Characterizing Spatiotemporal Patterns of Drought among the Pastoralists of Southern Ethiopia, Eastern Africa

This paper is an inquiry of investigating trends and spatiotemporal patterns of meteorological drought in semi-arid pastoral areas in southern Ethiopia. Monthly gridded rainfall and temperature data for thirty years between 1986 to 2016 were obtained d from Ethiopian Meteorological Agency. Nonparametric Mann Kendall’s test (MK) along with Sen’s slope estimator was employed to see trends of drought. In addition, Reconnaissance Drought Index (RDI) was used to characterize the drought patterns. The result indicates that under all time scale, drought events tend to show an increasing trend with varying magnitude and intensity. The highest drought severity category for major and small rainy months was identified during May 2000 with RDI= -2.28 and September 2000 with RDI= -2.55 . The drought magnitudes of the small rainy months range from -11.15 in November to -11.76 in September, while it ranges from -10.08 in April to -11.38 in March for the major rainy months. There are some variations of drought magnitude among the seasons while rainfall tend to show gradual decline as a manifestation of spatially varying drought severity, frequency and intensity in the watershed.


Introduction
Drought is an incessantly happening climatic phenomenon mostly related to the reduction of precipitation (Alok and Anil, 2018) and is time of prolonged below-normal supply of water (Rajib et al., 2013;Ndlovu and Brain, 2011). Although drought happens ubiquitously in the world; the intensity, frequency, severity and impacts vary spatiotemporally. The available estimates on drought impacts suggest that, during the period 1900-2013, there were 642 drought events reported across the world resulting in a huge toll to humanity, killing about 12 million people and affecting over 2 billion (Kiumars et al., 2016). There was a tremendous change in terms of its frequency, severity and geospatial coverage (IPCC, 2007;Shiau andHsiao 2012 andMishra andSingh, 2011). To this end, there are number of indications about projection of threats from the drought where the fatalities are also escalating. According to IPCC (2007) report drought is expected to expose between 75-250 million people to water stress by 2020 globally. By 2020, there will be a significant reduction in arable land in Africa, yields from rain-fed agriculture will decline by as much as 50%, which worsens the need of food stuff (Oxfam, 2008). Due to regional variability of rainfall, countries in Africa face varying drought incidences. The highest spatial and temporal variability of rainfall is found across most of the countries having a semi-arid climate within western, eastern and southern Africa (Misangi, 2014). Kenya has a tropical climate with two rainy seasons, one occurring in March, April and May and the second one in October, November and December (Peter et al., 2001 andFrancis et al., 2015). Over 80% of the country falls under arid and semi-arid lands (ASAL) which are prone to droughts International Federation of Red Cross and Red Crescent Societies (IFRCRCS, 2014 andFrancis et al., 2015). This can also be largely linked with the differences in intra-annual variability of the rain each region receives. Southern Africa receives most of the rainfall during October-March, whereas Sahel rainfall is concentrated during July-August summer monsoon period (Mavhura et al., 2013). The countries in the Horn of Africa usually receives rainfall in two seasons; October-December as short rainfall season and March-May a long rainfall season (Gebrehiwot et al., 2011;Georgis, 1997 andSeleshi andCamberlin, 2006). North Western Africa receives most of the rainfall during October to April (Temesgen, 2010).
In African countries there is variability and below average rainfall at regional basis, which gives birth to variation of drought severity and frequencies across the continent (Araya andStroosnijder, 2011 andAgnew andChappell, 1999). In the continent alone, the droughts that was occurred in 1972-1973, 1983-1984 and 1991-1992 were most intense and widespread (Masih et al., 2014). Regionally, Northern Africa receives very low rainfall and has a desert climate (Davies et al., 1991) in which the persistent dry years reduced societies coping capacity to droughts. Prolonged droughts were recorded in the recent past in 1999-2002in northwest Africa and between 1970s-1980s in western Africa (Dembele and Zwart, 2016, Agnew and Chappell, 1999. From 2010-2011, there was drought with varying intensity recorded in eastern Africa (Degefu and Bewket, 2014;Degefu et al., 2017, and Gebrehiwot et al., 2011) and by 2001-2003 in southern and south eastern Africa (Masih et al., 2014 andOxfam, 2008).
Being a country in the eastern Africa, Ethiopians has been facing different drought incidences with dozens of animal and human deceases. According to Zenebe et al., (2016), despite the fact that Ethiopia has a long history of drought, the frequency and spatial coverage of droughts have increased over the past decades since 1960s. Both the rise in temperature and prolonged deficiency of precipitation are major factors (Yuei and Getachew, 2019). The drought apparently escalated food shortages through declining agricultural productivity in the 1970s and 1980s (Terefe and Mengistu, 2012). The most severe drought Ethiopia faced was the notorious famines known as Kifuken 1 , which devastated major areas of the country during 1980s Ethiopian Environmental Authority (EEA, 1998 (Viste et al., 2013;MoA, 2012 andUSAID, 2016).
Nowadays, drought incidences put pastoralism and agro-pastoralists livelihoods at high risk (MoA, 2012). During the years 1960-1990s, Ethiopia used to face drought frequencies every eight-ten years (Tsega, 1997), but now the time interval reduced to every five years and even below (Zenebe et al., 2016 andUSAID, 2016). As vulnerability continued to increase, the impacts are similarly increasing for both pastoralist and farming community. There is continued failure of harvest far less than the expectations due to an extended extreme dry season and strong rain in the wet season, followed by a prolonged absence of precipitation, which is likely due to a manifestation of global warming (Gebrehiwot et al., 2011 andMoA 2012). The El Niño Southern Oscillation (ENSO) phenomenon hugely impacts Ethiopian rainfall (Peter et al., 2010;Camberlin, 1997 andWoldeamlak andConway, 2007). It hugely resulted in uncertain drought. For instance, a drought crisis in Ethiopia, triggered by erratic and severely declined rainfall in early 2015, has affected 9.7 million Ethiopians (Roop et al., 2016). This in turn made pastoralism and pastoral livelihoods unsustainable. Moreover, this kind of situation created the cycle of vulnerability and depletes the capability to overcome hunger and poverty (Abate, 2013). identifying trends and spatial patterns of drought helps intervene drought incidences and limit impacts. Thus, the aim of this paper is to analyze trends and spatiotemporal variation of drought in Burkitu watershed of Guji zone in southern Ethiopia.

Description of the Study Area
Dugda Dawwa Woreda is found in Eastern Guji zone Oromia regional state. It is one of the woredas of the Eastern Guji zone with area coverage of mentioned to be 2180 km 2 . Fincawa is the capital town of the woreda which is located at about 500 km from Addis Ababa and 25 kms from Bule Hora, the capital town of the zone. It is mainly inhabited by pastoralists with very small number of agro-pastoral households. Agro-ecologically, the area is known for its arid and semi-arid climate. The mean maximum and minimum temperature of the area is 27 December to February in the Winter followed by the short dry season (Adoolessa) occurs from June to August in the Summer season. In the district, the average annual rainfall ranges between 350 and 850 mm, with considerable spatial and temporal variability. Inconsistent rainfall results in great variability of forage and range production.

Fig 1 Location Map of the Study Area
2 Ganna is local name given to the time of relatively long rainy season 3 Hagayya is local name to short rainy season In few areas of the district, agricultural crops like maize, teff, sorghum and haricot beans are produced. It is estimated that 55.4% of the income in the zone is derived from the sale of livestock. The main livestock being reared are cattle, sheep, goats, sheep and camels. Cattle contribute much to the milk for home consumption and butter production used both for consumption and sold at market. Subsistence agriculture also contributes to a lesser extent to food needs in the district. Beside this, bee keeping practice and crop production activities in the high land areas of the district practiced in limited custom. In the very recent time, rearing livestock becoming challenge due to degradation of range land, population pressures, water shortage, recurrent drought and livestock diseases. In addition, petty trades are highly practiced at Finchawa town and few kebeles nearby to the town. Similarly, sale of livestock, production and sale of vegetables, charcoal, firewood, water, minerals (such as gold, marble and granite), incense and natural gum were common in the study area.

Data sources and acquisition
Drought analysis in this paper was holistically focused on investigating trends, magnitude, severity, intensity and spatial pattern in the watershed. Basically, calculations of drought pattern were based on the long-term meteorological data that were acquired from Ethiopian National Meteorological Agency (ENMA) for the year 1986-2016. Therefore, gridded data of rainfall and temperature were taken for thirty years and calculations were made for different time (1-month, seasonal and annual) scales.

Drought Indices
In drought characterization based on its magnitude and spatial analysis, different indices can be used. In the arid climate, Standardize Precipitation Index (SPI) and Reconnaissance Drought Indexes (RDI) are most commonly used. They are recommended to quantify the deficiency of precipitation at different time scales and across space (Kumar and Anil, 2018;Melaku, 2013;Oloro, 2006 andAlok andAnil, 2018). Often times, SPI criticized for it fails to precisely quantify drought condition of prolonged time over the areas with scarce rainfall. This is due to it solely uses precipitation as an input and not considering temperature and its influences. Therefore, in the areas where temperature is high throughout the year and rainfall is inconsistently received annually, RDI is appropriate for it considers potential evapotranspiration (PET). Therefore, RDI is a meteorological index used for drought assessment (Mohammad et al., 2011;Mohammed et al., 2018 andZarch et al., 2011) and it was developed to analyse water deficit in a more accurate way, as a sort of balance between input and output in a water system (Tsakiris et al., 2007c). Moreover, RDI can be effectively used to compare the drought conditions between areas with different climatic characteristics (Haied et al., 2017). It also has an advantage as it enables universal applicability over the other indices (Haied et al., 2017). By employing monthly or seasonal measures of PET from the general index of meteorological drought, extent of drought spatial pattern and severity for different time scale over the space can be mapped. Therefore, as the area under study is categorized under an arid climate, RDI were used to characterize the drought.

RDI computation
The RDI is expressed in three forms namely, the initial value (αk), normalized RDI (RDIn) and standardized RDI (RDIst) (Tsakiris and Vangelis, 2005). The initial value (αk) is presented in an aggregated form using a monthly time scale and be calculated on a monthly, seasonal or annual basis. Concerning the values, positive values of the RDIst indicates wet periods, whereas the negative values indicate dry periods compared with the normal conditions of the area. The computation of RDI was done by using Drinc Software which is based on the ratio of total precipitation (P) to potential evapotranspiration (PET) accumulated over the selected time scale (k). On the annual basis (k=12), we have: (1) Where Pij and PETij the precipitation and potential evapotranspiration of the j-th month of the i-th year.
The standardized RDI is computed by: with y (i) =ln(a (i) ), yk arithmetic mean and σ yk standard deviation.
The Hargreaves equation can be written as: where PET is the computed reference evapotranspiration (mmd -1 ); while Ra is the water equivalent of the extraterrestrial radiation (mm d -1 ). Similarly, Tmax, Tmin and T are the monthly maximum and minimum temperature in (°C), with T calculated as the average of Tmax and Tmin, where 0.0023 is the original empirical coefficient proposed by Hargreaves andSamani in 1985 (Seleshi andCamberlin, 2006).
According to Tsakiris et al., (2015), drought can be grouped in to different category based on the RDI values. An example for development of the αk relationship and fixing of ranges for drought classes is illustrated for Burkitu watershed in Dugda Dawa district. Using this relationship, the levels of drought have been fixed based on αk for Burkitu Watershed and given below in the (Table 1).

Characterizing drought events: Magnitude, Intensity and Severity
The analysis of the drought trend was done using a non-parametric Mann Kendall (MK) test.
The rank-based MK test is commonly used to assess the significance of monotonic trends in hydro meteorological time series (Mohammed et al., 2018;Mekonnen andWoldeamlak, 2014 andTabari, 2011). In the process of computation, determination of trend was run by taking values of Reconnaissance Drought Index (RDI). MK test was used by many researchers for trend detection (Kendall, 1975 andMohammed et al., 2018). It is widely used to detect trends of meteorological variables (Mekonnen and Woldeamlak, 2014). It has two advantages and thus preferably used by different researchers. First, MK test does not require the data to be normally distributed and secondly, it has low sensitivity to abrupt breaks due to inhomogeneous time series (Tabari, 2011). It entails an assumption of either increasing or decreasing trend in the element (data). According to this test, the null hypothesis, H0 assumes that there is no trend (the data is independent and randomly ordered) and this is tested against the alternative hypothesis H1, which assumes that there is a trend (Neha, 2012). In the analysis of trends of drought in observed rainfall time-series (1-month, seasonal and annual) was calculated on annual and seasonal basis. The M-K test statistic S is calculated using the following formula: Where, xi is the data time series ranked from i = 1, 2, ... n-1; and j = 2,3....n. Each of the data point xi was taken as reference point, which is compared with the rest of the data points, xj: sgn ( ⱼ − ) = { −1 for ⱼ < xi 0 for xⱼ = xi 1 for xⱼ > xi A positive value of S indicates increasing trend, and a negative value indicates decreasing trend. It has been documented that when n ≥ 8, the statistic S is approximately normally distributed with the mean zero and variance computed as: Where, m is the number of tied groups and ti is the number of ties in sample i.
The normal test statistic ZMK is computed as: The trend is said to be increasing, if ZMK is positive and computed ZMK statistic is greater than the ZMK value corresponding to a desired level of significance (0.01, 0.05 or 0.10); and the trend is said to be decreasing, if ZMK is negative and computed ZMK statistic is greater than the ZMK value corresponding to desired level of significance (Mekonnen and Woldeamlak, 2014). The Sen's slope estimator has been found to be a better and more power tool as compared to simple linear regression to detect the linear trend as it is unaffected by gross data errors and outliers (Kumar and Anil, 2018). The Sen's slope is estimated as the median of all pair-wise slopes between each pair of points in the data set. The pair-wise slope Ti of all data points is computed as: Where, xj and xj are data values at the times i and j (j >i), respectively. For n values in data set, there will be N= n(n-1)/2 number of slope estimates, and the median of these N values of Tij is represented as Sen's estimator of slope (Qi), which is calculated as: And Qi= 1/2 (TN/2 +T(N+2)/2), if N is even (7) Positive value of Qi indicates an increasing trend, while a negative value indicates a decreasing trend. In characterising temporal drought events, frequency, severity, magnitude and intensity were computed. In this analysis, drought frequency is a return period between drought events that have negative values of RDI. In Abebe et al (2020) drought magnitude is a drought event corresponds to the cumulative water deficit during the drought period below some threshold (RDI-values ≥ -1) and drought intensity is the ratio between drought magnitude (DM) and duration (DD) of the event given as: Where: DM= is drought magnitude, n=number of months with drought event at j time step, Dd= is drought duration.

Mapping spatial distribution of drought incidences
The RDI values produced by DrinC software were used as an input to ArcGIS to generate drought severity maps for the study area at 1, 3 and 12-month time scales.  The total drought events for the main rainy season are relatively lower as compared to the small rainy months. However, for the rain of small rainy months in the region is apparently inconsistent, strong drought with high RDI value is often commonly identified. Thus, duration of drought event is higher for small rainy months than the main rainy months. Annually, the year 2000 showed the highest drought manitude and severity with RDI value of above -2.0.
The highest drought severity category for major and small rainy months was identified for May there was a lower drought intensity but higher drought severity as compared to Autumn. Due to the lesser moisture in the months from December to February which proceeds the belg season, rain was highly affected as aridity increases in these months which tend to increase drought magnitude and severity. Moreover, the long dry season (Bona Hagayya) occurs from December to February play a significant role over the rain pattern in the area. Ostensibly, holding a drought magnitude of -24.9, drought intensity of -3.55 and severity of -2.56, belg season of the entire years had experienced spatiotemporally varying drought distribution in the the watershed. Cognizant to this, in pastoral zones observed rainfall declines during the belg season is reducing the quantity and quality of viable pasturelands (USAID 2012, Oxfam 2008 and Boku 2010).

Temporal drought variation
The drought condition for rainy months (main and small) of thirty years in different time scale, covering the period of 1986-2016 calculated. From the computation, a mix of dry and wet years has been observed. From the computation, several drought risk incidences with RDI ≤ -1.00 were detected. For both rainy seasons Spring and Autumn, the year from 1999-2000 was the driest in the history of drought in the study area. For the main rainy months, March, April, and May as well as spring (belg) season has shown temporal variation with varying severity and intensity (Fig 2). For instance, May 1986 and 1992 were characterized by severe drought events with RDI ≤-1.5 while spring season of 2000 were characterized by extreme drought event.
Similarly for small rainy months and the autumn season is also impacted by the short dry season (Adoolessa) occurs from June to August in the area. In sum, temporal variation of drought is slightly higher for small rainy season.  and seasons (RDI-3 and RDI-12) timescales were shown in (Table 3)

Fig 6: Spatial pattern of 1-month and 3-month (Autumn Season) RDI
For the small rainy season Tseday (Autumn), RDI of 1-month (September, October and November) and 3-month season also calculated. In 1-month RDI, Septemebr was observed as a month with moderate drought category that covered majority of the watershed unit while October is the month with severe drought category that nearly covered the whole watershed unit (Fig 6) 1986-1987, 1992, 1993, 2000 and 2004. In