Comparative analysis of long-term precipitation trends and its implication in the Modjo catchment, central Ethiopia

The hydrologic system and water resources of a catchment are widely determined by climate change and human activities. In recent decades, climate change demands even the revision of engineering designs that affect our lives seriously (Mohammed et al. 2018). Precipitation is the main weather variable in which its change can be related to global warming (Li & Luo 2011), and its spatiotemporal distributions directly affect the distributions of water resources in space and time (Aziz & Burn 2005). Hence, studying the hydro-climate trends is highly important for identifying the spatiotemporal variability and management of inadequate water resources for future economic development, flood control measures as well as agricultural planning activities. Hence, a detailed analysis of rainfall trends and variability is highly important to national and local water resources management as it contributes to a better understanding of the impacts of global climate change.

In Ethiopia, precipitation is the primary source of both surface and groundwater supply, and its significant variability directly affects the quantity and quality of the water resources. Therefore, analysis and understanding of the characteristics, amount, and distribution of this variable is essential for sustainable management and to improve the utilization of water resources of the country. Various studies have been conducted regarding trends and variability in Ethiopia, with a range of spatial (e.g., national, regional) and temporal (e.g., annual, seasonal, monthly) scales (Bekele et al. 2017; Gummadi et al. 2018; Worku et al. 2022). However, the studies conducted in different parts of the country did not report consistent trend results even for the same areas. For example, Cheung et al. (2008) investigated the trends and spatial distribution of annual and seasonal rainfall over Ethiopia and reported a declining trend in the annual and summer rainfall in eastern, southern, and southwestern Ethiopia. However, Rosell & Holmer (2007) and Viste et al. (2013) indicated an increasing trend result regarding the summer rainfall pattern in different parts of the country. Jury & Funk (2012) on their part reported a weak rising trend in the arid lowlands of the southeastern and a downward trend over the western highlands of the country. However, Seleshi & Camberlin (2006) found the absence of significant trends in the Kiremt (summer) and Belg (spring) seasons in many parts of Ethiopia except the eastern, southwestern, and southern parts. Recently, Worku et al. (2022) applied the M–K test and Sen's slope estimator to determine rainfall trends in southern Ethiopia using data for the period 1981–2018. They reported that a significant increasing trend in annual rainfall was detected in some of the stations such as Arero, Dehas, Dillo, and Miyo. Therefore, consistent rainfall trends and variability results are crucial not only for knowing the monotonic trends of an area but also for the proper estimation and analysis of water resources as well as for the planning of effective adaptation strategies.

Most of the previous trend-related studies have been conducted at the regional or basin scales (Gedefaw et al. 2018; Misrak et al. 2019; Harka et al. 2021) among others. However, rainfall variability and trend at the local scale or catchment level have not been given due attention and the issue was not well addressed in Ethiopia. Analyzing the spatiotemporal trends of hydro-climate variables such as rainfall is vital to look into the effects of climate change on agricultural activities as well as water resources planning and management at the catchment scale. The large-scale studies cannot be overlooked as well as not suitable to observe local variability that can affect agricultural activities, water resources, and hydrologic regimes of catchments. Moreover, the previous studies employed the conventional M–K trend test and Sen's slope estimator (SS) while quantifying the spatiotemporal trends in the country. However, these methods have limitations due to their assumptions of data distribution as well as sample size and trend calculation algorithms (Alifujiang et al. 2020). In addition, the classical M–K test requires the removal of the serial autocorrelation from time-series data that can alter the significant trend results (Wagesho et al. 2013). To overcome the stated drawback of this method, Şen (2012) developed the innovative trend analysis (ITA) method as a new statistical trend estimation approach, which is a flexible graphical method used to explore time-series trends to avoid errors when major hidden trends are detected. According to Şen (2012), the ITA method is independent of sample size, data distribution, and serial correlation. Moreover, the classical tests do not allow to identify the contribution of low and high values in the detected trend. The cited works indicated that the ITA method has some advantages compared with the conventional methods. For example, Wu & Qian (2017) applied linear regression, the M–K trend test and the ITA method to detect trends of seasonal and yearly rainfall extremes, and they concluded that the ITA method is the better one. Moreover, the ITA approach allows more detailed interpretations of trend detection, which has benefits for identifying hidden variation trends of precipitation and the graphical illustration of the trend variability of extreme events, such as ‘high’ and ‘low’ values of precipitation (Ay & Kisi 2015; Caloiero et al. 2018; Caloiero 2019; Alifujiang et al. 2020).

The objective of this study is, therefore, to analyze the spatiotemporal trends of rainfall amount of a highly agricultural catchment located in central Ethiopia by a comparative analysis of the classical M–K trend test, SS, and ITA method. The proposed study is crucial since previous studies in the region are scanty and fragmented. The nobility of the proposed study lies in the comparison between the classical M–K test and the ITA method for trend robustness. Moreover, unlike the past study in the same area (Eshetu 2020) to analyze similar issues using the classical M–K test and within the same region (e.g., Gedefaw et al. 2018; Misrak et al. 2019), this paper offers a better understanding on the distribution of annual and seasonal precipitation patterns over space and time. Finally, the study results will be helpful in identifying constraints related to climate change faced by the local farmers and provides references for further analysis of climate-related issues in an agricultural catchment.

The rainfall regime of the catchment is bi-modal type, with two distinct peaks in April and August months. On the basis of recorded data of the period 1981–2020, the annual rainfall of the area varies between 846 and 1,131 mm, with maximum value in the northern and northwestern parts (https://www.ethiomet.gov.et). The rainfall that occurs during the major rainy (Kiremt) season (June to mid-September) and Belg season (March, April, May (MAM)) covers, more than 80% of total annual precipitated rainfall amounts over the catchment. Based on recent studies (Eshetu 2020), the annual mean minimum and mean maximum temperatures vary between 10.6–11.6 °C and 21–27 °C, respectively, whereas the annual average temperature of the watershed varies between 16.9 and 19.3 °C.

The natural vegetation in the catchment mostly consists of shrubs and range lands; however, Eucalyptus trees are found close to the towns and around rural villages. Agriculture is the dominant sector and source of income in the catchment. The sector is dominated by small-scale farmers those practice rain-fed mixed farming with the help of traditional technology, adopting low-input and low-output production systems. Attributable to high population growth and its pressure, most arable land area is cultivated in the catchment. In recent time, more than 90% of the catchment area is cultivated. The agro-climatic conditions make it suitable for permanent and temporary crops.

Daily precipitation datasets were collected from the National Meteorological Service Agency of Ethiopia (https://www.ethiomet.gov.et) which is used in the study to understand the spatiotemporal trends of an agricultural catchment located in the central part of the country. The data consisted of daily precipitation observations covering the period 1981–2020 from seven meteorological stations within (Bishoftu, Chefe donsa, Edjere, and Modjo) and outside (Aleltu, Koka dam, and Zequala) the catchment boundary (Figure 1). List of stations, latitude, longitude, and elevation of the meteorological stations used in the present study are presented in Table 1.

Based on the agricultural activities of the area, we divided the daily data into four seasons: summer (June–August), is the major rainy time of the catchment, autumn (March–May), is the second contributor to annual rainfall, winter (December–February), and spring (September–November). The obtained daily precipitation data were converted to average monthly and then seasonal data for the analysis of temporal trends at the station level and spatial distribution of rainfall at the catchment scale.

The precipitation data were subjected to different quality control procedures such as checking missing records, homogeneity, and inconsistency tests. There are few records with continuity problems in the time-series. This may be due to instrumental failure, changing the record site, or the absence of the observer. Therefore, it is necessary to estimate and complete the missing data before using it for trend analysis. However, the choice of the methods is based on the percentage of data missed and choice of neighboring stations. When the amount of the data filled is less than 5%, linear regression can be used by identifying the relationship between the observed data of neighboring stations and that of the reference station (Aieb et al. 2019). In this study, the missing values were less than 3% (Table 1). Hence, linear regression was used to fill in the missing rainfall data. The other pre-processing analysis carried out was the identification of outliers, which were treated case by case using information from the date before and after the event. In the present work, the data of each station were plotted against time in days of the year format and subjected to visual examination for the presence of outliers and typing errors such as negative rainfall values.

The homogeneity tests are an important part of climate change analysis because the tests help to identify whether the variations in climate data are either due to meteorological, climatological, or external factors. Inconsistency and homogeneity of the rainfall data were checked using the Double mass curve (DMC) analysis and standard normal homogeneity test (SNHT) (Agha et al. 2017). By using the double mass analysis (DMA) method, cumulative annual rainfall for each station (suspect gauge) was plotted against cumulative mean annual rainfall for surrounding stations. Any deviation from the straight line would therefore mean that the data are not homogenous and inconsistent. In such a situation, a correction factor is calculated and used to correct erroneous data. The coefficient of determination (R²) was used to show the strength of the correlation and relationship between each station and the surrounding stations and hence confirm the consistency of rainfall data. The quality control results obtained from both the SNHT and DMC tests indicate that the rainfall data are homogenous, consistent, and of good quality, and can further be used for hydrological analysis (Table 2). Moreover, the relation between the suspect gauge and the other stations was checked, and the value of the coefficient of determination (R²) for the stations is large indicating the strong linear relationship between each station and other surrounding stations. Therefore, the identification of trend tests and their magnitudes was carried out on the homogenous rainfall data collected from the seven stations.

The coefficient of variation (CV) was used to evaluate the rainfall variability of the study catchment. CV measures the relative dispersion of time-series data from the mean. It can be estimated from the ratio of standard deviation (SD) to mean rainfall values. Overall, in using the CV, the degree of rainfall variability can be analyzed based on the following: if CV < 20 (low variability), if 20 < CV < 30 (moderately variable), high when 30% < CV < 40%, very high when 40% < CV < 70%, and extremely high when CV > 70% (Asfaw et al. 2018). Therefore, the CV was applied to understand the variability of annual and seasonal rainfalls in our study area.

It is important to note out that if the r₁ value falls within the confidence limits, there is no significant autocorrelation in the dataset and the M–K test can be applied on the dataset to examine the trend. On the contrary, if r₁ falls outside of the confidence interval, it can be said that the time-series exhibits a significant autocorrelation and the M–K test should be applied after the removal of the autocorrelation (Alashan 2020). Based on the result (Table 2), except rainfall record at the Modjo station all the stations have no correlations at the 5% level. The annual rainfall of the Modjo station is dependent at a 5% significant level and the significant autocorrelation was removed as per the suggested correction procedure for this particular station before using the data for trend analysis by the M–K test. The summary of the autocorrelation test results at the station level is presented in Table 2.

In the study, the ITA method, developed by Şen (2012), with later mathematical and graphical improvements (Şen 2017; Alashan 2018) was applied to detect the trends in rainfall time-series. The ITA method is chosen taking the advantage of its freeness from different types of assumptions and its applicability both for graphical and statistical trend analysis. The technique was used in different parts and proved as an effective method of trend analysis (Gedefaw et al. 2018; Malik et al. 2019). In the process, initially, the time-series is separated into two equal parts and arranged in ascending order. The first part of the time-series (X_i: i = 1, 2, 3, …, n/2) is placed on horizontal x-axis and the other time-series (X_j: j = n/2 + 1, n/2 + 2, …, n) is located on the vertical y-axis on the Cartesian coordinate system. Graphically, the ITA method provides visions in trend analysis, namely monotonic/non-monotonic increasing, monotonic/non-monotonic decreasing, and trendless conditions (Oztopal & Şen 2017). A monotonically increasing/decreasing trend can be detected if all points are distributed above/below the 45° (1:1 line), respectively; a non-monotonically increasing/or decreasing trend is when most of the points are distributed above/or below the 45° line; a monotonically decreasing trend can be detected if all points found below the 45° line; a non-monotonically decreasing trend can be detected if most of the points are below the 45° line, and a trendless condition is when the points are concentrated on the 45° line (Şen 2012).

Here, S_crt represents the critical slope for standardized time-series at ±1.96 for a 95% significant level (⁠⁠). If the ITA slope (S_ITA) value falls outside of the lower and upper CL values, then the null hypothesis of no significant trend is to be rejected at α significance level (Şen 2015). In a two-sided condition, the null hypothesis (H₀) indicates that no trend in time-series data, and the alternate hypothesis (H₁) describes that there is a trend in time-series data at α = 5% with Z = ±1.96. If ±S_ITA > ±CL_(1–α), then H₀ is rejected and H₁ is accepted. The positive and negative value of S_ITA indicates a rising and falling trend in the time-series data, respectively. The positive and negative value of S_ITA indicates an increasing and decreasing of trend in time-series data, respectively (Şen 2017).

For the presence of a significant trend, the calculated Z value is compared with standard normal distribution with two-tailed significance levels. In this study, the null hypothesis should be accepted at α = 5% level of significance. Moreover, an increasing (upward) trend should be interpreted if Z > 0 or the time-series experiences a decreasing (downward) trend if Z < 0, where as if Z = 0, a trend due to random fluctuation, both increase and decrease trends have similar impacts, hence entirely there is no trend in the time-series.

The annual and seasonal precipitation variability was assessed using the CV. The CV is a statistical measure of dispersion around the mean, which is the ratio of SD to the mean. Areas with high annual precipitation showed less inter-annual variation, whereas areas with low annual rainfall experienced high inter-annual variation. The annual precipitation distribution of the catchment showed from low (CV < 20%) to moderate (i.e., 20% < CV < 30%) variability, except at the Koka dam station which shows a very high variability (i.e., 40% < CV < 70%) in annual precipitation (Table 3). With regard to the seasonal rainfall, low to high variability was observed in all seasons (Table 3). For example, the CV value for summer (18–58%), winter (4 to >119%), and spring (32 to >77%). During the winter season, CV is greater than 50% in most stations except at Bishoftu and Chefe donsa, but this is not a crop season and has no significant impact on the cropping system in the area. Generally, summer and spring seasons rainfall patterns are highly variable over the area, which could affect the agricultural productivity and water resource sustainability over the catchment. Moreover, the high rainfall variability during the spring season can be an indication of low agricultural production in such highly populated area of the country. Comparatively, the seasonal rainfall of the catchment showed more variation than that of the annual variation of the period. The finding of this study is in line with the studies recently carried out in the area (Bekele et al. 2017; Gedefaw et al. 2018; Mahtsente et al. 2019; Eshetu 2020; Bayable et al. 2021). The highest seasonal rainfall variability was observed during winter time, which varies from 4.5 to 119 over the catchment. Previous studies by Degefu & Bewket (2014), Asfaw et al. (2018), and Mekonen & Berlie (2020) indicated that the winter season rainfall is highly variable in Ethiopia.

Figure 3(b) presents the spatial distribution and its variability of the summer season precipitation over the Modjo catchment. The summer season precipitation followed almost the same spatial distribution as that of the annual, except for some changes in the central parts (Figure 3(b)). During this season, the highest rainfall values were recorded in the western (which has the highest topography in the catchment), southeastern, and small parts of the northern region while the lowest rainfall values of the period were recorded in the southeastern, eastern and northwestern parts of the catchment (Figure 3(b)). This may have an impact on the agricultural productivity of the area as the area is one of the main farming of cereal crops. Figure 3(c) depicts the spatial distribution of rainfall during the autumn season (MAM) rainfall distribution of the period. The autumn (MAM) season is the second contributor of annual rainfall amounts. This season follows partially similar trends as that of the summer season. As can be seen, the spatial rainfall distribution in the western parts of the catchment received relatively better rainfall amounts, while the northern and southern, and northeastern parts received the lowest rainfall amount (Figure 3(c)). During this season, the highest rainfall values were observed in the western region, which ranges from 213–221 mm, whereas the rest parts of the catchment experienced a very low rainfall amount throughout the year. Winter (December, January, February (DJF)) is the season with the smallest rainfall amounts in the region (Figure 3(d)). This season is usually used for planning and preparation for the next cultivation season.

Many studies have been conducted to analyze the spatial variability of precipitation in different parts of the country (e.g., Wagesho et al. 2013; Alemu & Bawoke 2019; Harka et al. 2021) are among others. Spatially, the annual and seasonal rainfall variabilities were assessed by Worku et al. (2022) in the south central Ethiopia, and a decrease from the northeast and northwest parts of their study area. Fitsum et al. (2017) stated the spatial distribution of rainfall in Ethiopia is radically influenced by complex topography and elevation as the major controls of climate in the country.

In this study, comparative approaches of the classical and ITA methods were employed to examine the long-term rainfall trends of the Modjo catchment. SS is also considered to estimate trend magnitude. The annual trend analysis result under different methods is presented in Table 4. As per the M–K test, non-significant trends were dominant in the annual precipitation except at Zequala stations (Table 4), which revealed decreasing trend (Z = −3.13, p < 0.05) (Table 4). Thus, the decreasing significant trend at these stations could be attributed to climatic variability. All the other increasing and decreasing trends were not significant in statistical terms. However, increasing trend tendencies were dominant in the catchment. There is an increasing trend (Z = 1.90 ≥ 1.65 and p = 0.065 < 0.10) in the Modjo station according to the M–K test, but there is no trend according to these tests at a 95% significant level. Generally speaking, it was observed that the annual rainfall in high-altitude stations shows decreasing trend and the opposite is true for the low-altitude stations considered in the study area (Table 4). Similarly, the magnitudes of positive and negative SS are recorded in five and two stations, respectively (Table 4.) The SS value was maximum in the Koka dam (+6.28 mm/year) and the minimum trend magnitude was obtained in Zequala (−18.54 mm/year) stations (Table 4).

The statistical and graphical approaches of the ITA method have also been applied in order to show the robustness of trends in the study catchment. In this work, the significance level (α = 5%) was considered to measure the significance or non-significance of the rainfall trends on multiple scales. If the computed value of S_ITA is less than the critical value (S_crt), then the detected trend is a non-significant trend at a 5% significance level. According to the ITA method, there has been an increasing or decreasing trend in most of the stations as the S_crt value is outside the lower/upper CL in most of the stations Except at Aleltu and Zequala stations, the annual rainfall trends at the remaining stations were found to be insignificant at 5% level of significance under the ITA method (Table 4). The annual rainfall trend at Aleltu and Zequala stations showed a statistically significant decreasing rate of 3.74 and 13.98 mm/year, respectively, at a 5% significance level compared with the computed critical slope (S_crt) 1.063 and 1.706 mm/year in that order (Table 4). From the ITA analysis result, the rainfall trend of Modjo and Zequala stations showed maximum increment and decrement rates, with ITA slope of 8.05 and −13.98 mm/year, respectively. In general, similar increasing and decreasing trend results have been detected by the ITA method except for the Koka station, which showed a significant decreasing trend under the ITA method (Table 4).

The results obtained in this study agree with the findings of other papers. For example, the study on spatiotemporal variability in rainfall conducted by Arragaw & Woldeamlak (2017) indicated that annual rainfall pattern increases insignificantly in the central parts of Ethiopia. The findings of the result are also in agreement with that of Alemu & Bawoke (2019) who reported that the annual rainfall showed an insignificant rising trend in Ethiopian highland areas. In the study by Harka et al. (2021), most of the rainfall stations in the southern Rift Valley of Ethiopia showed an increasing trend in annual time scale. Contrary to our finding, some studies in the region (e.g., Mahtsente et al. 2019; Gebremichael et al. 2022) reported in their work that insignificant decreasing trends in the annual rainfall for the stations in common (i.e., Koka dam and Modjo stations). However, Eshetu (2020) analyzed the temporal trends in the same catchment during the past 30 years (1981–2010), and concluded that the annual precipitation shows the statistically insignificant trend for all of the stations under the M–K test. But in this work, based on the longer precipitation time-series, even though insignificant increasing trends are dominant over the catchment, a statistically increasing trend was also identified for the station in common (i.e., Edjere station) at the 5% significant level (Table 4). This difference may be due to the time period considered in our case as well as the quality control processes we followed.

Based on the agricultural activities of the area, a year is divided into four seasons, namely autumn, summer, spring, and winter (Mohammed et al. 2018). The spring season extends from March to May, the short rainy season (locally known as Belg); the summer season rainfall ranges from June to mid-September, (locally called as Kiremt), the period in which the main agricultural activity is conducted. The winter season has 3 months in Ethiopia (December, January and February) whereas the autumn season includes September, October, and November months. The summer, spring and winter seasons are rainy seasons in Ethiopia in general and in our study area in particular. Therefore, we considered the rainfall variation and trends for these seasons, and the autumn season does not contribute a significant amount of rainfall, hence is not considered in this work.

At the seasonal scale, upward trends were dominantly occurring in four of the seven stations during the summer season (Table 5). Under the M–K test, only two of the stations (Modjo and Zequala) show a significant increasing/decreasing trend. However, under the ITA test, a significant downward trend was detected only at Bishoftu (S_ITA = −0.35 mm/year) and Zequala stations (S_ITA = −10.35 mm/year). The summer season rainfall at the Modjo station showed an insignificantly increment rate of 6.49 mm/year at a 5% significance level as the critical slope (S_crt) was 1.132 mm/year under the ITA method (Table 5). According to the ITA method, the summer season rainfall trend of Zequla and Modjo stations experienced maximum decrement and increment rates with ITA slope of −10.35 mm/year and +6.49 mm/year, respectively. Based on the M–K test, significant trends were detected for summer rainfall only at Zequala (Z = −3.25, p < 0.05) and Modjo (Z = 2.92, p < 0.05) stations.

In the spring season (Belg: March– May), downward trends were dominant, as it occurred in four out of the seven stations (Table 5). It can be seen that the rainfall amount of this season experienced a downward trend except at Koka and Modjo that revealed weak increasing trends with SS values of 1.27 and 1.0 mm/year in that order. During this season, a significant trend was detected only at Aleltu and Bishoftu stations under the M–K test, whereas, in using the ITA method a significant trend was observed in most of the stations except at Koka and Modjo stations which revealed an insignificantly decreasing and increasing trends, respectively (Table 5). Although the rainfall during the spring season is extremely needed for agricultural systems, declining trends of rainfall in this season were dominant in the Modjo catchment. A similar result was also reported in the central highlands of Ethiopia (Rosell 2011), in which our catchment is located, and in the whole country (Gummadi et al. 2018).

The trend analysis result shows that the precipitation in winter (Bega: February–December) has a declining trend in all the stations using all the methods (Table 5). Based on the M–K result, the rainfall recorded for the winter season at the Modjo station had the highest statistically significant reduction (Z = −3.34, p = 0.0001) at the 5% significance level. Moreover, the S_ITA value of rainfall during winter was the maximum for the Modjo station (S_ITA = −0.40 mm/year), which showed a significant decline trend (S_crt = −0.2452 mm/year), at a 5% level of significance (Table 5). In general, the ITA, M–K , and Sen's slope analysis results of the winter showed more consistency temporal trends as compared to the rainfall trend of the spring and summer seasons. In contrast to our findings, a study conducted in a similar area (Eshetu 2020), reported that there is a non-significant trend in the seasonal rainfall from 1981 to 2010. According to the study, Belg (spring) season rainfall at all the stations experienced an insignificant decline trend over the past 30 years, which is consistence with our findings from the statistical point of view. The same paper also indicated that insignificant trend for the summer (Kiremt) season rainfall in all the stations over the Modjo watershed. However, in our findings, some of the stations (e.g., Modjo) during summer as well as (Aleltu and Bishoftu/Debre zeit) for Belg seasons show significant trends in the period 1981–2020 (Table 5). The inconsistencies between the results may be due to the data length, and quality control approach considered in the current study.

According to the plot in Figure 7, except for winter season, no trend was detected at the Chefe donsa station during summer and autumn seasons, because the points are concentrated around the 1:1 line. The rainfall amount of the summer season is with the clearest trend evidence, as all the low, medium, and high values were positioned in the upper triangle at the Chefe donsa station (Figure 7(a)). For the low rainfall values of the winter season negative trend tendencies can be clearly observed for this station (Figure 7(b)). However, a no trend condition was evidenced in autumn at the Chefe donsa station, with few exceptions of the high rainfall values (Figure 7(c)). The plot in Figure 8, except for summer season, no clear trend was identified at the Edjere station during winter and spring seasons. The rainfall amount of the summer season is with the clearest trend evidence, as all the low, medium and high values were positioned in the upper triangle (Figure 8(a)). Except for some high rainfall values during the spring season (Figure 8(c)), no trend tendencies were observed in winter and spring seasons at the Edjere station (Figure 8(b) and 8(c)). There was a negative trend of the high rainfall values of summer, spring and winter rainfall patterns at the Koka station, with the highest values positioned largely far from the 1:1 line (Figure 9). Positive trend has been detected in summer, which is more evident for the low rainfall values. At this station, the lower rainfall values during summer show an inclined trend but the higher values during this season distributed far from the 1:1, and located in the lower triangle (Figure 9(a)). Overall, no trend tendencies were evidenced for the low and medium rainfall values at this station during winter and spring seasons (Figure 9(b) and 9(c), respectively).

Modjo is the station with the clear increased trend tendencies for the high rainfall values in the summer season (Figure 10(a)), with no significant trend evidences for the low and medium rainfall values during winter and spring seasons of the period. At this station, except few high rainfall values, all the low, and medium values were concentrated around the 1:1 line during both winter and spring seasons. Hence, no trend was detected at the Modjo station during the winter and autumn seasons (Figure 10(b) and 10(c), respectively). All the seasons evidenced clear negative trends, with some exceptions concerning the low rainfall values in summer, winter, and spring seasons at the Zequala station (Figure 11). The findings from the graphical analysis are in line with the statistical estimations under the ITA method (Table 5). In summary, the rainfall trend analysis indicates an inverse relationship between rainfall amount and temporal trend, i.e., the area which received the largest rainfall amounts showed a decreasing trend. For example, the northwestern part of the watershed received the maximum annual rainfall of the study period, but revealed a declining trend.

The M–K , SS, and ITA methods have been used and compared in this paper. The main intent is to determine accurate trend tendency in time-series data. A comparison of results of the rainfall trends estimated by different tests is summarized in Table 6. Consistent results were found, except in some cases in time-series trend analysis among methods used in the study. Accordingly, significant trends were detected only in one (12.3% of total stations) in annual and in two (28.6%) summer season precipitation using the M–K test, whereas these were two in applying the ITA method (Table 6). At the same time, two (28.6%) and five (71.4%) of the stations for the spring season rainfall amount experienced statistically significant trend using the M–K test and ITA method, respectively, indicates the ITA method displays a more definite significant pattern than the M–K test. On the other hand, the number of significant trend identified by the M–K and ITA tests are six (85.7%) and seven (100%), respectively, for the winter season (Table 6). This indicates, there has been some differences regarding the number of significant trends among the methods used in the study. Therefore, all the statistically significant trends identified by the ITA method were not supported by the classical M–K tests.

In this study, the spatiotemporal trend analysis results indicate the M–K and ITA method have some contradicting results. For example, a statistically increasing trend at Koka during spring rainfall is detected as a statistically decreasing trend under the ITA method (Table 6). Additionally, the statistically insignificant rainfall trends of spring and winter seasons at the Edjere station as well as the insignificantly increasing trend of the Chefe donsa station by the M–K test detected all as statistically significant trends when applying the ITA method (Table 6). Similar contradicting results between the classical M–K test and ITA method were also reported in previous studies. This may be due to data distribution, sample size, and extreme values while using the classical tests (Mekonen & Berlie 2020; Mallick et al. 2021). It is important to point out that the ITA method is insensitive to the sample's size, data distribution, and serial autocorrelation effects as evidenced from previous studies (Şen 2012; Machiwal et al. 2019). Therefore, as it can be demonstrated graphically on the Cartesian Coordinate System to provide a clear visual display for further clarification, the ITA method is found to be more robust for trend analysis in time-series data than classical methods (Alashan 2020). Furthermore, taking the advantage of its freeness from the assumptions of data normality, serial autocorrelation, its independency from record length as well as a seasonal cycle (Cui et al. 2017; Kisi et al. 2018; Zhou et al. 2018).

Adequate precipitation amount is necessary to ensure high and quality yield in the agricultural sector, particularly for the rain-fed agriculture-dependent countries. Among the climate variables, precipitation is the key determinant in the agrarian counties like Ethiopia. Its spatial and temporal distribution may affect the cropping system, schedule as well as productivity. Changes in this variable are known to be among the foremost long-term threats to humans, with an impact on all sectors of life generally and agriculture in particular (Larbi et al. 2022). It is worth noting that the major traditional occupation in the study area remains agriculture which is nearly rain-fed. According to CSA (2014), the influence of agriculture is incomparable that employs about 80% of the labor force and accounts for 45% of the GDP and 85% of the export revenue in any single year over the country. Therefore, any variation in the occurrence and distribution in rainfall amount can significantly affect the agricultural productivity and food security of the region.

The Modjo catchment receives high precipitation amounts under the present condition. More than 60% of total annual rainfall is precipitated during the summer (JJA) season. In most of the stations, rainfall of the major rainy season (summer season) revealed an increasing trend in the considered period. Therefore, sufficiently available rainfall during this season leads to greater moisture availability and facilitates early planting of long-duration crops (Gummadi et al. 2018). Spring (MAM) season, contributes around 19% of the annual mean rainfall. However, the catchment exhibits an overall declining trend tendency during this season (Table 5). It is clearly observed that the spring season is the cultivation and growing time for short duration crops. Hence, the rainfall precipitated during this season is extremely important for agricultural activities and productivity in the region. Therefore, decreasing trends in the rainfall amount of March to May months (spring season) will have serious impacts on the production and productivity of agricultural systems in such a highly populated region. It also affects soil moisture availability and disturbs the summer season cropping schedule, in turn, it makes future agricultural activities more challenging in the area.

Both surface and ground water resources are vital natural sources in the study catchment. Even though most of the rainfall trends did not show significant changes in statistical terms, the general tendency is toward increasing both in the annual and summer season (Tables 4 and 5). For example, 71.4% of the stations showed increasing trends during the summer (JJA) season rainfall amounts (Table 5). Given the strategic importance of the Awash river (in which our study area is located), any change in the spatial and temporal rainfall can impact multiple users including the Capital Addis Ababa. Considering water resource abstraction as constant and insignificant, the increasing trend in the amount of rainfall during this season is expected to increase either surface runoff or ground water recharges.

The increasing trends in the major rainy season may have supported by increased frequency of occurrence and magnitude of extreme events, which will have expected impacts to increase flooding and soil erosion risks. This would also increase the chance of flooding and erosion hazards within the catchment. The two major hydro-meteorological hazards in the Awash river basin are erosion and flooding that affects life and properties every year (Dereje et al. 2015). Flooding, on its parts can cause siltation and sedimentation problems, while erosion destroys the fertility of farmlands, and in turn reduced land productivity. In addition, increase of summer rainfall amounts plays a significant role on sediment flows into the Awash river, which can impose sedimentation problems to the downstream Koka hydroelectric reservoir. This has also implied the increasing tendency of siltation risks on the hydroelectric reservoir that can reduce the power generation capability of the dam. Various studies conducted in different parts of the country (e.g., Ayele et al. 2016; Matewos & Tefera 2020) averred that changes and fluctuations in hydro-climatic variables significantly affected water resources and food security in the country.

This study was conducted to explore the spatial variability and temporal trends of annual and seasonal precipitation of the Modjo catchment, central Ethiopia. The temporal trends were analyzed using the comparative approach of the classical M–K test and ITA method. In the study, daily rainfall records from seven stations for the period 1981–2020 were used. The results show that the average annual rainfall of the catchment is between 846 and 1,131 mm, with moderate inter-annual variability across the studied stations. Generally, the seasonal rainfall variability was higher than the annual rainfall variability, as evidenced by high CV, which is closely related to its geographical location and complex terrain.

There were no systematic patterns of temporal trends and spatial variations of annual and seasonal rainfall amounts over the study catchment. Increasing trends for the annual rainfall patterns were observed in most of the stations, however, significant trends were detected only at Aleltu and Zequala stations. The summer (JJA) season rainfall showed raising trend in most of the stations, however, statistically significant trend was identified at Bishoftu, Modjo and Zequala stations. The spring (MAM) season rainfall showed decreasing trend in the studied stations except at Chefe donsa, Koka and Modjo stations. The declined trend of spring rainfall indicating tendencies for increasing of the number of dry days and the reduction of rainfall magnitude of this season in most of the stations. On the other hand, the winter season rainfall revealed statistically significant declined trends in all the stations under all the methods, except at Edjere that showed insignificant decreasing trend under the M–K test.

Spatially, mixed rainfall variability was experienced over the Modjo catchment. It was observed that the annual and seasonal precipitations were unevenly distributed over the area. The annual rainfall distribution of the period decreased from west to east. The northwestern part of the catchment received the highest annual precipitation of the period that ranges from 1,065 to 1,085 mm. Whereas the lowest annual precipitation was observed in the southwestern and southeastern parts, which varies between 950 and 970 mm.

Comparing the results from the two methods, the significant trend in annual rainfall occupies 14.3% in the M–K test and 28.6% in the ITA method, whereas this is, respectively, 28.6 and 71.4% during the spring season, which indicates that more definite significant patterns are displayed by the ITA method than the M–K test. Overall, the ITA method allows more detailed interpretations about trend detection than that of the M–K test. It has also the benefit for identifying hidden trends as well as graphical illustration of the trend variability of extreme events of annual and seasonal rainfall over the classical M–K test.

The rainfall patterns during summer and spring seasons have shown increasing and decreasing tends, which could have implications to accelerate extreme events (such as surface runoff and soil erosion) and low agricultural productivities, respectively. Hence, there is a need for planning of effective adaptation and mitigation strategies at local scale. the involvement of stakeholders together with government organizations is essential in planning efficient strategies in order to minimize the negative impact of climate change in such rain-fed agricultural regions. The river and various small streams are accessible in the catchment. For the production of crops, especially when the Belg season rain declines, implementing small-scale irrigation schemes using such surface as well as groundwater sources will be very important to enhance agricultural productivity. To compensate and effectively utilize the highly variable wet season rainfall, providing short duration crop varieties is recommended in the area. Finally, the study findings have an essential contribution to the detailed trend interpretations as it quantitatively assesses the annual and seasonal trends in the ‘low’, ‘medium’, and ‘high’, values.

The Ethiopian National Meteorological Service Agency (NMSA) is acknowledged for providing rainfall data used in this study. The authors would like to acknowledge Jimma and Assosa universities for the educational support. We are also pleased to thank the editor and reviewers of the manuscript for their time and constructive comments that we used to improve the quality of this manuscript.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for profit sectors.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.