Variability and time series trend analysis of temperature over 1981-2018 in semi-arid Borana zone, southern Ethiopia

Background: Understanding the climate variability at local scale could help suggest local adaptation responses to manage climate driven impacts. This paper analyzed the variability and trends of temperature over the period 1981-2018 in semi-arid Borana zone of southern Ethiopia using Mann-Kendall (MK) test and inverse distance weighted (IDW) interpolation technique. Gridded (4 km * 4 km) daily temperature data was used to study variability at temporal and spatial scales. Results: The results revealed that monthly temperature shows a warming trend where February was the warmest month for both maximum and minimum temperature. Seasonally, the highest maximum and minimum temperatures were observed during Bega. Minimum temperature shows a warming trend during all seasons unlike maximum temperature. Both minimum and maximum temperature shows not significant warming trend at annual timescale. The later decades (20012018) have shown a warming trend compared to a period ahead especially for minimum temperature. The southwestern and southeastern areas across the zone were warmer than any other areas in the region during the studied period. Conclusion: Temperature shows variability at shorter than longer timescales. There is a pronounced warming trend for minimum than maximum temperature. Warming condition advances from the northcentral parts towards the southwestern and southeastern areas. Internal variability was observed at temporal and spatial scales and therefore any adaptation responses to local climate variability should consider the microscale climate.


Introduction
The recent report of Intergovernmental Panel on Climate Change (IPCC) indicate that the global mean temperature showed a warming trend of 0.85 0 C (0.65-1.06) over the period 1880(IPCC 2013Birara et al. 2018). Many developing countries particularly those found in Sub-Saharan Africa are significantly affected by the global average temperature rise, termed as global warming and its consequences (IPCC 2012). The consequences of changing climate are widespread in these countries whose economy is in one way or another depend on climate sensitive sectors such as agriculture and livestock systems. Climate variability and change affects these systems by altering the pattern and distribution of climatic elements including temperature and rainfall.
Ethiopia is one of the largest countries of Africa characterized by diversified physiography.
There is a marked altitudinal variation that ranges from 125 m below mean sea level in Dallol Depression in the northeastern parts to 4,620 m above mean sea level indicating peak of Mt Ras Dashen in the north central highland group of the country. Climate of Ethiopia is highly modified by altitude than other factors where lowlands are known for warm to hot arid climate while the highland regions are characterized by cool to cold sub humid (humid) type of climate.
In addition to complex topography, the climate of the country is influenced by the seasonal migration a low-pressure zone namely Intertropical Convergence Zone (ITCZ), a region where the northeast and the southeast trade winds converge, following the overhead Sun (Segele and Lamb 2005;Korecha and Barnston 2007). The seasonal movement of ITCZ is, therefore, responsible for the occurrence of dry and wet conditions over the country during different seasons. The spatial and temporal variability in the climate system (temperature and rainfall) is hence governed by these factors where this variation can have a potential impact on various socio-economic systems such as agricultural systems, livestock systems, etc. Apart from this, climate variability and extreme events like drought exacerbates the conditions in pastoral and agropastoral communities inhabiting in lowland regions.
Climate variability study gained careful attention and the most important parameters either temperature and/or rainfall were studied at national/sub-national scale to watershed level. A range of studies has been carried out in this aspect (Gebrehiwot and van der Veen 2013;Jury and Funk 2013;Mengistu et al. 2014;Fazzini et al. 2015;Wassie and Fekadu 2015;Biraraet al. 2018;Asfaw et al. 2018;Wedajo et al. 2019;Tesfamariam et al. 2019;Esayas et al. 2019;Gedefaw et al. 2019;Mekonnen and Berlie 2020;Berhane et al. 2020;Alemayehu et al. 2020;Bayable and Alemu 2021;Belay and Demissie et al. 2021). Some of these studies used satellite-based data sources which provides long-term data over a century but coarse resolution; while others studied large-scale spatial variability such as country level or beyond.
On top of that, most of the papers were conducted in the highland areas of northwestern, northcentral, northeastern and western regions where the issue of temperature variability is not a main concern. The limitation with the usage of coarse resolution climate data and larger scale spatial analysis could be the difficulty of observing local level variability. This means the existing internal variability might not be captured by employing these data types. In the context of Ethiopia, the various agroecological zones respond differently to climate variability and hence, lowland areas with warm (hot) semi-arid (arid) type of climate are highly sensitive to climate variability than highland areas. Borana zone of southern Ethiopia is one of the regions frequently affected by climate variability and extremes where local scale variability study is lacking especially in this area. Therefore, it is ideal to emphasize on climate sensitive geographical regions. This paper is intended to study the variability and timeseries trend analysis of maximum and minimum temperature in the semi-arid Borana zone over the period 1981-2018. The research is conducted based on the fact that, understanding the climate variability as local scale could help suggest local adaptation responses to manage climate driven impacts. It also helps local communities, actors and decision makers to take planned intervention measures so as to reduce climate associated risks. Apart from these, the study result add value to the existing literature especially in the Borana where study on spatiotemporal variability of climate is lacking at high spatial resolution.

Study area description
This study was conducted in Borana zone which is one of the 21 administrative zones of Oromia regional state, Southern Ethiopia. Borana zone is located in the southern part of the country bordered by Kenya in the South, West Guji zone in the North, Somali region and Guji zone in the East and South Nations and Nationalities Peoples' Region (SNNPR) in the West.
Astronomically, the study area stretches from 3 0 30' N to 5 0 25' N latitude and 36 0 40' E to 39 0 45' E longitude. Yabelo is the capital town of Borana zone and located at about 570 kilometers South of Addis Ababa. The zone covers almost 48,360 km 2 out of which more than 75% is a lowland.
The study area exhibits four seasons namely Bega / 'Bona' the long dry period from December to February, Belg / 'Ganna' the long rainy period from March to May, Kiremt / 'Adolessa' the short dry spell from June to August and Meher / 'Hagayya' the short rainy period from September to November. The rainfall pattern of the region is different from most parts of the country. It is during Belg and Meher seasons that Borana zone receives most of its rain. The The region has a semi-arid savannah landscape, marked by gently sloping lowlands and flood plains vegetated predominantly with grass and bush land. The geology is composed of a crystalline basement with overlying sedimentary and volcanic deposits (Gemedo-Dalle et al. 2006;Lasage et al. 2010;). People are predominantly involved in small-scale subsistence agriculture production and mainly on livestock husbandry. These sectors are climate-sensitive and frequently hit by climate related hazard, which is of course drought. Small-scale farming is not widely practiced mainly due to the aridity that prevails over the study area and hence government introduced the farming practices as means of income diversification and to support the family.

Data types and sources
Gridded (4km * 4km spatial resolution) data for daily maximum and minimum temperature precipitation for all the points lying within the study area boundary for the period 1981 to 2018 were collected from National Meteorological Agency (NMA). Therefore, a total of 2,702 data points ( Figure 2) were considered as inputs and the mean values were used to analyze the variability and trends of rainfall at multiple timescales including monthly, seasonal, annual and decadal time periods. Therefore, the data generated were prepared for use in R software package to test trend and variability analysis. We prefer to use gridded data for a number of advantages including its accessibility and completeness. On the other hand, due to the remoteness of the location, meteorological stations are sparsely distributed in the study area with serious missing values in the dataset.

Serial correlation
One of the challenges in detecting and interpreting trend in timeseries data is the existence of serial correlation (autocorrelation), is where error terms in a time series transfer from one period to another (Yue et al. 2002;Birara et al. 2018). In other words, the error for one time period 'a' is correlated with the error of a subsequent time period 'b'. Autocorrelation is tested in this paper through calculating the autocorrelation coefficient at lag-1 and plotting the correlogram. It is said that, there is significant autocorrelation when the value for correlation coefficient, r, falls outside the range at 95% confidence interval. Therefore, we took a remedial measure to remove the effect of significant autocorrelation in the timeseries through the 'pre-whitening' procedure (Von Storch 1995). Based on this, for the data points (x1, x2, x3, …, xn), the 'pre-whitened' time series was obtained through (x2 -rx1, x3 -rx2, …, xn -rxn-1) procedure before applying Mann-Kendall trend test.

Mann-Kendall test
Mann-Kendall (MK) test, a popular non-parametric test is used in order to detect trend in climatic variables at 5% level of significance (Mann 1945;Kendall 1955). Non-parametric methods are more suitable to the detection of trend in hydrological data (Helsel et al. 2002). MK test is a rank-based test, used where autocorrelation is not significant and it can tolerate to outliers, distribution free and has higher power than the other test (Duhan and Pandey 2013). MK test was then proposed as the null hypothesis (H0), there is no trend in the time series and alternative hypothesis (H1), there is a monotonic trend which can either be an upward or a downward.
The MK test (Mann 1945;Kendall 1955) was first carried out by computing S statistic as: ( 1) where, n is the number of observations, and xj is the j th observation and Sgn denotes the sign function, defined as: (2) and variance defined by: where, n is the number of data, m is the number of tied groups (a tied group is a set of sample data with the same value), and it is the number of data points in the k th group.
Finally, the statistics of this test, designated by Z, is computed as: The test statistics Z is used as a measure of significance of trend. If the value of Z is positive, it indicates increasing trends, while negative values of Z show decreasing trends. A significance level of α = 0.05 (confidence level of 95%), is also utilized for testing either upward or downward monotonic trend (a two-tailed test) (Jhajharia et al 2012). If Z appears greater than Zα/2 where α depicts the significance level, then the trend is considered as significant.

Sen's Slope Estimator
If a linear trend is present in a time series, then the true slope (change per unit time) can be estimated by using a simple non-parametric procedure developed by Sen (1968). Sen's slope estimation can be calculated in the form of: where Q is the slope and b constant. To obtain slope Q in Equation 5, it is necessary to calculate the slope for all data with the equation: where, xj and xk are considered as data values at time j and k (j>k) correspondingly. The median of N values of Qi is ranked from small to large, with an estimated Sen's estimator of slope, given by: To obtain estimates of b in Equation 5, the values of n data from the difference (xi -Qti) are calculated. The median value is the estimate for b. Finally, Qmed is computed by a two-sided test at α = 0.05 (95% confidence interval) and then a true slope can be obtained and its value indicates the steepness of the trend.

Statistical data analysis
The study tried to capture the variability of maximum and minimum temperature at temporal and spatial basis. Temporally, variability is seen at monthly, seasonally, annually and decadal scales.
Descriptive statistics including mean, standard deviation and coefficient of variation were calculated at different scales for the parameter. Hare (2003) computed coefficient of variation (CV) using the following formula: Where CV represents the coefficient of variation, σ is the population standard deviation, and µ is the population mean.
Apart from ArcGIS which is used for spatial data analysis and R-software package for statistical and trend tests while Origin software version 17 was used to plot various temperature graphs in this study.

Analyzing spatial variation
The inverse distance weighted (IDW) interpolation technique in ArcGIS was employed to generate surface data for both maximum and minimum temperature from grid points at different temporal scales. To do this, time series temperature data from the grid points were analyzed by using ArcGIS 10.5 interface. Hence, spatial maps showing maximum and minimum temperature variability across the Borana lowland were produced at seasonal and annual timescales.

Variability and trends of monthly maximum and minimum temperatures
The mean maximum and minimum temperatures for the last 38 years were computed for the Borana lowland. The results show that, February was the hottest month (32.06 0 C) followed by January (31.49 0 C) and March (31.03 0 C) whereas the lowest mean maximum temperature was observed during the month of July (28.01 0 C). For most of the months, the trend for mean maximum temperature shows an upward trend where it is statistically significant at 95% level of confidence (P-value < 0.05) only for August whereas it shows not statistically significant decreasing trend (P-value > 0.05) for April, May and October (Table 1 and Figure 3).
On the other hand, the highest mean minimum temperature was observed in February (16.92 0 C) followed by March (16.69 0 C) during the entire period. Apart from this, the lowest mean minimum temperature was recorded in July (15.77 0 C) and August (15.93 0 C). The MK and Sen's slope test results revealed that all months except April shows an increasing trend where it is statistically significant for January, February, August and October at 95% level of significance (P-value < 0.05). In addition to this, the decreasing trend for the month of April is not statistically significant (P-value> 0.05) as observed in Table 1.   Meher season exhibited statistically significant increasing trend (P-value < 0.05) at 95% level of significance as indicated in Table 2.

Variability and trends of annual mean maximum and minimum temperature
The annual mean maximum temperature during the last 38 years was 29.66 0 C, where the highest mean maximum was 30.49 0 C (1994) and the lowest mean maximum was 28.68 0 C (1985). On the other hand, the mean minimum temperature for the same period was 16.31 0 C where the highest and lowest means observed were 16.93 0 C (2016) and 15.73 0 C (1985) respectively as shown in Table 3 and Figure 7. The range of temperature between the highest and lowest mean values over the studied periods were 1.81 0 C for maximum and 1.2 0 C for minimum temperature.
The test results also proved that, the trend of both annual mean maximum and minimum temperature shows an increasing trend but not statistically significant.  Therefore, the recent decade was a bit warmer than the preceding as graphically presented in Figure 9.

Discussion
Both mean maximum and minimum temperature were studied at various temporal scales covering monthly, seasonal, annual and decadal timescales as well as spatially covering the entire semi-arid Borana zone in this paper. The highest average maximum temperature was observed in February while the lowest was in July. The mean monthly maximum temperature during the study period was 29.66 0 C and it showed an increasing trend for most of the months where the increase was statistically significant only for August (P-value = 0.003). April, May and October showed not significant decreasing trend during the studied period. The mean monthly minimum temperature in Borana was 16.31 0 C, where the highest value was observed in February followed by March for the studied period. Apart from this, the lowest mean minimum temperature was recorded in July and August. The same months revealed the highest and the lowest values for both mean maximum and minimum temperature and these months were February and July. The range of temperature between the highest and lowest mean minimum temperature was estimated to 1.15 0 C, which is by far smaller than the value for mean maximum temperature. The mean minimum temperature shows a significant increasing trend for January, February, August and September but in April, the trend was decreasing and also not statistically shows not significant warming trend only for minimum temperature. Therefore, the later periods are warmer than the previous time periods. In general, temperature shows a slight variation at shorter timescales (months and seasons) than longer timescales (annual and decadal) and the later period show warmer condition than the previous in Borana lowland. Mekonnen and Berlie (2020) also observed the increasing trend of decadal minimum (0.098 0 C), maximum (0.041 0 C) and average (0.069 0 C) temperatures in the northeastern highlands. Asfaw et al. (2018) found the change of temperature to be 0.046, 0.067 and 0.026 0 C per decade for mean, minimum and maximum respectively during the period of 1901-2014 in the north central Ethiopia.
The spatial distribution of temperature during the various seasons shows similarity across the lowland for mean minimum temperature than maximum temperature. In the case of maximum temperature, Meher and Bega seasons for months September through February shows less variability as well as Belg and Kiremt (March to August) shows uniformity across the lowland.
From this it can be concluded that, temperature is not uniformly distributed across the Borana lowland. The mean distribution of mean maximum and minimum temperature shows that, warmer conditions were observed over the southwestern and the southeastern areas of the lowland respectively. In addition, relatively cooler conditions were observed for both maximum and minimum temperature in the north and northcentral areas of the study area whereas moderate temperature was observed in so many areas including eastern central and western parts of the Borana lowland.

Conclusion
This study presented the variability and time series trend analysis of observed maximum and minimum temperature for the time period extending from 1981-2018 in semi-arid Borana zone of southern Ethiopia. The observed monthly mean maximum and minimum temperature in the study area were 29.66 0 C and 16.31 0 C respectively. February and July were the warmest and coldest months respectively for maximum as well as minimum temperature. Most of the months have shown an increasing trend for both maximum and minimum temperature. Bega season was the warmest for both maximum and minimum temperature. All seasons have shown a nonsignificant increasing trend for minimum temperature than maximum temperature whereas Meher season shows a significant increasing trend for minimum temperature. The year 1994 and 2016 the warmest years for the respective mean maximum and minimum temperatures. The annual timescale has shown a not increasing trend for both maximum and minimum temperature in the study area. Temperature is less variable at decadal scale where the decade (2001-2010) has shown a significant warming trend for both maximum and minimum temperatures. Prior to that temperature showed a non-significant decreasing trend. It can be concluded that, temperature is highly variable at shorter timescales (monthly and seasonally) than at longer timescales (annual and decadal) in the study area. The minimum temperature pronounced more warming trend than maximum temperature. Temperature also varies across the semi-arid Borana where temperature have shown an increment from northcentral parts to the southwestern for maximum temperature and southeastern for minimum temperature. This spatial variability is attributed to the nature of topography in the study area which could modify the local climate. Therefore, any intervention measure such as adaptation planning in response to local climate variability should take in to account the temporal and spatial variability at microscale.