Characterizing the variability and trend of rainfall in central highlands of Abbay Basin, Ethiopia: using IMERG-06 dataset

ABSTRACT

This study examines the spatiotemporal variability and trends of rainfall in the central highlands of Abbay Basin using the Integrated Multi-satellite Retrievals for GPM (IMERG-06) dataset. Parameters on onset and cessation dates, dry spells, Coefficient of Variation (CV%), and Standardized Rainfall Anomalies were utilized to evaluate rainfall variability and seasonality. Trend analysis was carried out using Mann-Kendall test and Sen’s slope estimator. The results highlighted a consistent pattern of late onset and early cessation of rainfall. The CV values for annual (10.7%), kiremt (11.7%), and belg (10.6%) seasons rainfall indicate moderate variability, while bega (22.6%) rainfall showed higher variability. SRA findings reveal episodic fluctuations between wet and dry years. The trend analysis demonstrates a statistically significant increasing trend in annual (9.14 mm/year) and belg seasons (6.94 mm/year) rainfall. However, kiremt season exhibited statistically non-significant (p > 0.05) decreasing trend. The temporal variability in rainfall onset, cessation dates and duration significantly impacts agricultural processes such as tillage and planting operations, affecting crop production. While the study area holds potential for agricultural activities, climate variability may reduce productivity. Therefore, the study findings are valuable for agricultural developers, planners and water resource managers. Including additional weather elements is crucial to enhance the precision and reliability of these findings for practical applications.

KEYWORDS:

• Dry-spell
• IMERG-06
• Mann-Kendall
• seasonal
• variability
• west gojjam

Introduction

Climate change and variability are real, rapidly advancing and widespread threats that the earth is facing (Adeoti et al., 2016). For example, (IPCC, 2022) stated that the global surface temperatures have increased by approximately 1.1°C since the pre-industrial era and 1.5°C above pre-industrial era. Global sea levels have risen by about 15 cm during the 20th century, and the rate of rise has increased in recent decades (IPCC, 2019). There have been shifts in rainfall patterns, including increased rainfall in some regions and decreased in others (IPCC, 2014). Further, climate change has altered annual and seasonal features of rainfall along with its spatial distribution (Banerjee et al., 2020).

In Africa, precipitation variability is expected to increase, while precipitation is likely to decrease over most regions of sub-Saharan Africa (Nicholson, 2017). According to Banerjee et al. (2024), average land surface temperature increased by 0.01°C/year along with decreasing rainfall 1.89 mm/year. However, this system in sub-Saharan Africa is a risky proposition due to its low and irregular rainfall and exposure to long periods of drought (Tadesse et al., 2014). In most part of Ethiopia in the middle of the 21st century, temperature will increase greater than 1°C and annual rainfall will increase (in central highlands, southern, southwestern parts of the country, in Afar, Ogaden and Danakil depression), whereas rainfall will decrease in south in Negele area. Although annual precipitation over Ethiopia will increase in the middle of the 21st century, it will decrease during the main rainy season of June, July and August (Tamiru, 2020).

Agriculture is ranked as the most susceptible sector to climate variability impacts and so do the livelihoods of subsistence farmers and pastoralists (Sherbinin et al., 2008). This sector is the backbone of the Ethiopian economy, providing livelihoods for millions of people and playing a crucial role in poverty reduction (ATA) (https://www.ata.gov.et/about/accessed on December 21/2023. It employs over 70% of Ethiopia’s population and contributes around 35% to the country’s GDP (Dube et al., 2019; Yurdakul & Kamasak, 2020) and 85% of export revenue (Berhane et al., 2018). Climate variability, including increased temperature and changes in rainfall patterns, negatively affects crop yield in Ethiopia. In addition, the temporal distribution of rainfall during the growing season has an important influence on crop yield and can induce food shortage and famine (Araro et al., 2019; Bewket & Conway, 2007).

Temperature and rainfall records from meteorological stations show a significant warming trend and changes in rainfall patterns over the past few decades. For example, a study by (Zewdu et al., 2019) found a consistent increase in temperature. While (Bewket & Conway, 2007) found a decrease in total annual rainfall in parts of the Amhara region over a 40-year period. In addition, Ayalew et al. (2012b) showed that the onset date of the growing season has been undergoing a little shift, whereas the end date of the rain is noted to be shortened in most of the areas found in the Amhara region. The maximum temperature during the Kiremt season varies between 0.4 and 0.6°C decade⁻¹, and the belg season temperatures show more rapid increase greater than 0.6°C decade⁻¹ (Addisu et al., 2015).

Characterization of the spatiotemporal variability of meteorological elements is a prerequisite for a better understanding of the climate at regional and local scales (Asfaw et al., 2018; Gebrechorkos et al., 2019). Gauge-based rainfall measurement is tangible, direct and accurate. However, installing a dense gauge-based network to capture rainfall variability can be impractical (Dinku et al., 2018). Due to this, most of the time, gauge-based data are inadequate and incomplete (Kaspar et al., 2022). Satellite-based rainfall estimates could be a good alternative, where gauge-based observed rainfall data are not available or inconsistent (Banerjee et al., 2020; Wedajo et al., 2021). Huffman et al. (2020) suggested that the combination of microwave and infrared climate estimates can be a helpful approach that resulted from a high-resolution satellite precipitation algorithm.

A number of studies were conducted on the spatiotemporal variability of climate, for instance; Alemu and Bawoke (2020), Ayalew et al. (2012a), Ayalew et al. (2012b), Bewket and Conway (2007), Mengistu et al. (2014), Legese et al. (2016), Yohannes et al. (2017), M. Teshome (2017), Weldegerima et al. (2018), Mebratu and Prem (2020), Abeje et al. (2019), Addisu et al. (2015), Arragaw and Woldeamlak (2017), Gebrehiwot and Van (2013), Kassie et al. (2014), Senbeta and Olsson (2009). Most of these studies reported that the climate system of most of their study areas showed change and variability in the form of increasing temperature and insignificant but decreasing trends of rainfall. For example, the study made by Asfaw et al. (2018) showed statistically significant trends of declining annual and kiremt rainfall while the belg rainfall did not show a significant declining trend. Furthermore, they have also observed that there was intra- and inter-annual variability of rainfall in northcentral Ethiopia. Similar results indicated a decreasing, but insignificant trend for annual and seasonal (kiremt and belg) rainfall was noted by (Ayalew et al., 2012b; Bewket, 2009; Bewket & Conway, 2007; Rosell & Holmer, 2007). In contrast, Mengistu et al. (2014) reported increasing but statistically insignificant trends of the annual total, summer, autumn and winter rainfall, over the Upper Blue Nile river basin. The majority of these authors mentioned above used meteorological gauge-based and some of them used blended data for analyzing the climate system of their study areas.

While previous studies have reported changes/variability in temperature and rainfall, there was a need to conduct a detail and site-specific analysis focused on the spatiotemporal variability and trends of rainfall using satellite-based rainfall dataset, i.e. the Integrated Multi-satellitE Retrievals for GPM (IMER-06). This study provided more accurate and current assessment of the rainfall features that may have occurred in recent years for people where their livelihood depends on rainfed agriculture. Therefore, the key objectives of this study were to analyze the spatiotemporal variability and to examine the trend of rainfall over west Gojjam zone, central highlands of Abbay Basin, using the satellite-based datasets GPM (IMERG-06) for 2001–2020.

Materials and methods

Description of the study area

This study was carried out in west Gojjam zone, which is located between 10°16′ to 11°54′ Latitude and 36°31′ to 37°57′ Longitude in the central highlands of Abbay Basin (Figure 1(a)). West Gojjam zone has a total area of 13,311.9 Km² and the lowest and the highest peaks of the area are 684 and 3592 m a.s.l, respectively (Csa, 2017).

Figure 1. Location and agroecological map (a), mean monthly rainfall, minimum, and maximum temperatures (b) of west Gojjam zone, central highlands of Abbay Basin, Ethiopia.

Based on altitude, west Gojjam has four climatic zones known as Wurch, Dega, Weyna Dega and Kolla. Wurch (upper highlands) has a type of alpine climate (Weldegerima et al., 2023). The annual average temperature is less than 11°C. Annual rainfall is between 900 and 2,200 mm/yr, the zone exists at altitudes more than 3200 m (m) above mean sea level (amsl). Dega (highlands) is also known as the alpine vegetated climate. The annual rainfall is between 900 and 1,200 mm/yr. This zone is a coldish less than temperate zone with altitudes ranging between 2,300 and 3200 m amsl. Weyna Dega (midlands) or the temperate zone ranges between 1500 m and 2300 m in elevation. Annual rainfall lies in between 800 and 1,200 mm/yr. Temperature ranges between 16 and 20°C, and Kolla (lowlands) looks like the hot zone. It is much drier and warmer than the Weyna Dega zone. This zone received annual rainfall of less than 800 mm/y. This zone elevated between 500 and 1,500 m amsl (Figure 1(a)). Based on 30 years of temperature and rainfall records of National Meteorological Agency of Ethiopia, the average minimum and maximum temperature of the zone was 11.2°C and 25.9°C, respectively. The mean annual rainfall varies in between 1248.4 mm (2002) and 1699.7 mm (2017), mean annual and kiremt season rainfall of the area were 1365 mm and 997 mm, respectively (NMA, 2020). According to the population projection of Ethiopia, west Gojjam zone has a total population of 2,542,221, of which 2,220,038 people lived in rural areas (Csa, 2017). The majority of the population is living in rural kebeles engaged in rainfed agricultural activities. It is the main source of livelihood for this community (Gebreegziabher et al., 2016). This zone has different agroecologies and played a pivotal role for the agricultural activities, there by contributing a significant role for the country’s economy. However, as the activity mainly depends on rainfed, the patterns and variability of rainfall influences the sector, may lead to severe droughts in some places and water logging in others (Zeleke et al., 2023).

The study area has three distinct seasons, each season comprises of 4 months. For example, kiremt season includes, June, July, August and September; bega season has October, November, December and January; and belg season months are February, March, April and May (Abebe, 2006; Segele et al., 2015; Wbg, 2021). According to Omay et al. (2023), the main producing period of this area is Kiremt season (Figure 1(b)).

Data sources

Meteorological station data

Daily rainfall data were collected from 15 gauge-based stations over the period of 2001– 2020 from the National Meteorological Services Agency of Ethiopia to validate the data extracted from the satellite-based datasets (Taye et al., 2023).

IMERG-06 satellite data

The Integrated Multi-satellite Retrieval for GPM (IMERG-06) product combines precipitation observations using infrared and microwave sensors from a constellation of partner satellites, united by the GPM Core Observatory, to provide near real-time half-hourly precipitation estimates at 0.1 degree resolution for the entire globe; https://gpm.nasa.gov/data/imerg accessed 2/1/2023, (Huffman et al., 2020; Sharma et al., 2020). The product Integrated Multi-satellitE Retrieval, IMERG-06 final daily time step rainfall, was extracted the point-pixel data as per the station locations (Taye et al., 2023). The data available on IMERG-06 dataset begin from 2001.

Methods of data analysis

Validation of IMERG-06 rainfall dataset

Temporally and spatially direct, consistent and long-term ground-based observed rainfall data are generally accurate near the sensor location and important (Behrangi et al., 2011; Nguyen et al., 2019). However, the observed rainfall data are usually lacking of such quality and are subjected to missing data, this may be inconsistent distribution of meteorological station, limited number of meteorological station in the region, and interruption of data recording by different reasons (Githungo et al., 2016; Kaspar et al., 2022; Kidd et al., 2012; Maggioni et al., 2016; Sisay et al., 2017).

The performance of IMERG-06 was evaluated quantitatively against the reference rain gauge observation data using categorical and continuous statistical techniques. The categorical techniques, such as Probability of detection (POD), False alarm ratio (FAR), Critical Success Index (CSI), and Accuracy, were employed to evaluate the detection performance of the data set. Further, the estimation capacity of the data set was assessed using continuous statistical techniques like bias ratio, Relative bias, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Correlation Coefficient (CC) indices (Taye et al., 2023).

Spatiotemporal variability

Characteristics of rainfall features

Onset, and cessation date, duration and dry-spell length of rainfall

Characterizing rainfall features in terms of the starting date, and the cessation date of rainfall, the length of rainy season and dry-spell length in main rainy season is too important, especially for areas where their economic activities highly rely on rain-fed agriculture. The main rainy season of the study area is kiremt (Gummadi et al., 2018) and Figure 1(b).

Daily rainfall data were subjected to a detailed analyses using several sequences of statistical packages. For instance, the satellite data were analyzed with respect to the onset, cessation date, duration and dry-spell length of rainfall in kiremt season via R-Instat 0.7.3.1 (Roger et al., 2021). The outset, the Box and Whiskers plotting techniques were used for examination and visual illustration of inter-annual spread of the rainfall series with respect to the onset date, cessation date and duration of the main wet season rainfall (Banacos, 2011).

The onset date is the first day of the year when the dataset record 20 mm of total rainfall for at least for three consecutive days that were not followed by eight or more days of dry days (<0.1 mm day⁻¹) in subsequent 30 days (Segele & Lamb, 2005). On the other hand, the cessation date of the rainy season is defined basically as the first day of a dry spell (0.1 mm day⁻¹) at least 20 days duration after the occurrence of onset and the soil water reaches zero (Segele & Lamb, 2005; Stern & Coe, 1982). In addition, length of the rainy season is an important factor in determining the maturity of crops/seeds grown under different conditions (Bogale, 2023). Length of the rainy season was calculated by taking the differences between the cessation date of rainfall and onset date of rainfall at a given period of time (Mupangwa et al., 2011). Dry-spell length represents number of consecutive days without rainfall during kiremt season (Kebede et al., 2017). A dry spell is a period of 3 or more days of lack of rainfall during the wet season (Sanchi et al., 2021). The occurrences of dry spells within the kiremt season were identified for a threshold of less than 0.85/1 mm^−d within the season’s months (Abebe, 2006).

Pattern of rainfall

From daily rainfall data, monthly, seasonal and annual averages were derived. To summarize the data, descriptive statistics such as mean, Coefficient of variation (CV%), standard deviation (SD) and percentage rainfall contribution of seasons for the annual total were calculated. In addition, general linear model was employed to check rainfall variation between study stations, and within study years was either significant or not. Finally, the spatial distribution of annual, kiremt, bega and belg rainfall was visually illustrated using ArcGIS 10.6.

Coefficient of variation (CV%)

The coefficient of variation (CV) was calculated to evaluate the extent of spatial and temporal variability of mean annual and seasonal (kiremt, bega and belg) rainfall for each station. Coefficient of variation less than 20%, between 20 and 30% and greater than 30% indicate lower, moderate and higher variability of rainfall, respectively (Hare, 2003). It is computed as:

$\mathit{CV}=\frac{\sigma }{\bar{\mathrm{X}}}\times 100$

where CV is the coefficient of variation of rainfall, $\mathit{\sigma }$ is the standard deviation of rainfall, and $\bar{\mathrm{X}}$ is the mean of rainfall over the year of observations (2001–2020).

Standardized rainfall anomaly

The Standardized Rainfall Anomalies (SRA) were calculated and graphically presented to evaluate the inter-annual fluctuations of rainfall in the study area over the year of observations (2001–2020). The positive values represent periods of above-normal rainfall/wet years from the long-term mean while the negative values reflect below-normal rain/dry years (Muthoni et al., 2019). The SRA was calculated according to Agnew and Chappell (1999):

$\mathit{SRA}=\frac{(\mathit{pt}-\mathit{pm})}{\sigma }$

where SRA is standardized rainfall anomaly, Pt is an annual rainfall in year t, Pm is long-term mean annual rainfall over a period of observation and σ is the standard deviation of annual rainfall over the period of observation.

Trend analysis of rainfall

Mann-Kendall test

Unlike parametric methods, non-parametric tests are more flexible as they do not depend on normally distributed time series data (Gilbert, 1987). Mann-Kendall (MK) test, one of the non-parametric method, and Sen’s slope estimates were employed to examine the temporal trend and magnitude of annual and seasonal rainfall, respectively. The MK test uses the correlation between the ranks of a time series and their sequence. It computes the difference between the later measured values and all early measured values for a time series of interest over time (Kendall, 1938). When a data value from a later time period is higher than a data value from an earlier time period, the MK test statistic S is incremented by 1. On the other hand, if the data value from a later time period is lower than a data value sampled earlier, S is decremented by 1. The MK statistics S is computed as:

$\mathit{S}=\sum _{\mathit{i}=1}^{\mathit{N}-1}\sum _{\mathit{j}=\mathit{i}+1}^{N}sgn\left(X_j-X_i\right)$

where S is Mann-Kendall’s test statistics, X_i and X_j are the sequential data values of the time series in the years i and j (j > i), and N is the length of the time series. A positive S value indicates an increasing trend and a negative value indicated a decreasing trend in the data series.

where X_i and X_j are sequential data values for the time series data of length N;

$\mathit{Sgn}\left(X_j-X_i\right)=\{\begin{array}{l}1\mathit{if}\left(X_j-X_i\right)\text{>\hspace{0.17em}}0\\ 0\mathit{if}\left(X_j-X_i\right)=0\\ -1\mathit{if}\left(X_j-X_i\right)\text{<}0\end{array}$

For samples, N ≥ 10, the S statistic is approximately normally distributed and the mean of S is zero, and the variance of S is given as:

$\mathit{Var}(\mathit{S})=\frac{1}{N}\left[\mathit{N}(\mathit{N}-1)(2\mathit{N}+5)-\sum _{\mathit{i}=1}^m(t_i-1)(2t_i+5\right]$

where N is the number of observations in the time series, m is the number of tied groups, and t_i is the number of observations in the i^th group.

The significance of a trend is computed by the Z sore using;

$\mathit{Z}=\{\begin{array}{l}\frac{\mathit{S}-1}{\sigma },\mathit{ifS}>0,\\ 0,\mathit{if}\mathit{S}=0,\\ \frac{\mathit{S}+1}{\sigma },\mathit{ifS}\text{<}0\end{array}$

When the Z value exceeds either of the confidence limit lines, it shows a significant trend at a given significance level. Hence, the null hypothesis (H_o) is rejected and in place, the alternative hypothesis (H₁) is accepted (Mann, 1945).

Sen s estimator of slope

Sen s slope estimator (Sen, 1968) is employed to estimate the magnitude of the trends in time series data. If a linear trend exists in a time series, then the true slope of the trend is estimated using a Sen-Theil trend line (Sen, 1968; Theil, 1950), an alternative to linear regression, in combination with the MK test. The slope (T_i) of all data sets is computed as:

$T_i=\frac{X_j-X_k}{\mathit{j}-\mathit{k}}$, for i = 1, 2, 3 … N

where X_i and X_k are taken as data values at time j and k (j > k), respectively. The median of these N values of T_i is denoted as Sen s estimator of slope, which is expressed as:

$Q_i=\{\begin{array}{l}\mathit{T}\frac{(\mathit{N}+1)}{2},\mathit{N}\mathit{is}\mathit{odd},\\ \frac{1}{2}\left(\mathit{T}\frac{N}{2}+\mathit{T}\frac{(\mathit{N}+2)}{2}\right),\mathit{N}\mathit{is}\mathit{even}\end{array}$

where Q_i indicates the β value of the slope estimator in Sen s slope method. A positive value of Q_i suggests an increasing trend and a negative value of Q_i offers a decreasing trend in the time series.

Results and discussion

Validation result

The result indicated that IMERG-06 had better detection and estimation capacity. As it was clearly stated by Hartke and Wright (2022), indirect precipitation measurements of satellite products experience uncertainty and retrieval conditions. However, IMERG-06 has comparable performance in precipitation detection to gauge-based data at very low gauge densities. For the daily time scale, the value of POD and FAR, IMERG-06 had 0.73 and 0.31 values, respectively. During the kiremt season, IMERG-06 had a POD value of 0.82 and 0.18 FAR. On the other hand, the estimation performance of this data set at the daily time scale was equivalent to (r ∼0.8). In addition, in terms of RMSE, IMERG-06 provides relatively lower error, with the values for both daily (50.3 mm) and kiremt season (75.9 mm) than others and MAE (7.7 mm) rainfall for daily time scale (Taye et al., 2023). Furthermore, results of similar studies (Khodadoust Siuki et al., 2017; Wedajo et al., 2021) showed the same concept as IMERG-06 had a reasonable agreement with the gauge-based data and could be a reliable rainfall data source for climate analysis.

Spatiotemporal variation of rainfall

Characteristics of rainfall features

Figures 2 show the variability of the onset date, cessation date and duration of rainfall in each study station. The mean onset date of west Gojjam was 142 (22-May) day of the year (DOY) (Table 1). The mean onset date in the study stations ranged from 132 (12, May) at Gundil to 166 (15, June) at Finoteselam DOY. In addition, the mean onset date of most of the study stations exhibited spatial variation (Table 1). On the other hand, mean onset date of the year; 2001 (139), 2005 (138), 2008 (131), 2010 (133), 2014 (135), 2015 (136), 2016 (130), 2017 (134) and 2020 (134 DOY) was slightly similar and early compared with mean onset date of the study zone. While during 2002 (159), 2003 (162), 2004 (157) and 2009 (160 DOY) and were late (Table 2).

Figure 2. Box and whisker plots for onset date (a), cessation date (b), and length the rainy season (c) in west Gojjam zone, central highlands of Abbay Basin (2001–2020).

Table 1. Mean onset, cessation date, length of the rainy season and dry-spell length for study stations (2001–2020).

Study stations	Onset date	DOY for onset date	Cessation date	DOY for cessation date	Length of the rainy season	Dry-spell length
						Mean	SD
Adet	24-May	144	22-Oct	295	151	12	4
Ayehu	16-May	136	25-Oct	298	162	9	3
Bahir Dar	17-May	137	27-Oct	300	163	8	2
Bure	25-May	145	26-Oct	299	154	8	2
Dembecha	20-May	140	19-Oct	292	152	10	3
Durbete	7-June	158	1-Nov	305	147	7	2
Feresbet	7-June	158	4-Oct	277	119	31	7
F/Selam	15-June	166	19-Oct	292	126	8	3
G/Mariam	14-May	134	27-Oct	300	166	8	3
Gundil	12-May	132	29-Oct	302	170	6	1
Kembaba	16-May	136	27-Oct	300	164	10	3
Meshenti	19-May	139	29-Oct	302	162	9	3
Quarit	19-May	139	20-Oct	293	155	11	2
Sekela	14-May	134	22-Oct	295	161	9	3
Shindi	16-May	136	24-Oct	297	161	8	3
Tisabay	20-May	140	25-Oct	298	157	12	4
Wetetabay	16-May	136	1-Nov	305	169	7	2
Zegie	23-May	143	24-Oct	297	154	10	3
Mean	22-May	142	24-Oct	297	155	10	2
Min	12-May	132	4-Oct	277	119	6
Max	15-June	166	1-Nov	305	162	31

Table 2. Mean onset date, cessation date, length of the rainy season and total rainfall of main rainy season (2001–2020) in west Gojjam zone, central highlands of Abay Basin.

Year	Onset date	DOY for onset date	Cessation date	DOY for cessation	Length of the rainy season	Total rainfall (mm)
2001	18-May	139	19-Oct	292	153	1066
2002	7-June	159	19-Oct	292	134	999
2003	10-June	162	23-Oct	296	133	1124
2004	5-June	157	27-Oct	300	143	1037
2005	17-May	138	28-Oct	301	164	996
2006	14-May	135	5-Nov	309	175	1195
2007	19-May	140	22-Oct	295	155	967
2008	10-May	131	13-Oct	286	155	989
2009	8-June	160	12-Oct	285	124	822
2010	12-May	133	24-Oct	297	164	995
2011	19-May	140	16-Oct	289	149	995
2012	24-May	145	14-Oct	287	142	939
2013	22-May	143	2-Nov	306	163	1015
2014	14-May	135	29-Oct	302	167	965
2015	15-May	136	27-Oct	300	164	765
2016	9-May	130	2-Nov	306	177	981
2017	13-May	134	31-Oct	304	170	951
2018	20-May	141	19-Oct	292	151	1076
2019	25-May	146	26-Oct	299	153	980
2020	13-May	134	30-Oct	303	168	1090

Mainly the onset date of rain in 2003 and 2009 was relatively too late for almost all stations: such as the onset dates of Adet, Bahir Dar, Durbete, Feresbet, Tisabay and Zegie were 176 and 187; 157 and 172; 157 and 170; 175 and 184; 168 and 170; and 159 and 170 DOY, respectively (Table 2). Except for Durbete and Finoteselam, the lower extreme of the onset date of the study station was 122, early onset of rain, while the upper extreme of onset date, late onset of rain, of Finoteselam was 211 in 2015, Feresbet (194) in 2002, Adet (187) in 2009, Quarit (179) in 2003 (Figure 2(a)). The 10^th, 25^th, median, 75^th, and 90^th percentile of the box and whisker plots indicated that Gundil had early onset date, i.e. 122 DOY, among the study stations. While 25^th and 10^th percentiles of Finoteselam, median, 75^th and 90^th percentile of Feresbet exhibited late onset date, 158, 155 and 168, 189 and 182 DOY, respectively, than the others (Figure 2(a) and Appendix 1). The delayed rainfall onset in those years may be due to large-scale climate drivers like El Niño events or other teleconnection patterns affecting precipitation timing and distribution in the region. Research by (Wang & Zhou, 2008) highlights ENSO’s impact on East Africa’s rainfall patterns, leading to delayed or reduced precipitation in certain years. (Mason & Goddard, 2001) have also shown how the IOD can influence rainfall variability in the Horn of Africa, potentially delaying the start of the rainy season.

The findings of the present study are in line with the study by (Kebede et al., 2017; Markos et al., 2023), which showed that the mean onsets of the rainy season for the majority of study stations occur during the third meteorological decade of May. The above analysis of onset date indicated variation through all observation years in all study stations. While this study result in contrast with Omay et al. (2023), who stated that the kiremt season recorded widespread decrease in onset date variability over most parts of highlands of central and northern Ethiopia, this might be due to the difference in topographical and vegetation cover. The temporal variability of onset date of rainfall has an impact on all agricultural processes including tillage and planting operations, which in turn affect crop production (Wakjira et al., 2021). Therefore, information about the start of the rains is too crucial for appropriate agricultural decision making and reduction in yield of the crops is avoided by taking appropriate action (Abebe, 2006).

Figure 2(b) portrays that the mean cessation date of rainfall of the study stations ranged from 277 (4,October) at Feresbet to 305 (1,November) DOY at Wetetabay. The mean cessation date of the study zone was 297 (24,October) DOY (Table 1). In addition, the cessation date of the study area showed spatial and temporal variation (Tables 1 and 2). The boxes and whiskers at 25^th, 10^th, median, 75^th, and 90^th percentiles indicated that rainfall terminated relatively too early at Feresbet compared with the remaining stations. While Wetetabay showed relatively better at 25^th and 90^th percentiles, and Durbete exibited 10^th and 75^th percentiles of relatively better termination day among the study stations (Figure 2(b) and Appendix 1). The result of the study indicated that the rainy season of the study zone, i.e. 297 DOY, was terminated early relative to the 308 DOY that used to cease in most part of the Amhara region (Ayalew et al., 2012b).

The result of the study is in line with the results of Taye et al. (2013) who revealed that the mean cessation date of rainfall in western Amhara region which was between 266 to 304 DOY during their study period between 1979 and 2008. Too early cessation of rainfall was found in Feresbet station for almost all study years, i.e. mean cessation date was 276 DOY (3, October) (Table 2). This station cessation date of rainfall was similar to stations studied by Kebede et al. (2017) in the Upper Baro-Akobo, Ethiopia. On the other hand, too late cessation of rainfall was observed in Kembaba 333 DOY (29,November) during 2006 (Fig ure 2b). This is ascertain by the study of Omay et al. (2023) who stated that the study years 2011–2020 showed delay of 7–14 days in mean cessation dates over central and northwestern Ethiopia. The study of Segele and Lamb (2005) revealed that the tendency for delayed cessation for some station may be due to orographic effects.

The result of present study showed that the mean rainfall duration of the study stations ranged in between 119 and 170 days. From studied stations, Feresbet had the least duration (<120 days) of rainfall, whereas Gundil had the maximum duration of rainfall, i.e. 170 days (Table 2). The majority of studied stations had long-duration of rainfall during main rainy season, i.e. >150 days (except, Feresbet and Finote selam) (Figure 2(c)). The boxes and whiskers illustrated the variation of length of the rainy season, for example, short duration of rainfall recorded at Feresbet for 25^th and 10^th percentiles including median and at Finoteselam for 90^th and 75^th percentiles. Whereas Gundil had better length of the rainy season for 25^th, 10^th percentiles and for the median and Wetetabay was relatively best for 90^th and 75^th percentiles than others (Figure 2(c) and Appendix 1). This study result agreed with the findings of Ayalew et al. (2012b), Taye et al. (2013), and Omay et al. (2023) the majority of their study areas had long-duration of rainfall in the Amhara region and western Ethiopia. This as explained by Tarkegn and Jury (2020) is due to the earlier arrival and later departure of rains. The shortest duration of rainfall was observed in Feresbet in contrast to the findings of Gummadi et al. (2018) who studied the spatiotemporal variability and trends of precipitation and extreme rainfall events in Ethiopia from 1980 to 2010 and stated that the duration of the rainy season in the highland was high. In general, the 10^th −90^th and 25^th-75^th percentiles of the box and whisker plots indicated the presence of spatiotemporal variation of onset, cessation date and length of the rainy season among stations over the study period (Figure 2 and appendix 1).

The mean of study stations dry-spell length varies in between 6 and 31 days. According to Froidurot and Diedhiou (2017) and Gong et al. (2005), the dry-spells with less than 10 days referred as short dry-spell while the long dry-spells refer to the other. The shortest dry-spell length was found in Gundil and followed by Durbete and Wetetabay, whereas Feresbet recorded the longest dry-spell length, i.e. 31 days. Adet, Dembecha, Kembaba, Quarit, Tisabay and Zegie had almost the same length of the dry-spell (Table 1). This is similar with the study of Segele and Lamb (2005) who stated that central and western Ethiopia have the shortest dry-spells. Linear regression were used to estimate the temporal trend of the series for each station. The visual inspection (Figure 3(a)) indicated a clear increasing and decreasing trend, for example the trend line of dry-spell at Adet, Ayehu, Bure, Durbete, Finoteselam, G/Mariam and Shindi have indicated decreasing trend. While Bahir Dar, Kembaba, Meshenti, Sekela and Tisabay showed increasing trend. From the ternd lines, it has been noticed that the change is not significant. Further, study stations with short dry-spell length had less R² value. This indicated no trend of dry-spell length over the study period, such as, Wetetabay and Gundil. Similarly, the mean result indicated that dry-spell length over west Gojjam zone, central highlands of Abbay Basine showed almost no trend over the period of analysis (Figure 3(b)). On the other hand, Adet (12), Dembecha (10), Feresbet (31), Kembaba (10), Quarit (11), Tisabay (12) and Zegie (10) had long dry spells (Table 1). Long dry-spells have a strong influence on agriculture and increasing the occurrence of drought (Gong et al., 2005; Huth et al., 2000). Sanchi et al. (2021), also emphasized that dry-spell threatens agricultural productions and undermining the economic development of rural farmers.

Figure 3. Dry-spell length for all study stations (a) and, west Gojjam zone (b) (2001–2020).

In general, from the Dega agroecological zone, Feresbet had late onset, early cessation dates, and long dry-spell length, June 7, October 4 and 31 days, respectively, than Sekela and other Weyna Dega agro ecological zones. Among Weyna Dega agroecological zones Finoteselam, and Durbete had late onset dates and Adet had long dry-spell length (Table 1 and Figure 2(a)).

Pattern of rainfall

The results of the study revealed the temporal and spatial features of rainfall. The mean annual rainfall of study stations ranged in between 1029 mm rainfall at Feresbet and 1605 mm rainfall at Wetetabay (Table 3). On the other hand, mean of kiremt season rainfall for the study years (2001–2020) was ranged in between 765 mm rainfall in 2015 and 1195 mm rainfall in 2006. In addition, the year 2007, 2009, 2012, 2017 and 2014 received least amount of rainfall compared with the remaining study years (Table 2). This indicated that west Gojjam zone experienced spatiotemporal variation of rainfall.

Table 3. Mean annual and seasonal rainfall CV (%) and SD (mm) of all study stations, from 2001 to 2020.

Stations	Mean				Annual		kiremt		Bega		Belg
	Annual	kiremt	Bega	Belg	CV	SD	CV	SD	CV	SD	CV	SD
Adet	1210	906	127	177	10	126	13	114	32	40	38	67
Ayehu	1212	862	120	230	13	162	17	148	30	36	33	75
Bahir Dar	1485	1068	179	238	11	169	11	113	38	68	32	77
Bure	1243	870	137	236	9	115	13	114	31	43	32	77
Dembecha	1377	967	159	251	11	154	9	91	43	68	30	75
Durbete	1492	1141	154	197	9	131	10	110	32	49	39	77
Feresbet	1029	796	41	192	20	202	19	153	74	30	45	87
Finoteselam	1277	906	139	232	9	110	11	98	34	48	32	75
G/Mariam	1353	984	121	248	13	171	14	137	34	42	31	77
Gundil	1492	1076	162	254	10	142	9	95	30	48	34	87
Kembaba	1488	1093	175	220	11	168	13	140	31	55	37	81
Meshenti	1543	1140	177	226	13	203	13	145	36	63	44	99
Quarit	1419	1009	164	246	12	168	10	98	47	77	30	73
Sekela	1340	964	149	227	12	159	10	101	52	77	31	69
Shindi	1248	884	131	233	11	137	14	122	27	36	32	75
Tisabay	1311	963	149	199	13	164	13	128	38	56	43	86
Wetetabay	1605	1203	183	219	10	159	8	101	30	55	38	83
Zegie	1443	1120	140	183	11	156	11	127	36	51	45	82
Mean	1365	997	145	223	11	146	12	115	23	33	11	24

CV = Coefficient of Variation, SD = Standard Deviation.

West Gojjam zone received a mean annual rainfall of 1364.8 mm rainfall with a standard deviation of 146 mm rainfall. Data from IMERG-06 data set for all study stations exhibited that all stations experienced more than 1000 mm mean annual rainfall (Table 3). The mean rainfall of the kiremt season was 997 mm rainfall with a standard deviation of 115 mm, which contributed to the maximum share of the annual rainfall (73%) of the study area (Tables 3 and 4). The result of the present study is in line with the study by Gummadi et al. (2018), who stated that the summer season (June to September) contributed 74% of the annual rainfall for most parts of Ethiopia. On the other hand, the mean belg and bega seasons rainfall was 223 mm and 145 mm with a standard deviation of 23 mm and 33 mm, respectively. From the present study, it is 3 and 4 possible to note that belg rainfall makes a considerable contribution to the annual total; belg and bega seasons contributed 16.3% and 12.4% rainfall, respectively (Table 4). Similarly Abebe (2006) and NMA (2009) noted that the contribution of belg season rainfall over the north, northeastern and eastern highlands is ranging from 5% to 30%.

Table 4. The annual and seasonal MK trend analysis of rainfall in west Gojjam zone, central highlands of Abbay Basin (2001–2020).

Rainfall (mm)	Min	Max	Mean	SD	CV%	Mkz	P value	Sen’s slope	Contribution of rainfall to annual (%)
Annual	1029	1605	1364.8	145.6	10.7	0.345	0.042*	9.14	100
Kiremt	796	1202	997.4	115.3	11.6	−0.205	0.234	−1.88	73.08
Bega	41	183	144.8	32.8	22.6	0.158	0.363	1.93	11.04
Belg	177	254	222.5	23.5	10.6	0.404	0.017*	6.94	16.31

MK_z=Mann – Kendall; SD= Standard Deviation; CV= Coefficient of Variation; *statistically significant at 0.05.

The spatial distribution of the mean annual, kiremt, bega and belg rainfall is presented in Figure 4. The visual inspection of mean annual rainfall showed that the northern and northwestern parts of the study area received maximum rainfall while the southern and southeastern and western parts of the study area received relatively low rainfall (Figure 4(a)). Similarly, during kiremt and bega seasons, the highest rainfall was observed in the northern and northwestern parts of the zone (Figures 4b,c). In contrast, during the belg season southern, southeastern and western parts of the zone received the highest rainfall (Figure 4(d)). This is due to the migration of the Inter-Tropical Convergence Zone (ITCZ) and this resulted a strong inter-annual variability across the country (Wbg, 2021).

Figure 4. Spatial distribution of mean annual (a), kiremt (b), bega (c), and belg (d) rainfall in west Gojjam zone, central highlands of Abbay Basin (2001–2020).

Temporal variability

The temporal variability of annual and seasonal rainfall was assessed using a coefficient of variation, which showed a strong relationship with the amount of rainfall received during that specific period (Figure 5). The CV of annual, kiremt, bega, and belg seasons rainfall of the study stations varied between 9 to 20%, 8 to 19%, 27 to 74%, and 30 to 45%, respectively (Table 3). The highest variation was observed in the highland (Dega agro-ecological zone) area of the study, i.e. Feresbet station. During bega and belg seasons, the variation is greater than 30% in almost all study stations. The CV of annual (11.7%), kiremt (12.6%) and belg (10.6%) rainfall revealed moderate variability, whereas CV of bega (22.6%) rainfall had higher variability in the study area (Table 4). The result of the study is in agreement with the finding of a study by Anteneh (2022), who reported lower values of CV for stations and seasons with higher annual rainfall amounts than for stations with lower annual rainfall amounts.

Figure 5. Spatial distribution of (CV%) annual (a), kiremt (b), bega (c) and belg (d) rainfall in west Gojjam zone, central highlands of Abbay Basin (2001–2020).

The CV values of the spatial distribution of rainfall indicated that eastern part of the study area had high variability of annual and bega season rains (Figure 5a and c). Eastern and southwestern parts of the study area had a high coefficient of variation during the kiremt season (Figure 5(b)). On the other hand, the northern part of the study area had high variability during the belg season (Figure 5(d)). As compared to annual rainfall, the bega seasonal rainfall had high variability up to 74.2% (Table 4).

The mean deviation of annual, kiremt, bega, and belg season rainfall for each station was calculated, and the spatial distribution of the mean deviation is illustrated in Figure 6. Annual rainfall indicated high deviation from the mean rainfall in the northern and, central parts of the west Gojjam zone (Figure 6(a)). Whereas northern and eastern part of the study zone exhibited high deviation from mean during kiremt season rainfall (Figure 6(b)). In addition during the bega seasons northern, northeastern and southwestern parts of the study area had high standard deviation (Figure 6(c)). Northern, eastern and western parts indicated high deviation from the belg season rainfall (Figure 6(d)). This deviation indicated the presence of spatiotemporal variation of rainfall in west Gojjam zone.

Figure 6. Spatial distribution of SD (mm) of annual (a), kiremt (b), bega (c) and belg (d) rainfall in west Gojjam zone, central highlands of Abbay Basin (2001–2020).

General linear model was employed to show the spatiotemporal variation for annual, kiremt, bega and belg seasons rainfall. The result indicated that the variation of annual, kiremt, bega and belg seasons rainfall showed stastical significance at p < 0.01. This showed the presence of substantial spatiotemporal variation of annual, kiremt, bega and belg season rainfall over the study period and in almost all study stations (Appendix 2).

Inter-annual fluctuation of rainfall

Figure 7 illustrates the temporal variability of rainfall in the study area from 2001 to 2020. The presence of inter-annual variability in rainfall, as indicated by standardized rainfall anomalies, can be justified by the complex interactions of various climatic factors influencing rainfall patterns. The observed percentages of positive anomalies (50%) and negative anomalies (45%) suggest significant fluctuations in rainfall amounts from year to year, reflecting the dynamic nature of the climate system. The small percentage (5%) indicating no deviation from the long-term mean annual rainfall underscores the rarity of years with rainfall amounts closely matching historical averages, further highlighting the variability inherent in rainfall patterns (Figure 7(a)).

Figure 7. Standardized anomalies of annual (a), kiremt (b), bega (c), and belg (d) rainfall in west Gojjam zone, central highlands of Abbay Basin (2001–2020).

The highest positive (2.4) and negative (−2.1) rainfall anomalies were observed in the years 2006 and 2009, which were extremely wet and extremely dry years, respectively. This can be attributed to the influence of climatic phenomena such as El Niño or La Niña events, which can significantly impact precipitation patterns. Research by Gobie and Miheretu (2021) has shown that strong El Niño events, like the one observed in 2006, can lead to above-average rainfall in certain regions, while La Niña events, as in 2009, are associated with drier conditions.

The years 2002, 2003, 2009, 2012, and 2015 were dry, whereas 2006, 2014, 2018, 2019 and 2020 were wet years. During the wettest years, annual rainfall has been around 2.4 times the standard deviations above the long-term (2001–2020) average. On the other hand, annual rainfall was found to be 2.1 times the standard deviation below the long-term average during the dry years (Figure 7(a)). As indicated by Zhang et al. (2018), the possible reason for such variabilities might be due to atmospheric circulation patterns, such as the Indian Ocean Dipole, influencing extreme wet and dry years in East Africa. According to this author, variations in large-scale climate drivers can result in the observed anomalies in annual rainfall, with wet years exceeding standard deviations above the long-term average and dry years falling below.

Substantial positive anomalies were observed during the kiremt season in the years 2001, 2003, 2006, 2018, and 2020. The rainfall in the wettest years deviated from the mean rainfall by factors of 0.73, 1.34, 2.1, 0.84 and 0.98 (Figure 7(b)). Extreme negative anomalies were recorded during the kiremt season in 2009 and 2015, with deviations of 1.85 and 2.46, respectively. Positive rainfall anomalies during the kiremt season at all stations throughout the study period accounted for 60% compared to negative anomalies (Figure 7(b)). This finding aligns with a previous study by A. Teshome and Zhang (2019), which highlighted the erratic nature of standardized rainfall anomalies during the kiremt season in eastern and northwestern Ethiopia. Similar variability in rainfall was observed during the belg and bega seasons, with negative anomalies of around 40% and 55%, respectively (Figures 7c,d). Overall, the time series analysis of annual and seasonal rainfall in the west Gojjam zone indicates episodic fluctuations between wet and dry years. Similarly, the study of Abdi and Yimenu (2020) also revealed that both negative and positive rainfall anomalies were occurred between 1996 and 2016 in Quarit district of west Gojjam zone.

Trend analysis of annual and seasonal rainfall

The trend analysis of annual and seasonal rainfall revealed different patterns during the study period from 2001 to 2020. The Mann-Kendall (MK) test results indicated a statistically significant increasing trend in annual rainfall (9.14 mm/year), and the belg season (6.94 mm/year) (Table 4). A similar increasing trend, although not statistically significant, was observed during the bega season. Conversely, the kiremt season showed a non-significant decreasing trend at p < 0.05. These results align with Alemayehu et al. (2022), who reported varying degrees of increment in rainfall distribution in Northwest Ethiopia, excluding the kiremt season. Additionally, Alemu and Bawoke (2020) noted an overall increasing trend in annual and seasonal rainfall over the Amhara region, except during the winter season. In contrast, Asfaw et al. (2018) found a statistically significant decreasing trend in annual and kiremt season rainfall, with a decreasing but non-significant trend in the belg season in north-central Ethiopia. Similarly, the study of Abdi and Yimenu (2020) also reported that very slightly decreasing trend of rainfall was occurred in Quarit district of west Gojjam zone from 1997 to 2016.

Conclusions

The prime aim of this study was to examine the spatiotemporal variability and trends of rainfall in west Gojjam zone of the central highlands of Abbay Basin, utilizing the Integrated Multi-satellitE Retrievals for GPM (IMERG-06) dataset from 2001 to 2020. Giving crucial role of rainfall in the livelihood of people who rely on rainfed agriculture, this study focused on onsets, cessation date, length of the rainy season, and dry-spell length of rainfall for kiremt season, CV% and SRA were used to evaluate variability and seasonality of rainfall. Mann-Kendall s test and Sen s slope estimator were used for trend analysis of rainfall.

The results of the study showed that the mean onset and cessation date of the study stations exhibited spatial and temporal variation. The mean onset date of central highlands of Abbay Basin was late, whereas the mean cessation date of the study area was too early, i.e. 142 and 297 DOY, respectively. Growing season length and quality are affected substantially by the frequency and duration of dry spells within a season. The distribution of mean annual and kiremt season rainfall showed that the northern and northwestern parts of the study area received maximum rainfall while the southern and southeastern and western parts of the study area received relatively low rainfall. Annual (11.7%), kiremt (12.6%) and belg (10.6%) seasons rainfall revealed moderate variability, whereas bega (22.6%) season rainfall had higher variability in the study area. Further, Eastern and southwestern parts of the study area had high variation during the kiremt season. Inter-annual rainfall variability was evident for the occurrence of dry and wet years. The Kiremt season positive rainfall anomalies of all stations during the study period accounted 60% over negative anomalies. The annual (9.14 mm/year) and belg (6.94 mm/year) season rainfall exhibited statistically significant increasing trends, while the kiremt (−1.88 mm/year) season had a non-significant decreasing trend at p < 0.05.

Such information will facilitate Ethiopian agricultural planning and development in general, and west Gojjam zone in particular, helps to reduce the adverse effects of climate variability, and promote full capitalization when abundant rains occur. The study area time of production is Kiremt. Hence, emphasis was placed on Kiremt onset, cessation date, length of the rainy season and dry-spell length.

Further research incorporating additional climate elements and long-time series data is recommended for enhanced precision and understanding of local weather patterns and their implications for water resources and ecosystem management.

Availability of data

The data used for analysis for the current study are available up on the corresponding author and National Meteorological Station Agency (NMA) on reasonable request.

Acknowledgments

The authors acknowledge the National Meteorology Agency of Ethiopia (NMA) for providing the study with long-term gauge-based daily rainfall data. The authors also acknowledge the use of the IMERG-06 dataset from Integrated Multi-satellitE Retrievals for GPM (https://gpm.nasa.gov/data/imerge). We are grateful to Bahir Dar University for providing the first author with study leave.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/22797254.2024.2372856.