1 Introduction

Global climate has been changing for thousands of years with adverse impacts on different sectors [1, 2] and its frequency has become short in the recent decades as compared to the past [3]. This is evident through rising temperatures, frequent drought occurrence, flooding, and spatiotemporal variability of rainfall, as highlighted by Taye et al. [4]. Although the effects of climate change and variability are global [5], the negative impact is strong in developing countries whose livelihood is nature-based and climate-sensitive [6]. Low adaptive capacity, being heavily dependent on natural rainfall, and lack of agricultural technologies among others made the agriculture sector more vulnerable to climate change and variability in developing countries such as Sub-Saharan Africa [7]. Located in Sub-Saharan Africa (SSA), Ethiopia is one of the most impacted countries by extreme climates such as recurrent drought and floods affecting the lives of thousands of people [8, 9]. Drought in Ethiopia is frequently accompanied by intra and inter-seasonal fluctuation in total and distribution of rainfall [5, 10] and season failure due to disruption of the start season [11, 12]. Agricultural production in Ethiopia is heavily prejudiced by the seasonality of rainfall in a given growing season, bearing substantial consequences [13,14,15]. This sector employs 80% of the country's workforce, contributes to 60% of the Gross Domestic Product (GDP), and accounts for 80% of foreign exchange [16]. However, despite its vital role in the economy, agriculture in Ethiopia remains a precarious venture due to its reliance on unpredictable rainfall patterns [6, 10]. Over the past few decades, Ethiopian agriculture has been consistently challenged by recurring droughts. One notable example is the drought of 1984–1985, which had a far-reaching impact, affecting approximately eight million people and resulting in the loss of one million lives [6, 17]. The recent El Niño-driven drought in 2015–2016 has left 10.2 million people experiencing food insecurity, with 1.8 million needing food aid and 1.2 million requiring support for livestock and crops [18]. Studies evidenced that climate change-induced droughts are occurring more frequently, posing a food security risk every three years [11, 19]. Ethiopia's seasonal climate and diverse landscape contribute to variations in rainfall distribution [5, 7]. The central highlands and southern regions have experienced rainfall decreases by 15% and 20% during the Belg and Kiremt, respectively [20].

The seasonal climate of Ethiopia is characterized by three different seasons locally called; Kiremt(June–September), the main rainy season in most parts, Belg (February—May) second rainy season, and Bega (October—January) dry season [15, 21, 22]. The Kiremt season accounts for 65–95% of the rainfall and 95% of crop production [23]. Ethiopian rainfall patterns, particularly Kiremt rainfall are influenced by factors such as the Inter Tropical Convergence Zone (ITCZ), low-level Somali Jet, Indian Ocean Dipole (IOD), Azores high, tropical easterly jet, and El Niño Southern Oscillation (ENSO) [5, 22]. Researchers have utilized the aforementioned rain-bearing systems to forecast rainfall seasonality, as evidenced by studies carried out by Eshetie [24] and Viste et al. [25]. Previous studies on climate in Ethiopia, including the Amhara Region, have provided valuable insights into climate variability, adaptation, and the impacts of climate change on ecosystems and economies. Notably, researchers such as Dendir and Birhanu [26] and Malede et al. [27] have made significant contributions to this knowledge base. These studies mainly focused on country, regional, and watershed scales [7, 28]. However, as highlighted in earlier studies [5, 29], large-scale climate studies often overlook ground-level realities, where local factors like topography influence climate patterns, especially rainfall. In response, Alemayehu & Bewket [10] and Alemu & Bawoke [28] suggest that micro-scale climate analysis is essential for capturing local variations, enabling the development of location-specific mitigation and adaptation strategies.

In addition to the aforementioned gap, most of the prior studies also focus on inter-annual and inter-seasonal average-oriented climate analysis [30, 31]. Nevertheless, seasonal and annual rainfall totals alone could not be helpful for farm-level decision-making, as outlined by several studies [32, 33], yet intra-seasonal rainfall variables are crucial for farm-level decision-making. This is because crop-specific rainfall amounts can be recorded within a few days accompanied by consecutive dry days after successful season onset. Consistent with this, Wamba et al. [33] recommend analysis of intra-seasonal rainfall and temperature variability and trends so that farmers will get risk-oriented agro-meteorological information that can help with farm-level decision-making. Previous studies [34,35,36] conducted in the Amhara Region, where the present study is situated, have investigated agriculturally significant climate variables such as the start and end dates and duration of the growing season. However, it is worth noting that none of the districts included in the present study were covered in these previous studies. Despite various challenges, including the absence of long-term climate data, there is a great need for micro-scale climate studies in drought-prone districts that are often marginalized. Hence, this study aimed to examine the spatiotemporal variability and trends of intra-seasonal to inter-seasonal rainfall, and temperature in the drought-prone districts of northwestern Ethiopia. To achieve the overarching goal of this study, specific objectives such as 1) Assess the spatiotemporal variability of intra to inter-seasonal rainfall and temperature, 2) examine the interlinkage of El Niño-Southern Oscillation (ENSO) episodes and Kiremt (June to September) rainfall across the study districts and 3) Analyze the trends in intra to inter-seasonal rainfall, length of the growing season, and temperature were pursued in the drought-prone districts located in northwestern Ethiopia. This study is significant as it is the first research conducted in the study area, highlighting its novelty and importance. It specifically addresses gaps in the existing literature by focusing on intra-seasonal analysis, filling a knowledge gap in the field of study and for the study districts. The relationship between Kiremt rainfall and El Niño-Southern Oscillation (ENSO) periods is also examined in this study, suggesting ENSO is the best indicator for Kiremt rainfall. In addition, it examines temperature, rainfall, and growing season trends—all of which are significant factors in farm-level decision-making. The study intends to capture local variations and facilitate the development of mitigation and adaptation measures that are relevant to a given place by offering micro-scale climate analysis.

2 Materials and methods

2.1 Study area description

The study area is located in Amhara National Regional State, northwestern Ethiopia. It comprises six drought-prone districts, as shown in Fig. 1. The total population is 1,172,888, with 593,069 males and 579,820 females (Obtained from district planning office). The study area covers an area of 9340.903 km2, with an absolute location of 11°2′28.364"N to 12°32′50.421"N and 37°35′58.569"E to 38°43′1.307"E. It shares administrative boundaries with Wagehemera and North Wello zones in the north, South Wello in the east, Lake Tana, Bahir Dar, and East Gojam in the south, and Central Gondar in the west. The study area exhibits varying topography, with elevations ranging from 1200 m in the lowlands of the northeastern and southeastern regions to 4106 m at Guna Mountain, situated between the Misrak Este and Lay Gayint districts (Fig. 2). As illustrated in Fig. 4, the study area exhibits a predominantly Kiremt rainfall pattern, characterized by a monomodal type. This finding is consistent with prior research conducted in the northwestern region of Ethiopia, where the present study area is located. Those previous studies also identified the dominance of Kiremt rainfall in all the stations investigated ([35]. As shown in Fig. 7, annual rainfall ranges from 806.4 mm at Arbgebiya to 1342.0 mm at Mekaneyesus meteorological stations. The study area experienced peak rainfall in July, with approximately 330 mm, closely followed by August with 325 mm. The monthly maximum temperature ranges from 16.8 °C in August to 27 °C in March followed by April. Conversely, the monthly minimum temperature ranges from 11.0 °C in December and January to 14 °C in May followed by April. These patterns are illustrated in Fig. 3. Major crops grown include Maize, Barely, wheat, Teff, bean, pea, chickpea, lentil, vetch, sorghum, and millet. According to [37], mixed farming systems are practiced by over 85% of the farm households in the study area, and traditional farming technology is employed for land cultivation by more than 93% of the farm households. As shown in Fig. 7 a, annual rainfall ranges from 806.4 mm at Arbgebiya to 1342.0 mm at Mekaneyesus meteorological stations. Major crops grown include Maize, Barely, wheat, Teff, bean, pea, chickpea, lentil, vetch, sorghum, and millet. As noted in [37], mixed farming systems are practiced by over 85% of the farm households in the study area, and traditional farming technology is employed for land cultivating by more than 93% of the farm households.

Fig. 1
figure 1

Location map of the study area

Full size image
Fig. 2
figure 2

Digital elevation of the study districts

Full size image
Fig. 3
figure 3

Monthly rainfall and temperature

Full size image

2.2 Data types and sources

This quantitative study utilized six meteorological stations in each district for temporal analysis and ten additional grid points for spatial analysis (Fig. 1 and Table 1). Both gauge-based and high-resolution blended (4 km X 4 km) daily rainfall and temperature data for the period of 1992 to 2021 were obtained from the Ethiopian Meteorology Institute (EMI). According to [38], Ethiopia's meteorological station density falls below the recommended minimum by the Organization of World Meteorology, with fewer or no stations in lowland areas, including the present study area. Moreover, existing records suffer from long-term recording and quality issues [39, 40]. To address these challenges, the EMI collaborated with the International Research Institute to establish the Enhanced National Climate Service (ENACTS), which provides improved access to climate data for decision-making processes [38]. ENACTS-Ethiopia has demonstrated a strong correlation with ground gauge data and has been utilized in previous studies to examine rainfall variability [10, 41,42,43]. Hence, the present study used the long-term gridded daily rainfall and temperature data from ENACTS-Ethiopia for spatial analysis and to fill the data gap (Table 1). Additionally, El Niño-Southern Oscillation (ENSO) data for 1992–2021 was downloaded from a National Oceanic and Atmospheric Administration (NOAA) website to analyze its implications on the spatio-temporal rainfall distribution over the study area.

Table 1 Meteorological stations and database period
Full size table

2.3 Data quality control

Data Before undertaking data analysis, a detailed examination was performed on the daily climate data with great care to identify and correct any instances of decimal point misplacements, as well as errors such as data entry mistakes, duplications, and missing data. This scrutiny was performed through visual scanning on a case-by-case basis, as highlighted by Hadgu et al. [44] and Mekasha et al. [45]. The nonparametric rank-based Pettit test in Climate Data Tool (CDT) version 7.0 is used for detecting shifts in the mean value of the distribution of the variable under study [39, 46]. Outlier detection is done using Turkey Fence [44, 47]. The zeroth and first-order Markov chain model in INSTAT version 3.6 was also used to fill that gauge-based daily rainfall data gap [32, 48, 49].

2.4 Defining rainfall thresholds

Various approaches have been employed to determine rainy and dry days, depending on the specific purpose. For example, in meteorological contexts, a threshold of 0.1 mm of rain per day has been utilized to classify days as wet or dry [50]. However, some argue that this threshold may not be suitable for defining rainy or dry days from an agricultural perspective, as it may not significantly impact crop growth [51]. Consequently, in the present study conducted by [52], a benchmark of 1.0 mm is adopted to classify a day as rainy if the rainfall is equal to or exceeds 1.0 mm, and if not, it is considered a dry day. Several previous studies used this definition to determine a rainy day and a day with no rain [53,54,55].

The start date of the Kiremt growing season in this study adhered to the criteria developed by Segele and Lamb [56], requiring at least 20 mm of rainfall over three consecutive days without a seven-day dry spell within the next thirty days, serving as an indicator of soil moisture [52]. The threshold considered the climatological context and the occurrence of dry spells within the next 30 days [32, 51], and has been widely employed in previous studies [57]. The end date of the growing season is determined by the depletion of stored soil water, signifying the point at which the soil water balance becomes zero [44]. Threshold-based proxy methods were used to establish the start and end dates, with specific criteria tailored to local conditions. For instance, in West Africa, the period between the initial successful onset date and a dry spell lasting less than 20 days defines the end of the season [58]. In this study, the threshold-based proxy method developed by Segele and Lamb [56] was used to identify the start and end dates of the Kiremt growing season, taking into account soil moisture availability. This approach has been widely utilized in previous studies, including defining the cessation date of the Kiremt season [59]. Additionally, the study calculated the length of the Kiremt growing season by subtracting the start date from the end date [44, 48].

2.5 Variability analysis

The variability of intra-seasonal climate was examined using the Coefficient of variation (CV%). Several earlier researchers used the same method to examine climate variability [28, 60]. It was calculated as:

$$CV\% = \left( {\frac{\delta }{\mu }} \right)*100$$

where σ is the standard deviation, and μ is the mean value of observations. Reddy (1983) cited in Gedefaw et al. [61] defined the coefficient of variation as follows: Less than 20% exhibits a low variable, 20–30 shows a moderate variable and > 30% indicates a high variable of rainfall.

On the other hand, standardized Rainfall anomaly (SRA) is applied to determine dry and wet years at different thresholds in the present study. SRA was computed as:

$$SRA=\frac{{P}_{t}-{P}_{m}}{\upsigma }$$

where SRA = standardized rainfall anomaly, \({P}_{t}\) = the annual rainfall in year t, \({P}_{m}\) = long-term means annual rainfall throughout the observation, and σ = standard deviation of annual rainfall throughout the observation. Based on the calculated indices, drought severity classes are given as extreme drought (Z < -1.65), severe drought (-1.28 > Z > -1.65), moderate drought (-0.84 > Z > -1.28), and no drought. The same method was used in previous studies [34, 62, 63]. The Precipitation Concertation Index (PCI) is used to examine the distribution of rainfall at seasonal and annual time scales over the study districts. As [64] specified, PCI is scaled as uniform distribution, moderate distribution, irregular distribution, and highly irregular distribution with PCI values of < 10, 11–15, 16–20, and > 20, respectively. The same method was used in previous studies [28, 65].

$$PCI_{Seasonal} = \frac{{\sum\limits_{i}^{4} {P_{i}^{2} } }}{{\left( {\sum\limits_{i}^{4} {P_{i} } } \right)^{2} }}\quad {\text{for seasonal PCI}}$$
$$PCI_{Annual} = \frac{{\sum\limits_{i}^{12} {P_{i}^{2} } }}{{\left( {\sum\limits_{i}^{12} {P_{i} } } \right)^{2} }} = {\text{for annual PCI}}$$

Spatial analysis in the study utilized Inverse Distance Weighted (IDW) interpolation, a robust method recommended by researchers for areas with complex topography and strong local climate control, such as rainfall [66]. The analysis was conducted using Geographic Information Systems (GIS), specifically ArcGIS version 10.4.

2.6 Trend analysis

2.6.1 Mann Kendall trend test

Trends in seasonal and annual rainfall totals, wet days, start dates, end dates and duration of the growing season, and maximum and minimum temperature were investigated using the Mann Kendall (MK) non-parametric trend test. Trends in seasonal and annual rainfall totals, wet days, start dates, end dates and duration of the growing season, and maximum and minimum temperature were investigated using the Mann Kendall (MK) non-parametric trend test [67, 68]. As described by prior researchers [60, 69], the MK test is less sensitive to outliers and suitable to test a trend in a time series without specifying whether the trend is linear or not. The Mann Kendall's test statistic was given as:

$$S = \sum\limits_{i = 1}^{N - 1} {\sum\limits_{j = i + 1}^{N} {{\text{sgn}} (\mathop X\nolimits_{j} } } - \mathop X\nolimits_{i} )$$

where S was the Mann-Kendal’s test statistics; xi and xj were the sequential data values of the time series in the years i and j (j > i) and N was the length of the time series. A positive S value indicates an increasing trend and a negative value indicates a decreasing trend in the data series. The sign function was computed as:

$${\text{sgn}} (\mathop x\nolimits_{j} - \mathop x\nolimits_{i} ) = \left\{ \begin{gathered} + 1if(\mathop x\nolimits_{j} - \mathop x\nolimits_{i} ) > 0 \hfill \\ 0if(\mathop x\nolimits_{j} - \mathop x\nolimits_{i} ) = 0 \hfill \\ - 1if(\mathop x\nolimits_{j} - \mathop x\nolimits_{i} ) < 0 \hfill \\ \end{gathered} \right\}$$

The variance of S, for the situation where there may be ties (that is, equal values) in the x values is given by:

$$Var(s) = \frac{1}{18}\left[ {N(N - 1} \right]\left[ {2N + 5} \right] - \sum\limits_{i = 1}^{m} {\mathop t\nolimits_{i} } (\mathop t\nolimits_{i} - 1)(\mathop {2t}\nolimits_{i} + 5)$$

where m is the number of tied groups in the data set and ti is the number of data points in the ith tied group. For n larger than 10, ZMK approximates the standard normal distribution [60] and is computed as follows:

$$\mathop Z\nolimits_{MK} = \left\{ \begin{gathered} \frac{S - 1}{{\sqrt {{\text{var}} (s)} }}ifS > 0 \hfill \\ 0ifS = 0 \hfill \\ \frac{S + 1}{{\sqrt {{\text{var}} (s)} }}ifS < 0 \hfill \\ \end{gathered} \right\}$$

The presence of a statistically significant trend was evaluated using the ZMK value. In a two-sided test, the null hypothesis (Ho) was accepted at ZMK < Z 1-α/2 at a given level of significance Z 1-α/2 was the critical value of ZMK from the standard normal table. The rate of change was computed using Sen.’s slope estimator, which could be used with missing data and remain unaffected by outliers or gross errors [65, 70]. Sen.’s slope estimator was computed as:

$$\mathop T\nolimits_{i} = \frac{{\mathop X\nolimits_{i} - \mathop X\nolimits_{k} }}{J - K}fori = 1,2,3...N$$

where Xj and Xk are considered as data values at time k (j > k) correspondingly. The median of these N values of Ti is represented as Sen.’s estimator of slope given by

$$\mathop Q\nolimits_{i} = \left\{ \begin{gathered} \mathop T\nolimits_{{\frac{T + 1}{2}}} .......N = odd \hfill \\ \frac{1}{2}\left\{ {\mathop T\nolimits_{\frac{N}{2}} + \mathop T\nolimits_{{\frac{N + 2}{2}}} } \right\}.......N = even \hfill \\ \end{gathered} \right\}$$

Sen.’s estimator was computed as Qmed = T (N + 1)/2 if N appears odd and it is considered as Qmed = (TN/2 + T (N + 2)/2 if N appears even. In the end, Qmed was computed by a two-sided test at 100 (1-α) % confidence interval and then a true slope was obtained by the non-parametric test. A positive value of Q indicates an upward or increasing trend and a negative value shows a decreasing trend and if the values are zero, it shows the data fluctuates around the mean. R studio in R software was used to carry out the computations.

2.6.2 Modified Mann–Kendall trend test (MMK)

Serial correlation is a common issue in time series data. In statistical testing, checking the presence of Serial Autocorrelation in the results of the Mann–Kendall test of the time series data is crucial to eliminate problems in detecting trends in hydro-climatological data [71]. As mentioned, in non-parametric testing, the characteristics of the distributed samples must be serially independent (no autocorrelation) and randomly distributed. Autocorrelation (or serial correlation) for time series data was used to assess the correlation between the values in time series data and historical values. The presence of serial autocorrelation in time series data can affect the ability of the non-parametric MK test to properly determine the significance of the trend [72]. Researchers have proposed various approaches to eliminate the influence of Serial Autocorrelation, such as pre-whitening (PW). The PW approach was applied for statistical tests (Mann–Kendall and Spearman's rho) to remove the influence of serial correlation [71]. However, Yue et al. [72] found that the PW approach can affect the statistical significance of the trends and slope magnitude present in the original hydro-climate data series. To overcome these problems, a trend-free pre-whitening (TFPW) approach was proposed by researchers [73]. This study applied the TFPW approach to assess the significance of trends in serially correlated data. To check for serial correlation in the data, the study first found the serial autocorrelation coefficient (r_1) value using a two-tailed test at a 5% significance level for lag-1 as follows:

$${r}_{1}= \frac{\frac{1}{n-1} \sum_{t=1}^{n-1}\left({Y}_{t}-\left(\frac{1}{n}\sum_{t=1}^{n}{Y}_{t}\right)\right)\left({Y}_{t+1}-\left(\frac{1}{n}\sum_{t=1}^{n}{Y}_{t}\right)\right)}{\frac{1}{n}\sum_{t=1}^{n}\left({Y}_{t}-\left(\frac{1}{n}\sum_{t=1}^{n}{Y}_{t}\right)\right)}$$

where \({Y}_{t}\) is the series value at time t and n is the number of years in time series data. The null hypothesis of an independent series at a 95% confidence interval for lag-1 autocorrelation coefficient values computed by:

\({r}_{1}=\frac{-1 \pm 1.96 \sqrt{n-2}}{n-1}\) For the two-tailed test.

To decide whether the time series data is serially correlated or does not show serial correlation, we find the value of \({r}_{1}\) in the range of upper and lower boundaries of confidence interval. If the value of the serial autocorrelation coefficient found between the intervals (\(\frac{-1-1.96\sqrt{\left(n-2\right)}}{n-1}\le {r}_{1}\le \frac{-1+1.96\sqrt{\left(n-2\right)}}{n-1}\)), it can be concluded that the data is not serially dependent or doesn’t exhibit the presence of serial autocorrelation in time series. If the time series data confirmed no significant autocorrelation (non-serially correlated), the trend analysis can be carried out without the need for pre-whitening (PW) or trend-free pre-whitening (TFPW) approaches [71]. Otherwise, the researcher applied the TFPW approach to evaluate the serially correlated data [74]. The presence and absence of autocorrelation in the time series data were computed using open-source library packages, such as the "modified" package developed in R-language. For non-correlated data series, the normal Mann–Kendall (MK) test was recommended without any modification. For the case of correlated data series, the Modified Mann–Kendall (MMK) test was applied [75, 76]. The same approach was applied in previous studies [71, 76].

3 Results and iscussion

3.1 Annual patterns of rainfall and wet days

The data presented in Fig. 4, illustrates the distribution of annual rainfall and number of wet days per month in all six study stations. The results confirmed that all stations in the study area experience a mono-modal rainfall pattern extending from June to September locally known as the Kiremt season. July followed by August exhibits the highest rainfall amount for all stations. This finding aligns with a previous study conducted in the South Gondar zone which reported that 85% of the seasonal rainfall occurs in July and August [35]. Other studies confirmed that most parts of central and northwestern Ethiopia in which the present study is situated experience mono-modal rainfall patterns [22, 28]. The result also confirmed that there is a chance of getting rainfall during the Belg and Bega seasons in the study area, albeit with a lower amount of rainfall. Per the previous study [79], the availability of rainfall during the Belg and Bega seasons is beneficial for land preparation ahead of the Kiremt growing season. The moisture further can support the planting of long-cycle crops, provide pasture for livestock, and enhance water availability. However, despite its importance, the Belg and Bega rainfall needs close monitoring as it could pose a risk of false onset and damage to crops in the field [48]. From a different perspective, relying solely on total rainfall is not sufficient for farm-level weather-based decisions, especially for rain-fed agricultural practices [44, 72]. This is because a required rainfall amount can be concentrated within a few days. The identical pattern of wet days and rainfall, exhibited in Fig. 4 confirms uniform temporal distribution across all stations, regardless of its amount and duration. The result suggests that wet days can be taken as one of the best indicators of rainfall distribution for the study area.

Fig. 4
figure 4

Patterns of monthly rainfall and number of wet days

Full size image

3.2 Temporal variability and trends of monthly and seasonal wet days

Statistical summary and trends of monthly wet days are presented in Table 2. The analysis indicates a non-significant negative trend (− 0.149 °C/year) in wet days during June, suggesting a slight decrease. The mean number of wet days in June is 14, and the high coefficient of variation (CV%) of 42 indicates significant variability. In July, there is a significant negative trend (− 0.373 °C/year) with a mean of 25 wet days and a low CV% of 17, indicating a consistent decline with relatively stable patterns. August shows a non-significant positive trend (0.141 °C/year) with a mean of 25 wet days and a low CV% of 16, suggesting a slight increase and stable patterns. In September, there is a significant negative trend (− 0.317 °C/year) with a mean of 13 wet days and a moderate CV% of 36, indicating a consistent decline with moderate variability. These findings provide insights into changing rainfall patterns, informing decision-making in agriculture and water resource management. The data in Table 2 further displays the average number of wet days during the Kiremt season, ranging from 60 in Mekaneyesus to 82 in Simada. Simada shows relatively consistent rainfall distribution compared to other stations. All study stations have low CV values of 7–10%, indicating reliable wet days during the Kiremt season. The analysis reveals significant downward trends in Ebenat (0.359 °C/year) and Simada (0.494 °C/year), indicating a noticeable decrease in wet days. Conversely, Mekaneyesus shows a statistically significant upward trend (0.344 °C/year). The Belg season exhibits varying wet days, with Adiszemen at 12 and Nefasmewcha at 27, and high CV values of 32% to 64%, indicating high variability. Water availability for agriculture is affected due to erratic rainfall patterns. Mekaneyesus shows a statistically significant upward trend (0.277/year) during the Belg season. In the Bega season, wet days vary across the study area with high CV values of 37% to 64%. Ebenat and Simada experience a significant decrease with Sen.'s slope values of 0.282 °C/year and 0.301 °C/year. The high CV values and significant trends suggest unreliable rainfall for annual and seasonal planning. The findings emphasize the importance of adaptive strategies and decision-making that consider changing rainfall patterns as a result of wet day variability across the study area. This information informs agricultural practices, water resource management, and other rainfall-dependent sectors in the study area.

Table 2 Summary statistics and MK trend analysis of monthly and seasonal wet days (1992–2021)
Full size table

3.3 Spatial variability of start date, end date, and length of Kiremt growing season

Figure 5a, b illustrated the spatial distribution of the start date and cessation dates of the Kiremt growing season in days of the year (DOY), while Fig. 5c presented the length of the growing season in days. The results in Fig. 5a reveal that the growing season starts earlier, around the second week of June (160 s DOY) in the western parts of the study area. This early commencement is particularly noticeable in and surrounding areas of Adiszemen and Mekaneyesus. Conversely, in the remaining areas, the onset of the growing season is delayed, typically taking place around the first week of July (180 s DOY) in stations like Ebenat, various parts of Nefasmewcha, and Arbgebiya, which are situated in the northeastern, eastern, and southeastern regions respectively. The observed early start in the west and a late start in the eastern parts of the study area are supported by the conclusion made in prior studies [81, 82]. These results suggest that farmers in the regions with an early start can offer valuable insights into identifying crop varieties that are well-suited for their specific area. Farmers in the late start region ( e.g. Ebenat) can benefit from this knowledge by selecting appropriate crop varieties that can mature within their shorter growing season.

Fig. 5
figure 5

Spatial distribution of a start dates, b end dates, and c length of Kiremt growing season

Full size image

On the flip side, the results in Fig. 5b showed that the Kiremt growing season withdraws earlier from the stations, such as Ebenat situated in the northeastern side of the study area by the third week of September to the first week of October (270 s–280 s DOY). The illustration evidenced that cessation time becomes late as one moves towards the stations on the western side. For instance, in Nefasmewcha, the growing season recedes around the second week of October (280 s DOY) while in Adiszemen, Arbgebiya, and the northern portion of Mekaneyesus, it ends around the second week of October (290 s DOY). As indicated by [83], the Inter-Tropical Convergence Zone (ITCZ), a Kiremt rain-bearing system, arrives early and departs late in the southern parts of Ethiopia. This results in an early onset and late cessation of the rainy season in the western and southwestern regions of the present study area. Conversely, the northeastern parts experience a delayed onset and early departure of the ITCZ, leading to a late onset and early recede of the season in that area. Figure 5c illustrates that the delayed start and early cessation of the Kiremt growing season have led to a reduced duration in the northern and northeastern locations. Notably, Ebenat has experienced a shortened duration of only 80 days. To mitigate this issue, it is advisable to prioritize the cultivation of fast-maturing crop varieties in the affected regions. By doing so, farmers can optimize their yields within the limited timeframe of the Kiremt season, thereby enhancing agricultural productivity and bolstering resilience against evolving climatic circumstances.

3.4 Temporal variability and trends in start and end date and length of Kiremt season

The box and whisker plots in Fig. 6a–c present the chance of occurrence of the start date, cessation date, and duration of the growing season, respectively for all study stations. Based on minimum values of start date (10th tile), the Kiremt season typically begins any time before the first week of June (153DOY) in most stations, except Ebenat and Arbgebiya, experiencing late starting time, around the first week of July (183DOY). The 25th, 50th (median), and 75th tiles data revealed that the growing season consistently begins between the first week of June (153DOY) and the second week of July (193 DOY) over the study districts, starting early in Adisemen and late in Arbgebiya, as shown in Fig. 6a. These results are corroborated by previous studies conducted in Nefasmewcha, which further approve that the season typically starts in early July [35]. The result displayed in Table 3 reveals low variability (CV values of 3 to 9%) in the start date of the Kiremt season, indicating consistent patterns across all stations. This finding is substantiated by the research conducted by Getachew [84] who documented similar low variability in the start and end dates of the Kiremt season in Adiszemen and nearby districts. An increasing trend with the Sen's slope values of 0.002 to 0.453 confirmed that most stations experienced late starting over time. Moreover, Adiszemen exhibited a statistically significant upward Sen's slope (0.453/year) at P < 0.05, demonstrating a notable late starting over time. The observed late onset invites attention to address the potential risk of forward shifting over time. Prior studies indicate that due to the unpredictability of the onset season, rain-fed crop production usually be in danger [44].

Fig. 6
figure 6

Box plots for a start date, b end date, and c length of Kiremt growing season (Solid circle (medians), boxes (25-75th tiles) and whiskers (10th/90th tiles))

Full size image
Table 3 Summary statistics of Start and End dates and length of growing season (1992–2021)
Full size table

On the flip side, the minimum values (10th tile) of the cessation date in Fig. 6b indicate that the Kiremt growing season typically recedes any time after the first week of October (265–283 DOY) from the study area. The results indicated by the 25th, 50th, and 75th tiles reaffirmed that the growing season consistently recedes between the first week of October and the third week of October (DOY 276–300), demonstrating an earlier withdrawal from Ebenat, Nefasmewcha, and Arbgebiya compared to other stations. Notably, in Simada, the season extends until the first week of November (DOY 313), which is an exceptional case. These findings align with a previous study that has shown prolonged cessation of the growing season in Nefasmwecha [35]. The positively skewed distribution in Fig. 6b and the decreasing trends in Table 3 show a tendency for early termination of the Kiremt growing season in Ebenat, Arbgebiya, and Simada. Moreover, Fig. 6c illustrates occurrence probability and patterns of the length of the Kiremt growing season for all the stations. The median values (50th tile) of the season duration suggest planting crop varieties to plant crop varieties with maturing periods of 96, 148, 139, 114, 131, and 124 days at Ebenat, Adiszemen, Nefasmewcha, Arbgebiya, Simada, and Mekaneyesus, respectively. Consistent with the present results, the study carried out by Getache & Teshome (2017) confirmed that the Kiremt season typically lasts between 95 to 140 days in Nefasmewcha. The coefficient of variation (CV) values in Table 3, indicated higher variability compared to the start and end dates, ranging from 11 to 18%. This emphasizes the importance of exercising caution when selecting crop varieties before making decisions. Furthermore, the existence of negative values in Table 3 and the right-skewed distribution depicted in Fig. 6c specifies a tendency towards shorter durations of the growing season across the majority of stations.

3.5 Spatial variability of monthly, seasonal, and annual rainfall

Spatial variability of monthly, seasonal, and annual rainfall is exhibited in Fig. 7. Based on the results, the western parts of the study area used to experience a relatively higher amount of rainfall during the June to September months which is supported by the seasonal Kiremt rainfall pattern shown in Fig. 7f. The study area received a peak rainfall amount during July (237.4 mm to 423.0 mm) followed by August (235.0 mm to 371.5 mm), as depicted in Fig. 7a, b. Consistent with this the annual cycle results presented in Fig. 4 evidenced that the study area exhibits peak rainfall in July and August suggesting a monomodal pattern. The findings confirmed that rainfall amount decreases as one moves from August to September, demonstrating the ending of the growing season over the study area. This spatial variability highlights the importance of considering localized rainfall distribution for agricultural planning and water management strategies. The results in Fig. 7h demonstrate a noticeable increase in annual rainfall as one moves towards the southwest and northwest parts of the study area. For instance, stations such as Adiszemen, Mekaneyesus, and Nefasmewcha, located in southwestern and northwestern parts of the study area receive higher amounts of rainfall ranging from 1220 to 1320 mm. This result aligns with preceding studies that conformed that, Nefasmewcha, Mekaneyesus, and Adiszemen receive annual rainfall, ranging from 1052 to 1347 mm [35]. The southeastern, northeastern, and eastern sides, on the other hand, exhibit relatively lower annual rainfall, extending between 820 and 920 mm. In line with this, prior analysis confirmed northeastern and southeastern parts of the Amhara region receive lower rainfall, as compared to the northwestern, western, and southwestern parts [28]. The result further indicates that the western locations of the study area receive a higher amount of rainfall, ranging from 900 to 1100 mm, compared to the north-eastern and south-eastern parts, which obtain about 600 mm to 800 mm during the Kiremt season. This result shows that annual rainfall is a function of Kiremt rainfall across all the study stations. On the flip side, the study area receives a relatively small amount of rainfall Belg and Bega seasons. These findings indicate that Belg and Bega rainfall, despite their importance, can pose risks such as false onset for the Kiremt season and late rains during harvesting time for Meher crops [32, 44].

Fig. 7
figure 7

Spatial distribution of monthly, seasonal, and annual rainfall (1992–2021

Full size image

3.6 Temporal variability and trends of monthly and seasonal rainfall

According to the data provided in Table 4, the analysis reveals that the rainfall exhibits high variability with coefficient of variation (CV) values ranging from 35 to 77% from June to September across all study stations. On the other hand, the months of July and August show relatively lower to moderately variable rainfall across most stations. Notably, stations like Simada and Mekaneyesus experience low variability in rainfall with a CV value of 19% during August and at Mekaneyesus in July. When examining the trends in rainfall, Sen's slope values indicate mixed patterns that are not statistically significant in all months from June to September, except for a significant increase of 0.228 mm/year observed at the Simada station during August (at P < 0.05). This information suggests that the rainfall in the study area during the specified months is characterized by high variability, particularly from June to September. However, the trends in rainfall show no consistent and significant patterns at most stations over the study period (1992–2021). As can be observed in the table, most of the stations exhibited low to moderate variability in annual rainfall, excluding Adiszemen which had a high coefficient of variation (CV) value of 32% (Getachew). Nefasmewcha and Mekaneyesus, on the other hand, experience low variability [84]. The majority of study stations demonstrate an irregular concentration of annual rainfall, with Precipitation Concentration Index (PCI) values ranging between 18 to 20 [84]. Ebenat and Adiszemen experience higher irregularities in annual rainfall, highlighting their susceptibility to erratic patterns [76, 77]. Trend analysis shows increasing signals in annual rainfall for Ebenat, Adiszemen, and Mekaneyesus, with Adiszemen exhibiting a statistically significant positive trend [84].

Table 4 Summary statistics and trends of monthly, seasonal, and annual rainfall (1992–2021)
Full size table

Other stations show non-significant decreasing trends [78, 79]. Kiremt rainfall is the major contributor to annual rainfall, ranging from 55 to 90% [77, 80]. Adiszemen and Ebenat exhibit high variability in Kiremt rainfall, with CV values of 40% and 30%, respectively [84]. Adiszemen, with a high CV and irregular concentration (PCI value of 17), is prone to flooding [89]. The results revealed that farmers in the study area should carefully manage and monitor Belg and Bega rainfall, as highlighted by Ayalew [85]. These rainfall patterns, which contribute approximately 6% to 19% of the annual rainfall, require attention for both benefit and risk management. [85] Suggests that while Belg rainfall can serve as signals for farmland preparation for the Kiremt growing season, it also has the potential to pose a risk of false onset information. Furthermore, Ayalew [85] and Hadgu et al. [44] emphasize that rainfall during the Bega season, despite its importance for water and pasture availability, has the potential to pose a risk of late rain that can adversely affect crops in the field. Hence, farmers should use this information to seize opportunities and implement strategies to mitigate the associated risks. Furthermore, considering the higher coefficient of variation (CV) values for Belg (ranging from 40 to 102%) and Bega (ranging from 47 to 87%) rainfall, it is crucial to acknowledge their unpredictable nature and unreliability when making weather-related decisions across all stations.

3.7 Standardized rainfall anomaly vs. ENSO events: for Kiremt growing season (june to september)

The study demonstrates seasonal rainfall anomalies during the Kiremt season at six stations, as depicted in Fig. 8. Positive anomalies were observed in 43% of the years at Arbgebiya and 50 of the study years at Ebenat, Adiszemen, Nefasmewcha, and Simada. On the contrary, negative anomalies were exhibited at stations such as Adiszemen, Arbgebiya, and Simada for 33% and 43% of the years, respectively. The results highlight a prevailing trend of deficit rainfall in most of the study stations for the last three decades (1992–2021). This adverse situation can have significant implications for water availability, agriculture, and other related sectors [90]. The study findings revealed a correlation between rainfall patterns and the El Niño Southern Oscillation (ENSO) phenomenon. Specifically, below-average rainfall was found coinciding with above-average sea surface temperature in the equatorial Pacific (EliNino), while above-average rainfall aligned with below-average sea surface temperature in the equatorial Pacific (LaNina) throughout the study period (1992–2021), as shown in Fig. 8. Anomaly-based correlation analysis of ENSO episodes and Kiremt rainfall reported the same findings over Ethiopia [90]. The identical characteristics of Kiremt rainfall and ENSO events suggest a potential influence of ENSO on rainfall patterns in the study area. prior studies [22, 23] also demonstrated the sensitivity of Kiremt rainfall to the events of ENSO [25, 91]. Kiremt season (June to Sept) rainfall decreases (increases) during El Nino (La Nina) ENSO phases [25, 90]. As can be noted in the figure, extreme drought was observed in all study areas in 2009 and 2015. Similarly, severe drought events were noted in the years 1993, 1997, 2004, and 2014 across most stations, with moderate drought events occurring in 1995, 2002, 2004, and 2014. Consistent with previous studies, the years characterized by drought were associated with the El Niño phenomenon[11, 22]. The result confirmed that the study area has been suffering from drought with varying severity indices. The analysis further confirmed that El Niño Southern Oscillation (ENSO) events are reliable indicators for Kiremt rainfall with 80% correlation for the study area. Consistent with this, prior research highlighted that ENSO events serve as reliable indicators for Kiremt rainfall in Ethiopia [52, 92].

Fig. 8
figure 8

Standardized rainfall anomaly against ENSO: Kiremt growing season

Full size image

3.8 Spatial distribution of maximum and minimum temperature

The spatial distribution of maximum temperature in Kiremt, Belg, and Bega seasons for all the six weather stations is presented in Fig. 9a–c. The findings reveal that Ebenat and Libokemkem (Adiszemen) located in the northeastern and northwestern part of the study area experienced higher maximum temperatures. Stations, such as Nefamewcha, Mekaneyesus, Arbgebiya and Simada situated in and near the central highlands exhibit relatively low maximum temperatures. This could be attributed to the influence of Guna Mountain which extends an elevation of 4106 m above sea level, as illustrated in Fig. 2. The result aligns with the prior study reported that higher maximum temperature during the Belg season as compared to the Kiremt and Bega seasons [93]. The monthly distribution of maximum temperatures depicted in Fig. 3 reveals that months like March, April, and May, which fall within the Belg season, exhibit higher temperatures. This observation corroborates with previous studies conducted by Shekuru et al. [94] and Mengistu [95], which also reported that higher maximum temperatures are typically experienced during March, April, and May. The results in Fig. 10a–c show that higher minimum temperatures have been experienced over Ebenat, Libokemekem (Adiszemen), and the northern portion of Laygayint (Nefasmewcha), compared to other parts of the study area. Conversely, a relatively lower minimum temperature is observed over the central to southern parts of the study area. As compared to the Kiremt and Belg seasons, the Bega season exhibits the lowest minimum temperature in the ranges of 6 to 9 °C, as shown in Fig. 10c. This might be due to the intensification of dry and cold high-pressure areas over the northern and central highlands of Ethiopia during the Bega season [22, 88]. Stations in the northern parts of the study area, like Adiszemen and Ebenat, have higher temperatures in both maximum and minimum compared to other stations. This information can be useful for farmers and practitioners to adapt their annual and seasonal weather-based plans, such as irrigation scheduling and crop choices, in response to changing temperature patterns.

Fig. 9
figure 9

Spatial distribution of Maximum Temperature (Tmax in °c) during a Kiremt(JJAS), b Belg (FMAM), and c Bega (ONDJ) seasons

Full size image
Fig. 10
figure 10

Spatial distribution of minimum temperature (Tmin in °c) during a Kiremt (JJAS), b Belg (FMAM), and c Bega (ONDJ) seasons

Full size image

3.9 Temporal variability and trends of maximum and minimum temperature

The results presented in Table 5 indicated that the maximum and minimum temperature exhibits a consistent distribution across all seasons in and around the study stations, as evidenced by the low CV values ranging from 3 to 19% across all the seasons over the study area. Consistent with this, previous studies [84] conducted in northwestern Ethiopia show low variability in temperature. The low variability implies a more predictable pattern in both maximum and minimum temperatures, which can be beneficial in various contexts such as agriculture. It is worth noting that the Bega season noticed relatively higher CV values for minimum temperatures for all seasons, as shown in Table 5.

Table 5 Variability and trends in seasonal temperature (1992–2021)
Full size table

The Sens slope values displayed in Table 5 indicate that, in general, the study area exhibits consistently rising trends in both maximum and minimum temperatures in all seasons. As can be noted in the table, all study stations experience increasing changes in both maximum and minimum temperatures in all seasons. Based on the analysis, stations such as Arbgebiya and Simada exhibit statistically significant increases in both maximum and minimum temperatures at a 5% significance level. This indicates a warming trend in the region over the study period. The Sens slope values show that the maximum temperature significantly increased in the range of 0.014 °C/year at Ebenat to 0.421 °C/year at Arbgebiya across the study area (Table 5). Likewise, the minimum temperature showed an increment in the range of 0.027 °C/year at Ebenat to 0.485 °C/year at Simada over the study period. The findings confirmed that the study area exhibits increasing temperatures during the Kiremt (June to September) and Belg (February to May) seasons as compared to the Belg and Bega (October to January) seasons. As depicted in Fig. 3, relatively warm months are found in these seasons. In line with this, previous studies also outlined that the Belg and Kiremt seasons are warmer seasons [22, 66, 97]. On the other hand, the study area experiences relatively low temperatures with non-significant changes during the Bega season [8, 66]. This might be attributed to the influence of dry and cold high pressures from Siberia and the Azores, which enhances the northern highlands of Ethiopia, including the present study area [6, 23].

Overall, the study period unveiled a consistent rising trend in both maximum and minimum temperatures across most stations during the three distinct seasons. Previous studies conducted in the study area have also documented significant increasing trends in both maximum and minimum temperatures [55, 84, 98]. The observed significant increasing signals in all seasons over most stations evidenced that the study area has been suffering from rising temperatures. The increasing trends in temperature can have different impacts on various activities. Higher temperatures during drier seasons can increase water demand, cause crop stress, and create favorable conditions for pests and diseases [100, 101]. In addition, consistent rising temperatures during Belg and Kiremt seasons can adversely affect water and pasture availability for livestock, as discussed in previous studies [77, 102]. Higher temperatures can lead to increased water demand and crop stress, reducing agricultural productivity. Rising temperatures can also negatively impact livestock health and water/pasture availability, threatening the livelihoods of farmers and pastoralists. Adapting through measures like water-efficient irrigation, heat-tolerant crops, and diversified livelihoods will be crucial to building resilience in the face of these temperature changes.

4 Conclusions and recommendations

This study examined the spatial and temporal variability of rainfall and temperature in the drought-prone districts of northwestern Ethiopia, and the implications for rain-fed agriculture in the region. The results showed a dominant monomodal rainfall pattern, with the Keremt rainy season typically lasting from June to September. However, the Belg and Bega seasons were characterized by small and highly variable rainfall, suggesting unreliable precipitation that poses significant risks for agriculture. Some stations like Ebenat and Arbgebiya have a shorter growing season of around 3 to 4 months compared to 4 to 5 months at other stations. The region also suffers from recurrent El Niño-driven droughts, as the study districts are heavily impacted by the El Niño-Southern Oscillation (ENSO) climate phenomenon. Understanding and monitoring ENSO is crucial, as El Niño events can lead to severe reductions in major growing seasons (Kiremt, June to September) rainfall, threatening crop yields and livestock production. The study identified risks from the timing of Belg (February-May) and Bega (October-January) rains, which can negatively impact on-farm activities. Early Belg rains can mislead farmers about the start of the growing season, while late Bega rains can damage crops. Close monitoring is needed across the study area to manage these weather-related challenges. Furthermore, the study found significant upward trends in both maximum and minimum temperatures across all seasons, indicative of ongoing climate change.

To address these climate-related challenges, the study recommends several key adaptations. For rainfall variability, it suggests developing early warning systems linked to ENSO monitoring and promoting drought-resistant crops and water harvesting practices to cope. For rising temperatures, it recommends finding heat-tolerant crop and livestock varieties, improving water management and conservation, and adopting agroforestry. Achieving these adaptations will require a collaborative approach, with policymakers developing climate policies and providing support, farmers adopting climate-smart practices, and researchers working closely with stakeholders to generate new knowledge and technologies. This coordinated effort can help build a more resilient and sustainable agricultural sector in Ethiopia in the face of a changing climate influenced by phenomena like ENSO. The study recommends further research on climate vulnerability and climate adaptive strategies and related challenges so that local-specific effective intervention can be practiced.