majority of the locations but no trend in some locations within the Lake Tana Sub-basin. Wegesho et al. [
31] reported a decreasing trend in annual rainfall over the northern, northwestern, and western parts of Ethiopia, but an increasing trend over some locations in eastern Ethiopia. Similarly, Teyso and Anjulo [
16] reported both increasing and decreasing trends over different locations in the Gamo Gofa zone in south western Ethiopia.
For summer rainfall, the dominantly decreasing trends identified over the sub-basins agree with results of some publications. Wegesho et al. [
31] found a significant decreasing trend (at α = 0.1) over most parts of the Awash River Basin over the period 1951–2000. Addisu et al. [
38] identified different rates of change for monthly rainfall over the Tana sub-basin. The average of these rates in the summer season is negative, indicating a decreasing trend in the season. Bekele et al. [
18] also reported a non-significant decreasing tendency in summer rainfall over locations such as Ginchi, Debrezeit, Dire Dawa, and Messio. However, results from a number of publications contradict both decreasing and no trends identified in this study. Contrary to the identified decreasing trend in this study, Seleshi and Zanke [
14] found no significant trend in summer rainfall over a few stations (Addis Ababa, Dire Dawa, and Awash) for the period 1965–2002. Viste et al. [
15] reported a decreasing trend for summer rainfall over large parts of the country, including the Awash River Basin. Similarly, Cheung et al. [
1] and Wegesho et al. [
31] also found no significant trend (α = 0.05) during the summer season for the entire Awash River Basin. Furthermore, no significant trends were identified by Urgessa [
54] over three locations (Metahara, Gewane, and Asiyata) in the Awash River Basin, by Alemayehu and Bewket [
17] in most of the grid points in the central highlands of Ethiopia, and by Bekele et al. [
18] over 9 locations in the Awash River basin.
On the other hand, Mohammed et al. [
19] over two locations (Dessie and Haik), Bekele et al. [
18] over three locations (Addis Ababa, Koka, and Melkasa), and Rosell [
20], over four locations (Majete, Combolcha, Dessie, and Hayik) within the Awash River Basin found significantly increasing trends in the summer rainfall. As for the rate of change, estimates by Asfaw et al. [
21] and Viste et al. [
15] are a bit higher than the upper limit for the estimated range in this study. Asfaw et al. [
21] estimated a decline in summer season rainfall by 13.12 mm per decade for the Woleka sub-basin. Similarly, Viste et al. [
15] reported a decline in summer season rainfall at a rate of 13 mm per decade over large parts of the country, including the Awash River Basin. On the other extreme, Asfaw et al. [
21] estimated a very low declining rate of 1.93 mm per decade for the Woleka sub-basin.
For autumn rainfall, in agreement with the results, some research reported no significant trend over parts of the Awash River Basin. Seleshi and Zanke [
14] reported no trend in autumn rainfall in the central parts of Ethiopia consisting of some parts of the upland catchment of the Awash River Basin. Cheung et al. [
1] also found no significant trend in autumn rainfall in the entire Awash River basin, while Urgessa [
54] found no trend in three locations (Metahara, Gewane, and Asiyata) in the Awash River Basin. Viste et al. [
15] found that the autumn season rainfall shows no trend over large parts of the country, including the Awash River Basin. Similarly, Asfaw et al. [
21] reported no trend in autumn season rainfall over some locations in the Woleka sub-basin. On the other hand, in agreement with an increasing trend in Eastern Catchment of the basin, Seleshi and Zanke [
14] reported an increasing trend in autumn rainfall. Similarly, Cheung et al. [
1] also identified an increasing tendency in the eastern catchment, though it was not statistically significant. Bekele et al. [
18] also identified a significant increasing trend in the autumn rainfall at locations such as Dire Dawa in the Eastern catchment. However, in contradiction with the results, Liebmann et al. [
55] reported a decreasing March-May rainfall in the Horn African region. Similarly, some local studies also reported a decreasing trend in autumn rainfall in the central highlands of Ethiopia and over the Woleka sub-basin [
17,
20,
21].The estimated rate of change is relatively lower than estimates by some studies. Seleshi and Zanke [
14] reported an increasing trend in autumn rainfall at Dire Dawa at a rate of 22 mm per decade. Asfaw et al. [
21] estimated a decline in the autumn season rainfall at a rate of 15.03 mm per decade for the Woleka sub-basin. Viste et al. [
15] reported a decline in the autumn season rainfall at a rate of 12 mm per decade over a large part of the country, including the Awash River Basin.
There are nearly no publications available dealing with trend analysis of the winter season rainfall for the study area. This might be because of low rainfall in the season. In agreement with results in the study, both increasing and no trend were reported in the available publications. Liebmann et al. [
55] reported increasing October–December rainfall over the whole Horn of Africa region. Viste et al. [
15], for large parts of the country, including the Awash River Basin and Mengistu et al. [
28] for the Blue Nile River Basin, reported no trend for winter season rainfall.
As for the rate of change, the upper limit of the estimated range in this study is closer to the 3 mm per decade estimated by Viste et al. [
15] for parts of the country receiving major rainfall in summer, including the Awash River Basin.
However, comparisons of the results of this study with previous studies have been difficult due to differences in sources of data (observed or gridded data), data record length, approach of analysis (point or areal scale), and methods of analysis (parametric, such as linear regression, or non-parametric, such as MK test, or Sen slope estimator), as well as difference in seasonal patterns, in the number of seasons, and the months constituting each season. Especially, the difference in temporal and spatial extent of analysis between this study and previous studies has made the comparisons uneasy. Most of the previous studies relevant to the basin are based on short-data records [
1,
15,
18] and at point-scale or arbitrarily defined areas [
14,
17,
18,
19,
20,
31], while this study is based on long data record (1901 to 2016) and areal scale analysis of rainfall. Cheung et al. [
1] have also explained contradictory results in most of the earlier studies due to difference in spatial extent in addition to poor and inconsistent data quality.
It is noted that in some of the sub-basins where significant decreasing summer rainfall has been identified, a significant increasing trend has also been identified in either autumn or winter rainfall. Thus, given no significant trend in annual rainfall over all sub basins and the entire basin, it can be inferred that the decrease in summer rainfall over some of the sub-basins could be compensated for by an increase by either autumn or winter rainfall. For instance, the decrease in summer rainfall over Upland could be compensated for by an increase in winter rainfall. Similarly, the decrease in summer rainfall over of the Upper Valley and Western Highlands could be compensated for by an increase in winter rainfall. Cheung et al. [
1] have reported a similar shift in rainfall patterns. However, similar compensation effects between autumn and winter seasons has not been noticed. Overall, the results show that climate change has not induced statistically significant changes in annual rainfall in the basin over the period (1901–2016). However, the amount of seasonal rainfall has been significantly affected by the prevailing climate change over most parts of the basin. A shift in seasonal rainfall has also been noticed due to climate change.
5. Conclusions
The annual rainfall shows no significant change over the 116 years in all sub-basins and over the entire basin. However, trends of seasonal rainfall are different across study areas. The summer rainfall showed significant decreasing trends (α ≤ 0.1) over five sub-basins at rates varying from 4 to 7.4 mm per decade, but showed no trend over two sub-basins—the Lower Valley and Lower plain. The autumn rainfall showed no significant trends over four sub-basins but showed increasing trends over three sub-basins (Lower Valley, Lower plain, and Eastern Catchment) at a rate varying from 1.9 to 5 mm per decade. The winter rainfall showed no significant trends over four sub-basins but showed significant increasing trends over three sub-basins at a rate varying from 0.6 to 2.7 mm per decade. The rate of change is generally highest in summer rainfall, followed by autumn rainfall. The winter rainfall showed the least rate of change over the period (1901 to 2016). On the basin level, summer rainfall has shown significant decreasing trend (α = 0.05), while autumn and winter rainfall have showed no a significant trend.
Understanding the importance of summer rainfall as the main rainy season and the heavy dependence on rain-fed agriculture in Ethiopia and in the basin, the decreasing trend in summer rainfall may have meaningfully impacted the agricultural productivity in the basin.
It is noted that the decrease in summer rainfall over some sub- basins could be compensated for by an increase in either autumn or winter rainfall. However, a similar compensation effect between autumn and winter seasons has not been noticed. Overall, it is clear that climate change has been significantly affecting the trends and patterns of seasonal rainfall in the basin. The impacts of the changes in seasonal rainfall on agricultural and pastoral communities in the basin might need to be studied in detail.
All possible opportunities and challenges related to the prevailing trends, the rate of change, and shifts in seasonal rainfall need to be evaluated. However, at this point, it can be noted that the shift of summer rainfall to autumn or winter might have severe implications for agricultural areas that are receiving insufficient summer rainfall. On the other hand, the increasing autumn or winter rainfall might be considered supportive to livestock and crop production, both to the dry and wet regions. For instance, areas with long and severe dry seasons might get some relief due to the increasing winter rainfall. Overall, the annual rainfall has consistently shown no trend over the basin, indicating that the risk of droughts in the basin may be related to the seasonal distribution of rainfall. Therefore, it is recommended to expand water storage infrastructures, such as dams and reservoirs, and to improve management of water resources to prevent potential seasonal droughts owing to changing seasonal rainfall.