A comparative study of drought characteristics using meteorological drought indices over the central main Ethiopian Rift

Drought is a naturally occurring disaster that has far-reaching implications for the socioeconomic, environmental, and agricultural realms. Unlike other natural disasters, it builds up gradually and has an undetermined beginning and end (Dracup et al. 1980). Drought is one of the effects of recent climate-related extremes, according to the Intergovernmental Panel on Climate Change (IPCC), posing a considerable threat to human systems and ecosystems (IPCC 2014).

Droughts of extreme intensity or extended duration could both have significant implications, such as infrastructure collapse and environmental vulnerability, and they are expected to occur more frequently and have more severe socioeconomic consequences with global warming (Gu et al. 2020).

A drought index, according to Mishra & Singh (2010), is an important component in assessing the intensity, severity, duration, and impacts of drought. Moreover, Wilhite et al. (2000) also stated that drought indices are designed for certain situations and have unique characteristics. To control drought conditions, a number of drought indices have been established.

Khalili et al. (2011) investigated similarities and differences by comparing the Standardized Precipitation Index (SPI) with the Reconnaissance Drought Index (RDI). Using the ETo (Reference Evapotranspiration), they discovered that RDI is highly susceptible to climatic fluctuations. This is crucial because employing the RDI appears to provide a better purpose if the drought analysis was for agricultural uses.

Ren et al. (2022) evaluated the correlation between SPI and RDI on a monthly, seasonal, and annual scale in the Yellow River basin of the source region, China. They discovered that SPI and RDI have great consistency, demonstrating both their resilience and their usefulness. Also, Memon & Shah (2019) evaluated and compared the SPI and RDI drought indices in Gujarat, India's Panchmahals district. They showed a strong link between the indices in various time scales.

Salimi et al. (2021) compared the SPI and Standardized Precipitation Evapotranspiration Index (SPEI) in the Lighvan, Navroud, and Seqez basins, Iran. They concluded that the SPEI is more appropriate than the SPI for applications examining drought variability in study areas because it uses precipitation and evapotranspiration data. Bazrafshan et al. (2019) used drought indices such as SPI, RDI, and SPEI to analyze the trend of drought intensity, duration, and frequency in Iran's arid and semi-arid regions. Their investigation demonstrated that the SPEI index responded to changes in arid-warm climates faster than the other indices. However, in cold regions, SPI and RDI variations showed similar values.

Meteorological drought is defined as a period of no precipitation or lower-than-average rainfall that lasts long enough to trigger hydrological and agricultural hazards (Łabędzki & Bak 2014). Agricultural drought is typically characterized as a soil water deficit at a specific time period resulting in a considerable drop in crop yield (Łabędzki & Bak 2014). The most severe Ethiopian droughts were caused by a lack of major rainfall (Kiremt, June–September) as well as short rains (Belg, February–May) (Suryabhagavan 2017). The Central Main Ethiopian Rift (CMER), on which this study focuses, is largely an agricultural area or cropland (Desta et al. 2017) that relies mostly on a rain-fed agricultural system. Rainfall deficits during the crop's growing season have a significant impact on Ethiopia's agricultural drought.

There has been a lack of comparative evolution of multiple drought indices across the study region. For example, Bisrat & Berhanu (2019) and Yirga (2021) used the SPI to analyze drought characteristics. The SPI, on the other hand, has been established to monitor the drought events (McKee et al. 1993), but it is derived only using precipitation data. Furthermore, Mohammed & Yimam (2021) conducted spatiotemporal drought assessments across the study region using the RDI (based on precipitation and temperature data). However, the use of other known drought indices was not applied in this study.

The El Nino/La Nina–Southern Oscillation (ENSO) has a significant impact on global climate patterns. This naturally occurring phenomenon involves changing ocean temperatures in the central and eastern equatorial Pacific, coupled with changes in the atmosphere (WMO 2014). Based on ENSO data, it may be necessary to monitor drought. However, the association between extreme drought events and large-scale modes of climate variables (e.g., ENSO episodes) has received minimal attention in prior studies by associating with drought indices, which the current study tries to investigate.

Bayissa et al. (2018) and Salimi et al. (2021) noted that multiple drought indices provide a more comprehensive understanding of drought characteristics. The purpose of this study is to compare drought characteristics based on the three drought indices, namely SPI, RDI, and SPEI, across the CMER during the agricultural growing season from 1980 to 2017. In terms of scholarly contributions, this work adds to the body of knowledge on the subject by giving multiple drought indices for monitoring drought in the study region, where there has been little research. As a result, we anticipate that our research will give useful references and ideas for the future comparison and selection of drought indices.

The time series data from 1980 to 2017 show that the mean annual temperature in the Highlands is around 15 °C, whereas on the rift floor, it is around 20 °C. The highlands receive 1,150 mm of yearly rainfall, whereas the rift floor gets around 650 mm.

The study region's climate is divided into three seasons: the main rainy season, which runs from June to September (summer, locally known as Kiremt); the shortest rainy season, which runs from February to May (spring, locally known as Belg); and the driest season, which runs from October to January (winter, locally known as Bega) (MoWR 2008).

Figure 1 also depicts the topographic features for the research area created with the SRTM (Shuttle Radar Topography Mission) DEM (Digital Elevation Model) (a 30 m resolution) available from: https://www.usgs.gov/). According to this figure, CMER has a variety of topography, ranging from 1,546 to 4,188 meters above sea level (m a.s.l.). Across the research region, there is a considerable disparity between the highest and lowest points (up to about 2,642 m).

For this study, daily precipitation, maximum temperature, and minimum temperature data for six stations (Kulumsa, Butajira, Sagure, Ziway, Wulbareg, and Hawassa) were obtained from the Ethiopian Meteorological Institute (EMI) from 1980 to 2017. Stations were chosen based on their geographical position, data accessibility, and statistical longevity.

McKee et al. (1993) established the SPI to define and monitor drought. The probability density function ‘Gamma’ is also used to fit the long-term time series of rainfall data recorded at each station before the fitted cumulative function is transformed into a normal distribution to calculate the SPI.

⁠, the maximum likelihood estimation of ⁠, ⁠, where and n is the number of observations.

RDI, which is based on precipitation and potential evapotranspiration (PET), was designed to more accurately approach the water deficit as a form of balance between input and output in a water system (Tsakiris & Vangelis 2005). In this study, the Thornthwaite method was used to estimate PET.

Three terms are included in the RDI expression: (the initial value), RDI_n (the normalized RDI), and RDI_st (the standardized RDI).

SPEI as an improved drought index of SPI was first proposed by Vicente-Serrano et al. (2010). Based on the principle of water balance, the SPEI uses the difference between precipitation (P) and PET as the input condition to evaluate the dry and wet conditions of the area. Vicente-Serrano et al. (2010) provided a complete description and calculation details of SPEI theory.

In this work, we studied the time series of drought indices (SPI, RDI, and SPEI) in the research area's agricultural growing months or rainfall seasons, namely spring (February–May), summer (June–September), and six months (March–August).

In addition to precipitation, PET values are required for RDI and SPEI indices. The Food and Agricultural Organization (FAO) Penman–Monteith approach is the most widely used method for estimating PET in various parts of the world (Allan et al. 1998). However, in order to predict the potential crop evapotranspiration, this method requires many meteorological parameters. We failed to apply the FAO Penman–Monteith (PM) approach due to a lack of meteorological data in the study area; however, other methods that only required a little amount of weather data could be applied. As a result, the present study used the method of Thornthwaite (1948), which calculates PET using the mean monthly temperature and latitude of the sites. Adem et al. (2017) verified the efficiency of the Thornthwaite approach over the Ethiopian highlands, encompassing the study area, in comparison to the FAO–PM method.

In this study, a threshold level of −1 (negative one) was utilized to identify moderate, severe, as well as extreme drought events in accordance with the computed values of SPI, RDI, and SPEI. This was done in accordance with McKee et al. (1993) drought classes, which are shown in Table 1.

Table 1

DS category classification (Mckee et al. 1993; Tsakiris et al. 2007)

SPI, ⁠, or SPEI values .	Category of drought .
Above 0	No drought
0.0 to −0.99	Near normal
−1.0 to −1.49	Moderate drought
−1.5 to −1.99	Severe drought
−2.0 and less	Extreme drought

SPI, ⁠, or SPEI values .	Category of drought .
Above 0	No drought
0.0 to −0.99	Near normal
−1.0 to −1.49	Moderate drought
−1.5 to −1.99	Severe drought
−2.0 and less	Extreme drought

View Large

In this study, drought characteristics were examined based on the frequency and severity of droughts. Drought frequency (DF) is the number of droughts in a given time, computed by dividing the number of years or months of drought occurrences by the total number of years in that period. The cumulative sum of drought indices for drought events that occurred at the site during the specified time period is referred to as drought severity (DS) (Spinoni et al. 2014).

Von Storch & Navarra (1995) proposed to use pre-whitening on the time series data before applying the Mann–Kendall (MK) test to remove serial correlation. If there is serial correlation in the time series data, it results in a disproportionate rejection of the null hypothesis with no trend, when the null hypothesis should be accepted.

For large values of r₁, the ‘null’ hypothesis of climatic series randomness being serial correlation is rejected.

Except for the Butajira station, none of the meteorological stations showed a significant serial correlation at the 5% significance level. To remove the autocorrelation from this time series, the pre-whitening method was applied. After that, the trend can be estimated (using the modified MK test).

The MK test analysis (Mann 1945; Kendall 1975) is a useful nonparametric method for assessing the significance of monotonic (increasing or decreasing) trends in climatic variables. It is the trend analysis that is most frequently employed and it can be used to evaluate climatic variables. The MK test was thus employed in this study to determine whether there were any significant trends for the drought indices (SPI, RDI, and SPEI). The MK test contrasts the alternative hypothesis of a trend that is either increasing or decreasing with the null hypothesis that there is no trend.

H₀, the null hypothesis, states that a sample of data is independent and uniformly distributed. H₁ is the alternative hypothesis, which states that X exhibits a monotonic trend.

In Equation (25), positive Z statistics imply an increasing trend, whereas negative Z statistics indicate a decreasing trend. If the standardized statistic |Z| is greater than 2.58, the trend is significant at the 99% confidence level or 1% significance level; the trend is significant at 95% confidence or 5% significance if the standardized statistic |Z| is greater than 1.96.

In addition to trend detection, it is critical to evaluate trend magnitude. Sen's Slope (β) estimator (Sen 1968) was used to estimate the slopes of drought indices.

Despite the fact that the indices reflect a similar pattern of time series variations, the peak negative values are not the same, as seen in Table 3. The magnitude of the drought also contributes to an understanding of the extent of dry spells during drought years.

For example, the peak negative value (the most severe drought) during the spring season was recorded at Hawassa meteorological station in 2015 by RDI (−2.93).

In the summer season, the peak negative value was recorded at the Ziway meteorological station in 1987 by SPI (−3.39). During the six-month period (March–August), the peak negative value recoded for Butajira station by SPI (−2.81). The drought occurrences explained by SPEI fall into the severe drought category in the majority of the stations, but those explained by SPI and RDI fall into the extreme drought category.

Drought events are also analyzed in this study based on El Nino and La Nina years (Table 2). For example, in Table 3, we can see that the drought explained by SPI, RDI, and SPEI identified an extreme drought category for Sagure station in the summer and March–August (six months) in 2009 (a moderate El Nino year). Furthermore, during crop growing season (March–August) in 2015 (a very strong El Nino year), Kulumsa and Hawassa are classified as severe drought (based on the SPI and the SPEI) and extreme drought (based on the RDI). The Ziway station, in the summer season of 1987 (the strong El Nino year), was under the extreme drought category. These findings are comparable with earlier research conducted across Ethiopia (Segele & Lamb 2005; Diro et al. 2011). They discovered that El Nino years are related to deficit in rainfall (drought).

In order to validate the findings of this study, we further compared them with some previous investigations using other drought indices. Wolteji et al. (2022), for example, conducted satellite-based agricultural drought assessments in Ethiopia's Central Rift Valley region. They investigated for the actual evidence of drought occurrence during crop growing seasons from 2015 to 2019.

Furthermore, Bhaga et al. (2020) analyzed the occurrence and impacts of major reported droughts in Africa, including Ethiopia, during the 1970s. Their assessment results demonstrated that crop failure caused food security and famine in the years 1984–1985, 1997–1999, 2005, 2008–2009, and 2015–2020 as a consequence of the continuing drought across the country, Ethiopia. In 2017, for example, 7.7 million Ethiopians faced severe famine and required emergency food relief.

As shown in Table 4, the correlation results determined by the coefficient of determination (R²) among SPI, RDI, and SPEI for all sites in the spring (February–May), summer (June–September), and six-month (March–August) time scales range from 0.772 to 0.995. The correlation value for the Ziway station is relatively low. The majority of stations exhibit a substantial correlation between the three indices and were consistent throughout the research area.

Table 5 displays MK and Sen's Slope (SS) statistic tests for detecting the significance of the trends and magnitudes of slopes, respectively, for drought indices (SPI, RDI, and SPEI) for the period of 1980–2017 (38-years). The tests were considered as two-sided assumptions at 1 and 5% levels of statistical significance. According to this table, during the spring season, drought detected by RDI and SPEI at a 1% significant level for Butajira station and by SPI at a 5% significant level for both Butajira and Wulbareg stations indicated a significant decreasing trend (increasing trends in DS), with the rate of change ranging from −0.03 to −0.04/year, while the remaining stations showed nonsignificantly decreasing trends.

Table 5

Trend test statistics (Z) based on the modified MK as well as trend magnitudes estimated by for the three drought indices (SPI, RDI, and SPEI) from 1980 to 2017

Time .	Test .	Spring (February–May) .			Summer (June–September) .			6 months (March–August) .
		SPI .	RDI .	SPEI .	SPI .	RDI .	SPEI .	SPI .	RDI .	SPEI .
Kulumsa	Z	−1.01	−1.36	−0.70	0.45	−0.10	0.03	0.13	−0.68	0.28
		−0.01	−0.02	−0.01	0.01	0.00	0.00	0.00	−0.01	0.01
Ziway	Z	−0.88	−0.65	−1.94	1.28	0.63	−0.20	0.35	−0.33	−0.65
		−0.01	−0.01	−0.03	0.02	0.01	0.00	0.01	−0.01	−0.01
Butajira	Z	− 2.24*	− 2.77**	− 2.87**	0.28	−0.03	0.68	−1.01	−1.21	−1.01
		−0.04	−0.04	−0.04	0.00	0.00	0.01	−0.01	−0.02	−0.01
Sagure	Z	−1.28	−1.43	−1.76	0.10	0.00	−0.08	−1.03	−1.08	−1.33
		−0.02	−0.02	−0.03	0.00	0.00	0.00	−0.01	−0.01	−0.02
Wulbareg	Z	− 2.09*	−1.63	−1.68	0.35	0.78	0.70	−1.43	−1.06	−1.08
		−0.03	−0.03	−0.03	0.01	0.01	0.01	−0.02	−0.01	−0.01
Hawassa	Z	0.15	−0.78	−0.85	0.10	−0.55	−0.63	0.20	−0.58	−0.75
		0.00	−0.01	−0.02	0.00	−0.01	−0.01	0.00	−0.01	−0.01

Time .	Test .	Spring (February–May) .			Summer (June–September) .			6 months (March–August) .
		SPI .	RDI .	SPEI .	SPI .	RDI .	SPEI .	SPI .	RDI .	SPEI .
Kulumsa	Z	−1.01	−1.36	−0.70	0.45	−0.10	0.03	0.13	−0.68	0.28
		−0.01	−0.02	−0.01	0.01	0.00	0.00	0.00	−0.01	0.01
Ziway	Z	−0.88	−0.65	−1.94	1.28	0.63	−0.20	0.35	−0.33	−0.65
		−0.01	−0.01	−0.03	0.02	0.01	0.00	0.01	−0.01	−0.01
Butajira	Z	− 2.24*	− 2.77**	− 2.87**	0.28	−0.03	0.68	−1.01	−1.21	−1.01
		−0.04	−0.04	−0.04	0.00	0.00	0.01	−0.01	−0.02	−0.01
Sagure	Z	−1.28	−1.43	−1.76	0.10	0.00	−0.08	−1.03	−1.08	−1.33
		−0.02	−0.02	−0.03	0.00	0.00	0.00	−0.01	−0.01	−0.02
Wulbareg	Z	− 2.09*	−1.63	−1.68	0.35	0.78	0.70	−1.43	−1.06	−1.08
		−0.03	−0.03	−0.03	0.01	0.01	0.01	−0.02	−0.01	−0.01
Hawassa	Z	0.15	−0.78	−0.85	0.10	−0.55	−0.63	0.20	−0.58	−0.75
		0.00	−0.01	−0.02	0.00	−0.01	−0.01	0.00	−0.01	−0.01

*, **, indicate at 5 and 1%, significant levels, respectively.

View Large

Table 5 also shows that no significant decreasing or increasing trends were seen at any of the stations during the summer season and March–August (six months). Moreover, the drought indices across the research area showed nonsignificant trends in the majority of the stations. This finding is also agreed with Bisrat & Berhanu (2019). They found no significant trends in the spatial and temporal variability of the drought index across Ethiopia.

The DF and severity were higher (increased) in the spring and during a six-month period (March–August) for almost all stations in the recent past decades (1999–2017), as shown in Figures 5(a), 5(c), 6(a), and 6(c). Additionally, Figures 5(b) and 6(b) demonstrate that, for the Butajira and Kulumsa stations, DF and severity increased from 1999 to 2017 compared to 1980–1998 during the summer season.

Based on the trend analysis outlined in the preceding section and the DF and severity feature as shown in Figures 5(a), 5(b), 6(a), and 6(b) in the spring and in summer seasons, we found that, in most stations, the research area is more vulnerable (more prone to drought) to drying conditions in the spring than in the summer. This conclusion is in line with Jury (2014). He discovered that Ethiopia experiences more dry spells in the spring.

Figures 5(c) and 6(c) show the frequency and severity of droughts during a six-month period (March–August), respectively. These statistics show that both the frequency and severity of agricultural droughts have increased recently. This judgment is consistent with Wolteji et al. (2022). In the research region, they discovered the phenomena of recent increases in the occurrence of agricultural droughts.

Reduced rains (drought) during months when crops demand the most water can have an impact on agricultural output. Because Ethiopia's economy is heavily reliant on rain-fed agriculture, frequent droughts cause severe economic losses, slow GDP (Gross Domestic Product) development, and crop failure. As a result, rainwater gathering, supplementary irrigation practice, and drought-resistant crop varieties should be developed in the research area to offset the harmful effects of drought on agricultural production.

In this study, we demonstrated that the three drought indices behaved similarly during the historical time series variations from 1980 to 2017. However, they differ in their classification of DS (moderate, severe, and extreme drought). As indicated in Table 3, for example, the peak negative value for Butajira station in the northwestern part of the study area from March–August (six months) was in 2017, the computed value of SPI is −2.81 (extreme drought), RDI is −2.78 (extreme drought), and SPEI is −1.94 (severe drought category). Furthermore, for Kulumsa in the northeastern portion of the study area, from March–August in 2015, both the SPI (−1.77) and SPEI (−1.96) identified the severe drought category, and the RDI (−2.06) revealed the extreme drought category. These analyses suggest that employing multiple drought indices for drought monitoring across the study area has advantages.

According to Table 5, the trend results for the three drought indices in the spring season, for example, show nonsignificant decreasing trends (the decreasing trend is not statistically significant at the 1 and 5% significance levels), despite the fact that two stations are significantly decreasing (Butajira and Wulbareg), and the decreasing trends in the drought indices imply an increase in DS and occurrence.

In addition to precipitation, the RDI and SPEI use PET, which can be estimated with temperature, to reflect the consequences of drought in the context of global warming. However, temperature data for calculating PET are sparse on spatiotemporal scales across the research region as compared to observed or gauged precipitation data. As a result, the drought explained by RDI and SPEI would be calculated at a few sites scattered around the research zone.

SPI is simple to calculate and requires only rainfall data, which is relatively available on a spatial and temporal scale across the research area. It can be a plausible solution in data-scarce areas. Furthermore, the significant correlation of SPI with other indices (RDI and SPEI) has aided in the understanding of other drought indices. This study emphasizes the significance of selecting appropriate drought indices while taking data limitations and DS classification into account.

This study explores and compares drought characteristics based on meteorological drought indices such as SPI, RDI, and SPEI for six meteorological sites (Butajira, Ziway, Kulumsa, Sagure, Wullbareg, and Hawassa) in the CMER region from 1980 to 2017. In this study, Sen's Slope estimator was used to evaluate the magnitude of trends, and the modified MK test was used to estimate statistical significance, which was set at 1 and 5%.

The results showed that in most places, the drought indices identified no significant increasing or decreasing trends throughout all time periods. During the spring season, however, drought identified by SPI, RDI, and SPEI for Butajira station in the northwestern part of the basin and by SPI for Wulbareg station in the western part indicated a significant decreasing trend (increase in DS) with a rate of change ranging from −0.03 to −0.04/year.

In this study, we also looked at two parameters, namely the frequency and severity of drought, in order to assess drought characteristics between two distinct periods (1980–1998 and 1999–2017). According to an analysis of drought characteristics and trends, drought incidence has increased in recent decades, which might be disastrous for the agricultural activity that depends on rain. Furthermore, the research region is more vulnerable to drought in the spring than in the summer at most sites. We also discovered that the extreme dryness in the study area was linked to ENSO episodes.

The remarkable relationship between the drought indices at different time scales suggests that they are well suited to the study area. Previously, researchers mostly employed a single index to examine drought conditions across the research area, with no evaluation of the applicability of the various indices. As a result, the current study, which employs a number of drought indices, could assist water resource planners better understand and monitor drought situations. The findings of the study could enhance the knowledge of water resource planners in understanding drought occurrences in different time periods and locations.

The authors would like to thank the Ethiopia Meteorological Institute for providing meteorological data for this study. We are also grateful to the editor and three anonymous reviewers for their insightful remarks and suggestions, which have increased the quality of this paper.

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

The authors declare there is no conflict.