Spatiotemporal variability and trends in rainfall and temperature in Alwero watershed, western Ethiopia

Climate analysis at relevant time scales is important for water resources management, agricultural planning, flood risk assessment, ecological modeling and climate change adaptation. This study analyses spatiotemporal variability and trends in rainfall and temperature in Alwero watershed, western Ethiopia. Our analysis is focused on describing spatial and temporal variability of rainfall in the study area including detection of trends, with no attempt at providing meteorological explanations to any of the patterns or trends. The study is based on gridded monthly rainfall and maximum and minimum temperature data series at a resolution of 4 × 4 km which were obtained from the National Meteorological Agency of Ethiopia for the period 1983–2016. The study area is represented by 558 points (each point representing 4 × 4 km area). Mean annual rainfall of the watershed is > 1600 mm. Annual, June–September (Kiremt), March–May (Belg) rainfall totals exhibit low inter-annual variability. Annual and October-February (Bega) rainfalls show statistically significant increasing trends at p = 0.01 level. May and November rainfall show statistically significant increasing trends at p = 0.01 level. March shows statistically significant decreasing trend at p = 0.1 level. The mean annual temperature of the watershed is 25 °C with standard deviation of 0.31 °C and coefficient of variation of 0.01 °C. Mean annual minimum and maximum temperatures show statistically non-significant decreasing trends. Bega season experienced statistically significant deceasing trend in the maximum temperature at p = 0.01 level. The year-to-year variability in the mean annual minimum and maximum temperatures showed that the 2000s is cooler than the preceding decades. Unlike our expectations, annual and seasonal rainfall totals showed increasing trends while maximum and minimum temperatures showed decreasing trends. Our results suggest that local level investigations such as this one are important in developing context-specific climate change adaptation and agricultural planning, instead of coarse-scale national level analysis guiding local level decisions.


Background
Climate change is increasing the occurrence and magnitude of rainfall extremes that cause increased drought and flood risks (Chen et al. 2014). The influence of current climate variability on crop production is large in developing countries like Ethiopia where agriculture is primarily dependent on rainfall (Alemayehu and Bewket 2016a). Rainfall in Ethiopia shows high variability across space and time owing to the complex topography varying from 120 mbsl to 4620 masl (Reda et al. 2015) and latitude of the country varying from 3 to 15 o N and 33 o E to 48 o E (Gamachu 1988). Robust information on the seasonality of rainfall in the country is important to tackle its adverse economic and social consequences including on agriculture (Korecha 2013) and for local level climate change adaptation planning (Alemayehu and Bewket 2017a). The rainfall pattern in Ethiopia is strongly seasonal (World Bank 2006). It is characterized by wet and long rainy seasons on the one hand and long dry and Open Access *Correspondence: arragawalex@gmail.com 1 Department of Geography & Environmental Studies, Debre Berhan University, Debre Birhan, Ethiopia Full list of author information is available at the end of the article short rainy seasons on the other hand (World Bank 2006;Korecha 2013). Consequently, the high and low rainfall phenomena can give rise respectively to flood and drought conditions with adverse economic and humanitarian crises (World Bank 2006). In Ethiopia, the performance of the agricultural sector and the rainfall pattern show strong correlations (Admassu 2004;Lemi 2005;World Bank 2006;Bewket 2009;Conway and Schipper 2011;Alemayehu and Bewket 2016a). Rainfall shortage often leads to famines (Alemayehu and Bewket 2017b).
There are three seasons in Ethiopia. These are (i) the June-September (main rainy season, locally known as Kiremt); (ii) the March-May (short rainy season, locally known as Belg); and the October -February (dry season, locally known as Bega) (Seleshi and Zanke 2004;Alemayehu and Bewket 2017a). Understanding the characteristics of Kiremt and Belg season rainfall is useful for improving agricultural productivity and mitigating food security challenges (Alemayehu and Bewket 2016a, a;Mulugeta et al. 2019) and water resource development of the country (Degefu and Bewket 2014;Suryabhagavan 2017). Kiremt rainfall which is the largest in terms of its amount and geographical coverage is less variable in most parts of the country compared to Belg and Bega season rainfall (Shankoa and Camberlin 1998). It supports the main cropping season production, locally known as Meher. Kiremt rainfall shortage or change impacted the agricultural productivity in the country (Mulugeta et al. 2019).
The Belg season rainfall is characterized by high temporal and spatial variability, which has implications on Belg season crops and food security of households (Alemayehu and Bewket 2016a;Bekele-Biratu et al. 2018). The change and shift in Belg season rainfall leads to devastating droughts affecting socioeconomic welfare and environmental resources (Haile et al. 2019). Alemayehu and Bewket (2017a) summarized previous studies on rainfall and temperature variability and trends in Ethiopia covering different temporal and spatial scales. They concluded that studies on rainfall did not show clear trend for the country. The possible explanations to that are: (i) trends in annual and seasonal rainfalls are affected by topography (Bewket and Conway 2007;Mengistu et al. 2013); (ii) trend analysis is largely sensitive to quality of data, the choice of study periods and stations considered by the different studies (Seleshi and Zanke 2004;Bewket and Conway 2007;Mengistu et al. 2013); (iii) the northsouth oscillation of the Inter Tropical Convergence Zone (ITCZ), the El Ni˜no-Southern Oscillation (ENSO) phenomena (Shanko and Camberlain 1998;Seleshi and Zanke 2004;Gleixner et al. 2017) and fluctuations in sea surface temperatures (SSTs) mask trends (Shanko and Camberlain 1998;Gissila et al. 2004;Segele and Lamb 2005;Viste et al 2012;Jury and Funk 2013); and (iv) previous studies on variability and trends in rainfall and temperature frequently takes place at large scale (Ademe et al. 2020) and contributed to mixed patterns of change.
The focus of this study is to investigate the spatiotemporal variability and trends in rainfall and temperature in Alwero watershed in the western part of Ethiopia using a dense network of 4 × 4 km gridded data (558 points) reconstructed from weather stations and meteorological satellite records which spatially covers the watershed. In countries like Ethiopia where a major underlying vulnerability factor is the heavy dependence of the economy on rain-fed smallholder agriculture (Conway and Schipper 2011;Alemayehu and Bewket 2016b), timely and accurate availability of climate information is useful for sustainable climate risk management (Alemayehu and Bewket 2017a). A better understanding of the changes in long-term annual and seasonal rainfall at local level is useful for adaptation intervention (Mulugeta et al. 2019). Local level investigations such as this one are important in developing context-specific climate change adaptation and agricultural planning, instead of coarse-scale national level analysis guiding local level decisions.
The study area in particular is the largest agricultural investment hub in the country where a detailed previous study does not exist. In the following section, we present a brief description of materials and methods of the study, and this is followed by the results and discussion section. Conclusions are presented in "Data and methods Section".

Description of the study area
Alwero watershed is found in the Abobo district of the Gambella National Regional State, western Ethiopia (Fig. 1). The Gambella Region is largely lowland (Kolla) agroecological zone with hot to warm humid climate. Mean annual rainfall is about 1200 mm. Mean annual temperature is 26.7 °C (Degife et al. 2019). The level of natural resource use is low due to sparse population density. The soil, topography and climate conditions have made the Region to be one of the most fertile parts of the country and highly suitable for growing various types of tropical crops. In the early 2000s, about 3.5 hectares of land from southwest part of the country which is also the present study area were leased to both domestic and foreign investors (Rahimato 2011). Since 2008, the Region was prioritized as one of the biggest destination for large scale agricultural investment.
More than 1.2 hectares of land were transferred to both domestic and foreign investors (Rahimato 2011).
The altitude of Alwero watershed ranges from 385 to 2531 m asl (Fig. 1). About 58 and 40% of the watershed lies in altitude ranges of 385-500 m asl and 500-1837 m asl, respectively. Only about 2% of the watershed contains hills and escarpments located in the eastern edge of the watershed with an altitude range of 1837 to 2531 m asl. The eastern edge of the watershed is characterized by steep and very steep slopes whereas the central part is dominated by a mix of flat, gently sloping and sloping terrain. Alwero watershed has two sub-watersheds; upper (eastern) and lower (western). The altitude of the upper (eastern) sub-watershed ranges from 619 to 2531 m asl. The altitude of the lower (western) sub-watershed ranges from 385 to 619 m asl.
According to the Central Statistical Authority (CSA) (2013), the total population of Abobo district is 22,420 out of which 11,531 are males and 10,889 are females. Of the total, about 35% resides in urban area (Abobo town) and the remaining 65% lives in rural areas. The livelihood of the population in the watershed is largely dependent on small-scale farming, fishing, retailing and casual jobs in large-scale farms, construction and government and non-government offices.

Data and methods
The study is based on gridded monthly rainfall and temperature data series at a resolution of 4 × 4 km for the period 1983-2016. The study area is covered by 558 points (each representing 4 × 4 km area). The data are created by blending station records with meteorological satellite estimates and were obtained from the National Meteorological Agency of Ethiopia. This is the best available dataset for the country which is homogeneous and recommended for climate analysis (Dinku et al. 2014). In recent years, gridded rainfall and temperature datasets have been used in Ethiopia with different data sources and spatial resolutions (Mengistu et al. 2013;Dinku et al. 2014;Alemayehu and Bewket 2017a;Mulugeta et al. 2019;Gebrechorkos et al. 2020;Ademe et al. 2020; Matewos and Tefera 2020).
Coefficient of variation (CV), precipitation concentration index (PCI) and standardized rainfall anomaly (SRA) are used to examine inter-annual and intra-annual variability of rainfall. The CV measures year-to-year variation in the data series. According to NMA (1996) classification, CV less than 0.20 is less variable, CV between 0.20 and 0.30 is moderately variable and CV greater than 0.30 indicates high variability. The PCI values measure extent of seasonality of rainfall; and it is given as indicated in De Luis et al. (2000): where: Pi = the rainfall amount of the ith month; and Σ Pi 2 = summation over the 12 months. PCI values of less than 10 indicate quite uniform monthly distribution of rainfall, values between 11 and 20 indicate (1) PCI = 100 × Pi 2 / Pi 2 moderate concentration, and values above 21 indicate high concentration. SRA is commonly used as a simple index to characterize drought at different time scales, or to identify abnormal wetness or dryness (Guttman 1999). SRA values are given as; where SRA = standardized rainfall anomaly, Pt = annual rainfall in year t, Pm = is long-term mean annual rainfall over a period of observation and σ = standard deviation of annual rainfall over the period of observation. According to Agnew and Chappel (1999), SRA values indicate extreme drought (SRA < − 1.65), severe drought (− 1.28 > SRA > − 1.65), moderate drought (− 0.84 > SRA > − 1.28), and no drought (SRA > − 0.84). Linear regression was used to detect trends in rainfall. It is given as; where y is dependent variable, m is the slope, x is independent variable and b is the intercept. F-distribution test was applied to determine the statistical significance of the trends. The minimum and maximum temperatures were also analyzed using the above techniques. Surface data were generated from the gridded monthly rainfall and temperature data using kriging interpolation technique with the help of ArcGIS 10.1.

Annual and seasonal rainfall patterns
Mean annual rainfall of Alwero watershed is 1665.5 mm (Table 1), and its inter-annual variability is low with a CV of 8.7%. The Kiremt and Belg rainfall also show low interannual variability (CV, 9.2% and 19.7%, respectively). Bega rainfall shows moderate variability (CV, 24%). Previous studies on variability and trends of annual and seasonal rainfall reported moderate and high variability in their respective study areas (Bewket and Conway 2007;Ayalew et al. 2012;Alemayehu and Bewket 2017a, b;Hadgu et al. 2013;Urgessa 2013;Ademe et al. 2020).
As indicated by the PCI values, rainfall shows moderate concentration in few months of the year (10.8). It can be observed that above average mean monthly rainfall is recorded from May to October. These months have recorded above 130 mm rainfall (Fig. 2). About 80% of annual rainfall is concentrated in these six months which ranges from 11% (October) to 14.4% (August). August supplies the largest amount of rainfall in the watershed. July and September also represent high contributions to the annual totals. Our results corroborate the findings of (Alemayehu and Bewket 2017a). ( As elsewhere in Ethiopia, Kiremt rainfall contributes the largest to the annual rainfall (56%), followed by Belg rainfall (25%). Unlike other parts of the country, the contribution of Bega rainfall is high (19%) as well. The highest monthly rainfall is 395.9 mm which accounted for 22% of the annual and 89% of Belg rainfall totals, respectively (Table 1). Unlike other areas of the country where high concentration of rainfall is recorded in the month of August followed by July, the highest monthly rainfall is recorded in May which is a dry month in most parts of the country. However, a recent study by Matewos and Tefera (2020) in the drought-prone districts of rural Sidama, central rift valley region of Ethiopia observed that April and May are the wettest months.
Annual and Bega rainfalls show statistically significant increasing trends at p = 0.1 and p = 0.01 levels, respectively. Kiremt and Belg rainfalls show no significant trends. These results contradict the findings of Shanko and Camberlin (1998) Matewos and Tefera (2020) who reported declining trends of annual and seasonal rainfall totals in their respective study areas in different parts of the country. The statistically significant increasing trend of Bega rainfall supports the findings of a recent study by Matewos and Tefera (2020) who reported a significant increasing trend of Bega rainfall in the drought-prone districts of rural Sidama, central rift valley region of Ethiopia.
Regarding monthly rainfall trends, November and May showed statistically significant increasing trends at p = 0.05 level. October shows significant increasing trend at p = 0.1 level, and March shows significant decreasing trend at p = 0.1 level (Fig. 2). June to September show non-significant increasing trends. December, January and February showed statistically non-significant increasing trends. March contributes the highest to the overall declining trend of annual rainfall while November makes considerable contribution to the overall increasing trend of annual rainfall.
In Ethiopia the driest year is 1984, which is the wellknown drought year. The wettest year is 2006, which is the major flood year. The country has witnessed large  Standardized monthly rainfall anomalies were also computed for the wettest and driest periods to assess extents of wetness and dryness in the watershed. As shown in Table 1, the driest and wettest years over the period of observation are 1986 and 2011, respectively. In the driest period the highest and the lowest SRA values are 1.45 and − 1.25, respectively, while the highest and the lowest SRA values during the wettest period are 1.26 and − 1.19, respectively. The highest and the lowest SRA during driest years are observed during June and January, respectively. May and February observed the highest and the lowest SRA values during the wettest years. About 67% of observations showed below average rainfall during the driest period, but only 25% of the observations fall under the different drought categories. About 50% of the observations showed below average rainfall during the wettest period. From these, nearly 33% fall under the different drought categories. This shows that rainfall concentration is not uniform and few months remain dry even in years of good rain conditions. Figure 4 and Table 2 show the spatial distribution of annual rainfall over the period of analysis. East and southeast part of the watershed receive high amount of annual rainfall. More than half of the watershed receives annual rainfall of 1550-1700 mm. Another large proportion of the watershed (26%) receives annual rainfall of between 1411 and 1550 mm. More than 20% of the watershed receives annual rainfall of between 1700 and 1850 mm. About 1% of the watershed receives annual rainfall of > 2000 mm. The spatial distribution of Belg and Bega seasons reveals similar pattern in which rainfall increases from west to east. Almost all parts of the lower sub watershed receive Kiremt rainfall > 950 mm over the period of observation. The northern part of the upper sub watershed receives Kiremt rainfall of 768-850 mm.

Trends in annual and seasonal rainfall in the upper (eastern) sub-watershed
Mean annual rainfall of upper sub-watershed is 1686.2 mm with coefficient of variation of 8.9% which Fig. 4 Spatial distribution of annual and seasonal rainfall in Alwero watershed Alemayehu et al. Environ Syst Res (2020) 9:22 shows low inter-annual variability. Kiremt and Belg rainfalls show low inter-annual variability of 9% and 19%, respectively. Only Bega rainfall shows moderate interannual variability with coefficient of variation of 24%.
Kiremt rainfall contributes the largest to the annual rainfall (55%), which is followed by Belg (26%) and Bega (19%), respectively. The driest and wettest periods over the period of observation are 1986 (1419.3 mm) and 2016 (1987 mm), respectively. Annual rainfall shows statistically significant increasing trend at p = 0.05 level, while Bega rainfall also shows statistically significant increasing trend at p = 0.01 level. Belg and Kiremt rainfalls show statistically non-significant increasing and decreasing tendencies, respectively (Fig. 5).
More than 85% of the upper sub-watershed receives annual rainfall of between 1411 and 1850 mm. The other 12% of the upper sub-watershed receives annual rainfall of between 1850 and 2000 mm. The remaining 2% of the upper sub-watershed receives annual rainfall of > 2000 mm ( Fig. 4 and Table 2).
Annual rainfall shows considerable inter-annual variations since 2000. Positive annual rainfall anomalies accounts for 50% of the total observations for the period 1983-2016. The 1980s is the driest decade. Bega rainfall anomalies show similar pattern with annual rainfall anomalies. Positive Bega rainfall anomalies accounts for 50% of the total observations. Kiremt rainfall shows considerable inter-annual variations throughout the period of observation. Kiremt rainfall has the smallest proportion of positive anomalies (44%). The 2000s is the driest decade. Belg rainfall shows considerable inter-annual variations. However, large proportion of positive rainfall anomalies is observed (56%). Since 2007 positive rainfall anomalies are observed except the years 2010 and 2012.

Trends in annual and seasonal rainfall in the lower (western) sub-watershed
Mean annual rainfall of the lower sub-watershed is 1600.6 mm. Annual rainfall shows low inter-annual variability (10.9%). Kiremt rainfall also shows low inter-annual variability (12.5%). Belg and Bega rainfalls show moderate inter-annual variability with coefficient of variation of 29 and 32%, respectively. Kiremt rainfall contributes the largest to the annual rainfall (67%). Belg and Bega contribute the remaining 19 and 14% of annual rainfall, respectively. The driest and wettest periods over the period of observation are 1995 (1179 mm) and 1991 (1857 mm), respectively. Annual and Kiremt rainfalls show statistically non-significant increasing trend, while Bega rainfall shows statistically significant increasing trend at p = 0.01 level (Fig. 5).
Large part of the lower sub-watershed (75%) receives annual rainfall of between 1411 and 1550 mm, and about 23% receives annual rainfall of between 1550 and 1700 mm. A small portion of the lower sub-watershed receives annual rainfall of between 1700 and 1850 mm ( Fig. 4 and Table 2).
Annual rainfall shows considerable inter-annual variations from 1983-2009. Positive annual rainfall anomalies accounts for 59% of observations. The 2000s is the wettest decade. Positive anomalies are observed since 2007 except the year 2009. The 2000s is the driest decade for Belg and Bega rainfall. The proportion of positive and negative anomalies for Belg and Bega rainfalls are 44% and 56% of the observations, respectively. Positive anomalies for Kiremt rainfall account for 53% of observations. The 1990s is the wettest decade.

Temperature trends in Alwero watershed
In the watershed, the mean annual temperature is 25.0 °C. Over the study period, the lowest and highest mean annual temperatures are experienced in 2016. The lowest temperature is 14.5 °C (January) with standard deviation of 0.096 and coefficient of variation of 0.05. The highest temperature is 39.3 °C (March) with standard deviation of 0.1 and coefficient of variation of 0.03. Mean annual minimum temperature ranges from 16.7 °C (2001) to 18.5 °C (1984), and the long term mean is 17.5 °C with standard deviation of 0. 4 and coefficient of variation of 0.03. Mean annual maximum temperature ranges from  (1999), and the long term mean is 32.5 °C with standard deviation of 0.46 and coefficient of variation of 0.01.
Unexpectedly, the mean annual minimum and maximum temperatures show statistically non-significant declining trends. Statistically non-significant but decreasing trends in all seasons are also observed in the minimum temperature. Bega season experienced statistically significant deceasing trend in the maximum temperature at p = 0.01 level. Belg season experienced statistically non-significant but decreasing trend in the maximum temperature. Statistically non-significant but increasing trend is also observed for the Kiremt (Table 3).
Our finding of declining trends in the mean maximum and minimum temperatures contradicts the results reported by NMA (2007), McSweeney et al. (2008), Taye and Zewdu (2012), Ayalew et al. (2012), Mengistu et al. (2013), Mekasha et al. (2014), Addisu et al. (2015), Alemayehu and Bewket (2017a), Asaminew et al. (2017), Suryabhagavan (2017), Haile et al. (2019) and Matewos and Tefera (2020) who reported warming trends in the minimum and maximum temperatures in different parts of the country. Our research is unable to explain this result. Figure 6 shows the year-to-year variability in the mean annual minimum temperature. The 1980s was the warmest decade compared to the 1990 and 2000s. Positive and negative anomalies account for 38% and 62% of the total observations, respectively. Seasonal anomalies in the minimum temperature have similar patterns to anomalies in the annual minimum temperature. Kiremt season anomalies are positive since 2005 except the years 2008 and 2011. The maximum number of negative anomalies is observed in Bega season (59%).
The year-to-year variability in the mean annual maximum temperature is shown in Fig. 7. From 1998From -2005 standardized anomalies are positive, and then negative  Figure 8 shows the spatial distribution of mean maximum and minimum temperatures. The mean maximum and minimum temperatures decrease from west to east.   (Tables 4 and 5 and Fig. 9). Seasonal anomalies in the maximum and minimum temperatures have similar patterns to mean annual maximum and minimum temperatures of the respective watersheds.

Temperature trends in the lower (western) sub watershed
The mean annual temperature in the lower sub watershed ranges from 26.6 °C (2009) to 28.13 °C (1985). The mean annual temperature over the period of observation is 27.39 °C with standard deviation of 0.38 °C and coefficient of variation of 0.01 °C. The highest mean maximum temperature is 36.59 °C (1991). Conversely, the lowest mean minimum temperature in the lower sub watershed over the period of observation is 18.61 °C (1990). The mean annual and seasonal maximum temperatures show decreasing trends. The decreasing trends in Bega season and annual maximum temperatures are statistically significant at p = 0 01 and p = 0.05 levels, respectively. The mean annual  and seasonal minimum temperatures show statistically non-significant declining trends except Bega season which shows statistically non-significant but increasing trend. Large proportion of the lower sub watershed (93%) experiences mean minimum temperature of 18-21 °C while the remaining 7% of the lower sub watershed experiences mean minimum temperature of 16-18 °C. Almost 96% of the area experiences mean maximum temperature of 33-36 °C. The remaining 4% of the lower sub watershed experiences mean maximum temperature of 30-33 °C (Tables 4 and 5 and Fig. 9).
The innovative element of this study is the use of gridded monthly rainfall and maximum and minimum temperature data series at a resolution of 4 × 4 km which are created by blending station records with meteorological satellite estimates. This is the best available dataset for the country recommended for climate analysis. The methodology used is well established, standard, proven, and widely used by many researchers. It has the advantage of making good results. However, the methodology and results are new to the study area. Our analysis focuses on local scale climatology and the findings are useful for climate risk management. The findings can be applicable to other parts of rural Ethiopia as well. In particular, the analysis is useful for irrigation and investment as the area is the largest agricultural investment hub in the country.