Climate change and its implications for rainfed agriculture in Ethiopia

The world's climate has been changing for several thousands of years with a widespread impact on human and natural systems (Kotir 2011; Marohasy et al. 2017; Birara et al. 2018; Yadav 2018). However, its change has become more rapid and unusual in recent decades relative to that of the past, which can be shown through observations of increasing temperature, sea-level rise, increase in the emission of greenhouse gases (GHGs), frequent floods and droughts, and changes in the amount, distribution and patterns of rainfall (Asare-Nuamah & Botchway 2019). Although climate change is global in its extent and impacts, Africa has been identified as the most vulnerable continent to climate change due to low adaptive capacity and high reliance on climate-sensitive sectors such as rain-fed agriculture (Conway & Schipper 2011; Kotir 2011; Calzadilla et al. 2013; Gebrechorkos et al. 2019; Girvetz et al. 2019). Rainfall variability and warming of temperature are being perceived as the two most important variables of climate change, imposing a crippling effect on the productivity of the agricultural sector and sustainable economic development in Africa, particularly in sub-Saharan African (SSA) countries (Carter & Parker 2009; Conway & Schipper 2011; Calzadilla et al. 2013; Serdeczny et al. 2017; Abera et al. 2018; Asfaw et al. 2018; Gebrechorkos et al. 2019).

Ethiopia is one of the SSA countries that is extremely vulnerable to the impacts of climate change and variability (Conway & Schipper 2011; Birara et al. 2018). Recurrent droughts combined with changes in the amount and spatial distribution of seasonal and annual rainfall are among the major climate-related disasters in Ethiopia (Seleshi & Zanke 2004; Korecha & Barnston 2007; Addisu et al. 2015; Zeleke et al. 2017; Weldearegay & Tedla 2018), thereby significantly affecting the productivity of rainfed agriculture and the economic and social development of the country. Agriculture is the main sector of the Ethiopian economy supporting about 49% GDP, more than 80% employment, and 80% foreign exchange (CSA 2014). With only about 5% of the total cultivated land being under irrigation (Awulachew & Ayana 2011), agriculture in Ethiopia heavily depends on natural rainfall (Addisu et al. 2015). Therefore, any change in the amount and distribution of rainfall would severely threaten agricultural productivity, and have immediate implications for food production and security across the country.

The variability of monsoon rainfall in Ethiopia, mainly in the north-western part of the country, is mostly associated with the El Niño-Southern Oscillation (ENSO) events, the warming of equatorial waters in the Pacific Ocean (Korecha & Barnston 2007; Richman et al. 2016). The warming of the ocean temperature leads to the variation in pattern and distribution of rainfall, particularly that of monsoon or summer rain, and changes the direction of the Inter-Tropical Convergence Zone (ITCZ) by altering wind pattern and moisture fluxes (Kotir 2011). According to Gleixner et al. (2017), ENSO events combined with sea surface temperature (SST) anomalies in the Southern Atlantic and Indian Oceans significantly affect the rainfall distribution in Ethiopia by displacing and weakening the rain-producing air mass. This evidence provides a picture of how Ethiopia is highly exposed to changes and variations in rainfall and temperature at varying space and time. It has also been predicted that increasing rainfall and surface temperature in most of the East African countries will highly affect the availability of water and agriculture in Ethiopia (Fentaw et al. 2018; Muhati et al. 2018). To this end, a clear understanding of the temporal trends and spatial distribution of past and projected rainfall and temperature is crucial for proper planning and decision making. As the government of Ethiopa strives to expand agricultural production, reliable and timely climate change information is essential for planning and formulation of appropriate mitigation mechanisms.

While several studies have been conducted on observed climatic varability in many parts of Ethiopia (Bewket & Conway 2007; Wagesho et al. 2013; Degefu & Bewket 2014; Addisu et al. 2015; Asfaw et al. 2018; Birara et al. 2018; Mekonnen et al. 2018), scant evidence is available on future rainfall and temperature projections and change analysis using CMIP5 dataset. The earlier studies on climate variability and changes in Ethiopia have also been focused either on regional or national levels (e.g. Seleshi & Zanke 2004; Korecha & Barnston 2007; Segele et al. 2009; Jury 2010; Jury & Funk 2013; Fazzini et al. 2015; Gleixner et al. 2017; Lewis 2017) putting less emphasis on local scale or catchment-level climate analysis. Local scale studies on climate trends and their implications are imperative to provide valuable guidance for policymakers on how to prepare for climate change adaptation and mitigation, better use and management of water resources, and agricultural developments. Therefore, the main aim of this study is to understand and analyse the variability and changes of past and future temperature and rainfall conditions and their implications for rainfed agriculture in the Rib watershed, northwestern highland Ethiopia. The specific objectives include: (1) to assess the variability and trends of rainfall and temperature over the last few decades; (2) to project the rainfall and temperature changes in the next few decades; and (3) to analyze the possible impact of climate variability and changes on the rain-fed agriculture in the Rib watershed.

The Rib watershed is located in north-western highland Ethiopia between 11°40′ and 12°20′ north latitude and 37°30′ and 38°20′ east longitude with a drainage area of 1,975 km². The area is characterized by irregular topography, valleys, and gorges with elevation ranging from 1,758 m in the western part of the watershed to over 4,100 m in the south-eastern part (Mount Guna Massif) (Figure 1). This diversity in topography enables the watershed to receive adequate rainfall to satisfy crop water demand and produce different crops and livestock. The mean annual rainfall in the watershed is about 1,503 mm, with a mean annual temperature of 15.6 °C. The climate of the watershed can be divided into three seasons: summer or main rainy season, locally known as Kiremt season (June–September), dry or Bega season (October–February), and short rainy or Belg season (March–May). The rainfall in the study area is generally erratic and unevenly distributed with more than 80% of rainfall occurring during the main rainy season. The agriculture of the watershed is characterized by mixed crop-livestock systems. The major crops grown include maize, barley, wheat, beans, and rice.

The long-term monthly rainfall and temperature data for both current and future timescales were considered in this study. The daily precipitation and temperature records for twelve weather stations located within and around the Rib watershed were obtained from the National Meteorological Agency (NMA) of Ethiopia for the period 1986–2017. After testing the consistency, spatial representativeness, and continuity of data for each station, only six stations (Deberetabor, Kimirdingay, Yifag, Alember, Gassay, and Addiszemen), which have a relatively long period of data (at least 30 years) and have no more than 10% missing values, were selected and used for further analysis. Table 1 shows the location of the selected stations and their data values.

The future climate projections and analysis were carried out based on the output of five global climate models (GCMs) (i.e. HadGEM2-ES, BCC-CSM1-1, NorESM1-M, MIROC5, and GFDL-ESM2M) derived from the CMIP5 archive under two representative concentration pathway (RCP) scenarios, RCP4.5 and 8.5. RCP4.5 and 8.5 were purposely selected in order to determine the future climate of the area under moderate and extreme greenhouse gas concentration conditions, respectively. The GCMs and RCPs selected in this study have been also considered in previous studies in Ethiopia (Alemseged & Tom 2015; Abera et al. 2018; Bekele et al. 2018; Worku et al. 2018) and proved to be the best models and scenarios capable of predicting an accurate future climate. The downscaled products of the five models for the period from 2018 to 2050 were downloaded from the MarkSim data distribution center. Although MarkSim has been extensively tested (Jones & Thornton 2013), we assessed the performance of the models by comparing their simulated products (1986–2017) with observed data of the same period using statistical measures such as root mean square error (RMSE), correlation coefficient, bias, and coefficient of variation (CV).

Table 2 indicates the statistical metrics test results of the performance of the selected GCMs in capturing the mean annual rainfall of the Rib watershed. The results confirmed that all the selected models performed well in capturing accurate future rainfall compared to the observed rainfall, but to varying extents. The CV of observed mean annual rainfall was found to be 15.85%, indicating low dispersion of annual rainfall from the mean. The CV values computed for the selected models, particularly those of HadGEM2-ES and MIROC5, also showed a small dispersion of annual rainfall from their mean values. The rainfall values simulated by the GCMs also revealed a positive and high correlation with the observed rainfall (r > 0.26). Although no important biases were detected in all models, two models (GFDL-ESM2M and MIROC5) underestimated the annual rainfall in relation to the observed rainfall, whereas the remaining three models and the Ensemble mean overestimated the annual rainfall. Based on the RMSE test, among the evaluated models, HadGEM2-ES performed best (RMSE = 116 mm/year) followed by the BCC-CSM1-1 model (RMSE = 254 mm/year).

The seasonal and annual variability of rainfall in the watershed was analyzed using the CV and PCI. Table 3 summarizes the results of seasonal and annual rainfall variability across the Rib watershed over the past few decades. The mean rainfall for Kiremt season varies from 881 to 1,214 mm while the mean Belg rainfall varies from 84 to 184.6 mm. This shows that much of the rainfall in the watershed comes during the Kiremt season (June–September), which contributes more than 80% of the rainfall. Most of the analyzed stations show moderate to high seasonal rainfall variations. As shown in Table 3, the Kiremt season shows relatively lower variability and low concentration of rainfall compared to that of Belg and annual periods.

The trends in annual and seasonal rainfall were assessed by applying Sen's slope and MK tests at 5 and 10% levels of significance (Table 4). The annual and kiremt rainfall trend analysis results show an increasing trend at four stations out of the total six, whereas the Belg rainfall showed a declining trend at four stations. Most of the trends detected for both annual and seasonal rainfall were statistically insignificant at the 5 and 10% levels of significance. The maximum statistically significant increase for annual rainfall was observed at Debretabor station (4.98 mm/year, at a 10% level of significance), whereas a statistically significant maximum increase for Kiremt rainfall was 3.95 mm/year observed at Addiszemen station.

The spatial distributions of the annual and seasonal rainfall and their PCI values are shown in Figure 2, indicating high spatial variability in rainfall amounts and concentrations. For both annual and seasonal scales, the southern parts of the watershed receive a significant amount of rainfall while the northern and north-western parts of the watershed receive a relatively lower amount of rainfall. On an annual scale, precipitation concentration in the Rib watershed can be described as strongly irregular on the northern and north-western and uniform on the southern and south-eastern parts. For both annual and seasonal scales, we noticed that the highest values of PCI belong to less rainfall receiving areas, implying more rainfall irregularity in small rainfall receiving areas compared to that of high rainfall receiving areas.

The changes in long-term minimum, maximum and mean monthly and annual temperatures were analyzed to detect the variabilities and trends across the watershed for the period 1986–2017. The results of descriptive statistics of monthly minimum, maximum, and mean temperature calculated for the average of six stations are summarized in Table 5. The results indicate that the monthly temperature in the watershed has exhibited significant variation during the last few decades. The temperature in the watershed was relatively low during the monsoon season (June–September) and high during pre- and post-monsoon months.

Table 6 summarizes the trends of the annual minimum, maximum and mean temperature in the study area. The increasing trends of temperature were detected for all stations although most of the trends are statistically insignificant at 5 and 10% levels of significance.

Projected changes in rainfall and temperature until the middle (2050) of the 21st century under RCP4.5 and 8.5 were analyzed based on the outputs of the five most commonly used GCMs. The projected mean annual and seasonal rainfall for the year 2050 is presented in Figure 3. The mean annual rainfall projected under RCP4.5 varies from 1,386 mm (BCC-CSM1-1) to 1,828 mm (GFDL-ESM2M), whereas mean annual rainfall under RCP8.5 varies from 1,466 to 2,028 mm. For model ensemble, the mean annual rainfall over the whole watershed varies from 1,546 mm (RCP4.5) to 1,718 mm (RCP8.5). With ensemble models, the projected mean Kiremt rainfall varies from 1,265 mm (RCP4.5) to 1,317 mm (RCP8.5), whereas the Belg rainfall varies from 134 mm (RCP4.5) to 129 mm (RCP 8.5).

The percentage changes in projected annual and seasonal rainfall were computed relative to the baseline rainfall, the rainfall during the observed period (1986–2017). The computed results show that the rainfall in the Rib watershed is expected to increase for both RCPs (Figure 4). Under RCP4.5, the projected change in mean annual rainfall varies from +5.06 (BCC-CSM1-1) to +38.57% (GFDL-ESM2M), whereas the change in mean annual rainfall under RCP8.5 varies from 11 (BCC-CSM1-1) to 54% (GFDL-ESM2M). For model ensemble, the mean annual rainfall in the future is expected to increase by 17% (RCP4.5) and 30% (RCP8.5). The model ensemble projected about a 20 and 25% increase in Kiremt rainfall under RCP4.5 and 8.5, respectively, while the Belg rainfall is projected to decline by 4.8 and 8% under RCP4.5 and 8.5, respectively. Except for GFDL-ESM2M, all models predicted decreasing trends in Belg rainfall in 2050, compared to the baseline period.

The projection results from the selected model outputs under both scenarios reveal a significant increase in temperature over the Rib watershed in the middle of the 21st century, compared to the baseline period (1986–2017) (Figure 5). The projected mean annual minimum temperature ranges from 10.30 °C for GFDL-ESM2M (RCP4.5) to 12.69 °C for NorESM1-M (RCP8.5). On the other hand, the projected mean annual maximum temperature varies from 25.27 °C for GFDL-ESM2M (RCP4.5) to 28.63 °C for NorESM1-M (RCP8.5).

For model ensemble, the minimum temperature is projected to reach 11.12 °C for RCP4.5 and 12.16 °C for RCP8.5 by the middle of the 21st century. The maximum temperature for model ensemble is expected to reach 25.94 and 27.91 °C under mid-range (RCP4.5) and high-range (RCP8.5) scenarios, respectively. Under RCP4.5, the mean temperature for the whole watershed in 2050 is anticipated to reach 18.53 °C and shows a substantial rise to 20.03 °C under RCP8.5. In general, all selected models showed that the temperature projections under RCP8.5 are higher than those of RCP4.5. The rates of change in future temperature were analyzed by comparing the projection results to the baseline period (1986–2017) (Table 7).

The highest mean annual minimum temperature projected under RCP4.5 is 2.08 °C (MIROC5) while the GFDL-ESM2M model under RCP8.5 estimated mean annual minimum temperature is 3.20 °C. For all models, the mean annual maximum temperature shows a significant increase under RCP8.5 compared to that of RCP4.5. The highest mean annual maximum temperature under RCP8.5 is 3.68 °C (NorESM1-M), while the GFDL-ESM2M model estimated the lowest mean annual maximum temperature of 2.17 °C. In general, although the rate of changes in temperature varies across the models and scenarios, projection results show that there is a high certainty of unforeseen increase in minimum and maximum temperature in the watershed by the middle of the 21st century. The increases in temperature will be notably higher under RCP8.5 than that of RCP4.5.

Current records indicate that both seasonal and annual rainfall patterns across the watershed vary extremely and exhibit high temporal and spatial variability. The watershed receives the maximum amount of rainfall in the Kiremt season (June–September), particularly in the months of July and August, which alone accounts for more than 50% of the Kiremt rainfall. Most parts of the watershed experienced high variability or less reliability of rainfall (CV > 30%) over the last few decades. In all the stations, the CV values of Belg rainfall exceed that of Kiremt rainfall, implying higher variability of Belg rainfall in the watershed than Kiremt. Earlier studies (Seleshi & Zanke 2004; Viste et al. 2013; Alemayehu & Bewket 2017; Asfaw et al. 2018) also indicate more Belg rainfall variability than Kiremt rainfall in most parts of Ethiopia.

The spatiotemporal distribution of rainfall was also assessed using PCI. On an annual scale, rainfall distribution across the watershed can be described as strongly irregular, as shown by the PCI values exceeding 20. On a seasonal scale, the Kiremt was the season where precipitation has been regularly distributed within months (lower PCI values, ranging from 7.5 to 12), while the Belg season precipitation distribution was highly irregular (PCI values between 18.5 and 24.5). Birara et al. (2018) also found high irregularities in Belg rainfall distribution compared to Kiremt rainfall in the Tana basin region of Ethiopia. The strong irregularity of annual rainfall distribution observed in the watershed (particularly in the northern and north-western parts) can be therefore highly attributed to precipitation concentration during the Belg season.

The annual and Kiremt rainfall shows a general increasing trend at four stations while the Belg rainfall shows a decreasing trend at four out of the six stations. However, most of the trends were statistically insignificant at 5 and 10% levels of significance. Previous studies (Alemayehu & Bewket 2017; Asfaw et al. 2018) also indicate a statistically insignificant increasing trend of annual and Kiremt rainfall and a decreasing trend of Belg rainfall in central and north-central highlands of Ethiopia. However, our results contradict earlier studies by Seleshi & Zanke (2004) and Wagesho et al. (2013) which reveal decreasing trends of Kiremt and annual rainfall in most parts of Ethiopia. This contradiction may be due to the differences in the time period, size of the geographical area studied, number of stations, and data quality and availability (Degefu & Bewket 2014).

Projections based on model ensemble suggested that the Kiremt rainfall will probably increase by 20% by 2050 under RCP4.5 and by 25% under RCP8.5 relative to the baseline period (1986–2017). However, the Belg rainfall is projected to decline by 4.8 and 8% under RCP4.5 and 8.5, respectively. Our results on future rainfall projections are generally in agreement with previous findings (Hadgu et al. 2015; Abera et al. 2018; Fentaw et al. 2018) which indicate a high probability of an increase in annual and Kiremt rainfall and a decrease in Belg rainfall in different parts of Ethiopia. However, our results contradict previous reports by Arndt et al. (2011) and Kassie et al. (2014) which show the high possibility of a decline in annual and Kiremt rainfall in Ethiopia in the future. According to Emori & Brown (2005) and the IPCC fifth assessment report (IPCC 2013), the dynamic changes in future rainfall would be highly attributed to the human-induced rise of global surface temperature.

Agriculture in highland Ethiopia is predominantly dependent on rainfall. Therefore, any variation in rainfall amount, distribution, and trends will have a direct impact on agricultural production and thus significantly affect the lives of rural smallholder farmers who depend largely on agriculture as their main source of income. A high concentration of rainfall over a few months (July and August) results in more frequent flood events and soil erosion, thereby posing a threat to agricultural production. Flooding and waterlogging in the farmlands significantly reduce crop yield through anaerobic stress in the roots (Fiwa et al. 2014), particularly in lower slope areas. Flooding not only inundates the agricultural fields and destroys the crops, but also could highly influence the quality of grazing lands and irrigation facilities at the downstream watershed (Maharjan & Joshi 2013). Excessive rainfall also leads to a high rate of surface runoff and soil erosion that causes the loss of fertile topsoil in the high slope areas and sedimentation in low slope areas in the absence of proper soil conservation structures (O'Neal et al. 2005; Fiwa et al. 2014; Kassie et al. 2014). Loss of soil fertility consequently leads to losses in agricultural production and rural livelihood. Because of the high concentration of rainfall in Kiremt season, while other months remain dry or receive little rain, the farmers in the watershed are unable to grow food crops more than once on their small parcel of plots in a given year, which leads to food insecurity and poverty.

A projected increase in annual and seasonal rainfall would have mixed implications for future agriculture in the watershed. On the one hand, an increase in growing season (Kiremt) rainfall could considerably contribute to future agriculture under proper soil conservation practices to control associated hazards such as flooding, excessive runoff, and soil loss. On the other hand, unless properly managed, the projected increase in Kiremt rainfall will lead to excessive flooding, runoff, and soil loss, thus contributing to a reduction in agricultural yield. High variability and irregular distribution of Belg rainfall observed in the watershed could affect not only the agriculture of the Belg season but also the agricultural performance of the main rainy season by influencing the early plowing operation and soil moisture availability issues. The declining trend of Belg rainfall may also influence the farming practice of the watershed by limiting crop choice and enhancing the loss of biodiversity (Hadgu et al. 2015).

Our results show a greater warming trend across the Rib watershed for both current and future scenarios. The mean annual temperature in the study area has increased by 1.07 °C over the last four and a half decades, with an average rate of 0.24 °C per decade. Earlier studies also indicate the occurrence of rapid warming across Ethiopia at varying rates, but consistent with global and African trends. According to Kotir (2011), over the countries lying in the Nile Basin, the mean annual temperature increased in the range of 0.2–0.3 °C per decade during the second half of the 20th century. The studies by Birara et al. (2018) and Kassie et al. (2014) also show an increasing tendency of temperature in different parts of Ethiopia during the last few decades, with an average rate of between 0.12 and 0.54 °C per decade. Model-based predictions of future climate also clearly suggest the warming of the watershed during the first half of the 21st century. Prediction under model ensemble indicated the rise of average temperature by 1.64 °C (RCP4.5) and 2.49 °C (RCP8.5) in the watershed by the middle of the 21st century, relative to the period (1986–2017). Similar results were found in other parts of Ethiopia, where the future temperature is projected to increase, regardless of the models and emission scenarios used (Conway & Schipper 2011; Kassie et al. 2014; Abera et al. 2018; Fentaw et al. 2018).

The higher increase in growing season temperatures may adversely affect crop production, farm income, and food security in many ways, basically when combined with high inter-annual and intra-seasonal variability of rainfall. The projected warming would reduce the grain yield of cereal crops of the watershed which is already experiencing significant reduction due to human-induced soil erosion (Moges & Bhat 2017). A study by Maharjan & Joshi (2013) indicates that an increase in temperature significantly affects mean yield responses as well as yield variability of maize, millet, and sorghum in many African countries. Heat stress reduces grain yield by raising evaporation and reducing water availability (Kang et al. 2009; Hatfield & Prueger 2015), particularly in low rainfall receiving downstream areas. It can also cause flower abortion and reproductive failure of crops (Hadgu et al. 2015). Moreover, the high temperature could influence crop growth, development, and yield through increased pressure from existing and new pests, weeds, and diseases (Pathak et al. 2018). An increase in Belg season temperatures can also influence irrigation farming by reducing surface and sub-surface water availability.

The rise in temperature not only influences crop production but also profoundly affects the growth and production patterns of livestock. Heat stress highly affects the productive and reproductive performance of livestock by compromising the reproductive system. An increase in temperature also results in the spread of new and emerging diseases which have a serious effect on livestock productivity and survival (Musemwa et al. 2012). The rise in temperature also adversely affects livestock production through competition for natural resources, quality, and quantity of feeds and biodiversity loss while demands for livestock products increase (Garnett 2009).

This study has presented a detailed analysis of the current and future rainfall and temperature variability and trends in the Rib watershed, north-western highland Ethiopia. A focus has been also given to describe the possible implications of observed and projected climate variability and changes in the farming system of the watershed. The results show that the watershed is highly exposed to spatiotemporal variations in the magnitude and direction of rainfall and temperature. Similar trends of change were found for the current and future rainfall – an increase in annual and Kiremt rainfall and a decrease in Belg rainfall. Similarly, the analyses of recorded and projected temperature data indicate increasing trends in mean minimum and mean maximum temperature in the study area. The variation in the amount, distribution, and trends of rainfall and warming temperature could have a direct implication for the productivity of rainfed agriculture and the livelihoods of rural farmers who depend largely on agriculture. The observed rainfall and temperature variability and changes would cause excessive runoff and soil loss, exaggerate crop and livestock diseases, and increase evaporation and reduce water availability, thereby significantly influencing agricultural productivity, food security, and rural livelihoods. The findings of the present study would offer useful information to better understand the spatial variability and temporal trends of rainfall and temperature which are essential for water resource management and farming practices.

All relevant data are included in the paper or its Supplementary Information.