Drought characterization and severity analysis using GRACE-TWS and MODIS datasets: a case study from the Awash River Basin (ARB), Ethiopia

Drought is one of the most complex climate extreme events that affect the agricultural productivity and socio-economic development of a region (Van Loon 2015; Mera 2018; Pendergrass et al. 2020). East Africa including Ethiopia has been experiencing more frequent, long-lasting, and disastrous droughts that impact the lives of millions (Gebremeskel et al. 2019; Nguvava et al. 2019; Thomas et al. 2019). The chain of mountains and rugged topography in East Africa caused unusual climate variability and considerably lower total annual rainfall than the rest of equatorial Africa, which plays a vital role in recurring droughts (Liebmann et al. 2014). Additionally, the warmer climate in the future is capable of increasing the duration and intensity of droughts in various places across the globe (Pendergrass et al. 2020). Most of the droughts in Ethiopia have been linked with El Niño events (El Kenawy et al. 2016; Suryabhagavan 2017; MacDonald et al. 2019). Droughts in Ethiopia have resulted in crop failure, death of livestock, human losses, conflict, diseases, epidemics outbreaks, and water shortage (Gebremeskel et al. 2019; Nguvava et al. 2019). Hence, timely characterization of historical droughts in Ethiopia using recent advancements in remote sensing is crucial. Drought in Ethiopia used to occur every 3–5 and 6–8 years in the northern part as well as every 8–10 years for the entire portion of the country, but recently more frequent droughts have occurred even in 3 or 2 consecutive years (Mera 2018). The recent historical droughts in 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, and 2011 have affected a number of 8.74, 13.46, 2.5, 5, 8, 6, 14, 14, 10, 12, and 12 million people with a cost of humanitarian aid of 92.33, 496.41, 58.92, 285.5, 393.12, 276.0, 1,077.8, 707.75, 488.8, 822.5, and 187.9 million USD, respectively (Shitarek 2012; Mohammed et al. 2017; World Bank 2017; USAID 2018). In Borena district alone, 247,000 and 26,000 animals were lost due to the droughts in 2006 and 2008, respectively (Shitarek 2012; Mohammed et al. 2017; World Bank 2017).

Drought assessment and monitoring have been carried out using remote sensing datasets, especially where ground-based observed datasets are not sufficient (AghaKouchak 2015). Remote sensing approaches are cost-effective and efficient to quantify droughts at different scales (Sinha et al. 2019). Recent advancements in satellite remote sensing help to generate new datasets that are used to propose new drought indices. For example, the Gravity Recovery and Climate Experiment (GRACE) satellite was released in 2002/03 that used to measure fluctuations in the terrestrial gravity field on a monthly time scale used to generate recent drought indices (Tapley et al. 2004; Thomas et al. 2014). First, the long-period mean gravity field should be deducted from the monthly values, and then the monthly variation of the gravity field can be expressed in terms of mass change unit that is correspondent with water height (Ditmar 2018). In other words, the monthly change in the gravity field is primarily due to the water mass variation over the continent. Thus, the variation in the terrestrial gravity field is equivalent to the fluctuation of landmass water storage (TWS), which is a new reliable means of assessing TWS variation (Xie et al. 2018). Moreover, TWS is a measure of integrated water storage in a vertical, which comprises surface water, groundwater, vegetation water, and soil moisture.

The importance of utilizing GRACE data for drought quantification and characterization have verified in various places worldwide (Yirdaw et al. 2008; Chen et al. 2009; Frappart et al. 2012; Long et al. 2013; Sinha et al. 2019; Yu et al. 2019). That is to say, conventional drought assessment that relies on the prolonged deficit of rainfall (meteorological drought), depletion of soil moisture that limits the availability of water to crops (agricultural drought), and a shortage of streamflow (hydrological drought) were not sufficient to quantify the integrated land water storage over a larger basin (Sinha et al. 2019; Sun et al. 2018). Therefore, drought assessment as well as characterization using recently proposed drought indices and satellite information is crucial to provide important information for resource management along with supplementing the previous studies.

Furthermore, the global warming effect has not been well represented in drought indices of the commonly used Standardized Precipitation Index (SPI) and Palmer Drought Severity Index (PDSI) (Beguería et al. 2014; Vicente-Serrano et al. 2018). Thus, to incorporate the current warming effect, an improved version of the SPI that includes temperature, which is the Standardized Precipitation Evapotranspiration Index (SPEI), was proposed by Vicente-Serrano et al. (2010). Plus, actual evapotranspiration (AET) and potential evapotranspiration (PET)-based drought indices, such as the Evaporative Stress Index (ESI) and the newly proposed Evaporative Demand Drought Index (EDDI) have been utilized to capture rapidly evolving drought events (Anderson et al. 2007; Hobbins et al. 2016). Both the ESI and EDDI are indirectly sensitive to soil moisture anomalies that are used to characterize rapidly evolving drought events though droughts are generally slowly developing and receding phenomena (Otkin et al. 2018; McEvoy et al. 2016; Wilhite & Pulwarty 2017). These indices are appropriate for drought early warning and preparedness although the ESI has a limitation to being implemented in various regions (McEvoy et al. 2016; Yao et al. 2018).

The most recent droughts have been frequently affecting the rift valley, eastern, and southeastern parts of Ethiopia, where 90% (104,737 km²) of the ARB is located with the reduction of 30% crop production in 2015/16 drought event alone (Shiferaw et al. 2014; Mera 2018; Thomas et al. 2019; Berhane & Tesfay 2020). The main economic activity of Ethiopia, including the ARB, is reliant on rain-fed agriculture. In addition to rain-fed agriculture, livestock keeping is a significant source of livelihood in the ARB, particularly for pastoralist societies living in the downstream part, which is also highly susceptible to drought. The ARB is the most populated, utilized as well as developed river basin in the country, which has been seriously affected by water scarcity and recurring droughts (Edossa et al. 2010; Murendo et al. 2010; Hailu et al. 2017). The impact of drought on agriculture and livestock production has also been disastrous in the ARB. Drought characterization including magnitude, duration, severity, and intensity is vital for drought prediction (Wilhite & Pulwarty 2017). Therefore, it is essential to characterize drought in the ARB based on its severity, frequency, duration, and magnitude which could help to minimize production losses.

Various indices of drought have been proposed for drought characterization and monitoring. However, some indices are area-specific, and their performance should be verified prior to applying them to different areas (Bayissa 2018). Drought indices have their own merits and demerits so drought monitoring using multiple indices can better characterize drought events than using a single index (Bayissa 2018; Yao et al. 2018; Davarpanah et al. 2021). Moreover, depending on data availability multiple indices can be used to better characterize drought events than using an index as each drought type can be sensitive to various parameters such as temperature, rainfall, evapotranspiration, and PET (Yao et al. 2018; Davarpanah et al. 2021). In this study, the most frequently used drought indices such as the SPEI, SPI, and ESI as well as the recently proposed ones like the EDDI and the Water Storage Deficit Index (WSDI) were applied. The objectives of this study are (1) to evaluate the spatiotemporal variability of TWSA and estimate the Water Storage Deficit (WSD) to use it as a means to characterize droughts in the ARB; (2) to compare the WSDI with other drought indices, namely the SPI, SPEI, EDDI, and ESI; and (3) to examine the applicability of recently proposed drought indices over the ARB by comparing them with the conventional drought index of the SPI at different time scales.

Most previous studies in the ARB have focused on drought assessment including its causes (Edossa et al. 2010; Masih et al. 2014; Mera 2018; MacDonald et al. 2019). These studies have applied commonly used drought indices that are not sufficient to assess the integrated land water storage and evaporative demand of the atmosphere. Characterization of drought using multiple drought indices that incorporate the significance of temperature and new developments in satellite remote sensing is crucial for ARBs that have inadequate observed datasets. Moreover, the various consequences of droughts within Ethiopia are specific to different places so characterizing drought on a basin scale is important. Although it is difficult to explicitly quantify and fully estimate the impacts of droughts, their characterization by including more information is necessary. Therefore, it is essential to utilize the new developments in remote sensing data and proposed drought indices for the better characterization of droughts in the ARB that could be helpful to enhance water resources management and minimize potential fatalities.

The water shortage has been severe in recent years due to climate change, population growth, and urbanization which has affected the development activities in the ARB (Dost et al. 2013; Adeba et al. 2015; Hailu et al. 2017). The three seasons in the ARB based on rainfall distribution are (i) the main rainy season (MRS), nationally called ‘Kiremt’ from June to September (JJAS), (ii) the slight rainy season (mRS), nationally known as ‘Belg’ from February to May (FMAM), and (iii) the dry season, also called as ‘Bega’ from October to January (ONDJ) (Degefu et al. 2017). These seasons are mainly due to the rotation of the earth around the sun, and the monsoon system in the area that direct the atmospheric circulation. The rugged topography also plays a significant role in the regional rainfall distribution (Viste 2012).

The AET and PET Moderate Resolution Imaging Spectroradiometer (MODIS) Terra data were obtained from the National Aeronautics and Space Administration (NASA) and the United States Geological Survey (USGS) of https://e4ftl01.cr.usgs.gov/MOLT/MOD16A2.006/. The MODIS Terra dataset, i.e. MOD16A2, an 8-day temporal scale (8-day average) was used for this study with a spatial resolution of 500 m (Running et al. 2017a, 2017b). The AET and PET data used for this study were obtained from 2002 to 2017. The Penman–Monteith equation was used to generate the MOD16A2 product of the AET, while the Priestley–Taylor equation is used to estimate the PET (Mu et al. 2007; Mu et al. 2011; Seleshi 2018). The detailed MODIS flow chart of generating AET and PET products of MOD16A2 is presented in the user's guide (Priestley & Taylor 1972; Running et al. 2017a, 2017b).

The observed rainfall and temperature datasets for 15 years (2002–2016) were acquired from 35 stations of the National Meteorological Institute (NMI), Ethiopia (Figure 1). The data obtained from NMI were preprocessed and quality checked.

Vicente-Serrano et al. (2010) proposed the SPEI which is highly responsive to global warming and has been used to characterize droughts. The computation of the SPEI is like the SPI, except that the SPEI includes temperature, which can be computed as the difference between rainfall and PET (rainfall-PET) (Vicente-Sarrano et al. 2010; Homdee et al. 2016). The SPEI is calculated by adjusting a data series of (rainfall-PET) to fit with the log-logistic distribution. The difference between rainfall and PET is referred to as the climatic water balance with the atmospheric evaporative demand (Beguería et al. 2014). The drought severity is associated with evaporative water demand because of evapotranspiration as the temperature rises could better be identified using the SPEI (Homdee et al. 2016).

ESI is sensitive to moisture stress that indicates terrestrial water availability (Zhang et al. 2019). The ESI value ranges from zero to one that is associated with dryness or wetness conditions of the land surface. The higher (positive) ESI means the atmospheric demand of evapotranspiration is met by available soil moisture and vegetation, whereas the lower ESI means the land surface meets little or none of the atmospheric ET demand (Christian et al. 2019). However, after normalization the ESI values range from − 3.5 to +3.5 and the positive ESI value indicates less or no drought occurrence, whereas the negative ESI value shows the occurrence of drought for a given location and period (McEvoy et al. 2016; Vicente-Serrano et al. 2018). Normalization by PET helps to minimize the AET variation due to the seasonal difference of available energy as well as vegetation coverage, which further emphasize on the currently available moisture to vegetation regardless of previous moisture conditions (Anderson et al. 2016; Farahmand 2016).

EDDI is a newly proposed drought index only relying on PET (Hobbins et al. 2016). It measures the drying potential of the atmosphere that induces vegetation stress on the ground (Hobbins et al. 2016; McEvoy et al. 2019). In other words, it uses to monitor the atmospheric evaporative demand that leads to the onset and development of drought when extreme atmospheric anomalies like high temperatures or lack of rainfall persist for several weeks (Christian et al. 2019). Similar to the rainfall (SPI), SPEI, and AET/PET(ESI), the monthly PET was adjusted to a normal distribution, and the cumulative probabilities were normalized. Positive EDDI value denotes a drier condition than normal that might lead to drought, whereas the negative values indicate a wetter condition (Hobbins et al. 2016). Like ESI, SPEI and WSDI, EDDI was accumulated to 1-, 3-, 6-, 9-, and 12-month time scale.

These drought severity classes are used to describe the magnitude of drought events over the ARB is indicated in Table 1.

The minor rainy and dry season spatial variations of TWSA over the ARB are illustrated in Figure 2. As displayed in Figure 2, the mRS TWS notably reduced across the basin with a reduced rate of 0 to −13.5 cm/year, except for 2014 and 2016. The maximum reduction of mRS TWS was observed in the western, northwestern, and southwestern portions of the basin as compared to the eastern and downstream of the ARB. The highest reduction rate mainly varying from −6.5 to −13.5 cm/year was noticed in the western part of the ARB, whereas the reduction rate in the downstream portion of the basin varied from 0 to −5.5 cm/year. Besides the evident seasonal variation of TWS that it increases in the MRS and reduces in the mRS and dry seasons corresponding to the rainfall pattern, the overexploitation of groundwater for irrigation and other purposes was a reason for the depletion of TWS in the mRS.

As shown in Figure 2 above, the dry season reduction of TWS was less severe than the minor rainy season. Additionally, the number of years that showed a decrease in TWS was less as compared to the mRS. Overall, a Belg season (FMAM) TWS decrease before 2010 and increase after that can be linked with ENSO that more La Niña years before 2010 as La Niña decreases the FMAM rainfall in the basin. After 2010 there were more El Niño years plus high temperatures in the Indian ocean (Dipole Moment of Indian Ocean (DMI)) that increase the FMAM rainfall in the basin (Getahun et al. 2021). Moreover, studies revealed that water storage in the African landmass is getting higher after some time due to climate change/variability (Reager et al. 2016; Hasan et al. 2019; Hasan & Tarhule 2019). The dry season TWS reduction rate is less than −3.5 cm/year in most of the years, except for 2004. Generally, as time went by, there was a clear increase of TWS in the ARB for both mRS and dry seasons. Similar to the MRS and annual series the MK test detected a significantly increasing trend of TWSA for the minor rainy and dry seasons with an increasing rate of 4.5 and 3.1 mm/year, respectively as indicated in Figure 3 and Table 2. Correspondingly, the dry season rainfall trend was detected to increase insignificantly, while the mRS rainfall was identified to decrease insignificantly. Besides, the groundwater abstraction, an insignificant decreasing trend of rainfall may contribute to the severe reduction of TWS in the minor rainy season (mRS).

In summary, the ARB experienced a significant increasing trend of annual and seasonal TWSA from 2002 to 2017, while the annual and seasonal rainfall exhibited an insignificant trend. However, the mRS hasexperienced a severe reduction of TWS as compared to other seasons.

The drought severity results presented in Table 3 were calculated using Equation (4) of the method section. As indicated in Table 3, among the thirteen drought events detected in this study period, the maximum drought duration was 15 consecutive months (from Jan-2005 to Mar-2006) with the highest drought severity that recorded a total water deficit of −411.8 mm. The second long-lasting drought periods that lasted for 11 months were from 2003/03–2003/12 and 2009/04–2010/02 with a severity of −300.1 and −162.0 mm, respectively. On the other hand, because of the short duration, the highest average total water deficit of −48.14 mm, and the second-highest drought severity was observed from 2004/05–2004/12 which lasted for 8 months with a total water deficit of −385.1 mm. The study period's highest peak deficits were recorded in June 2008 and December 2002 lasted 6, and 3 months with a magnitude of −72.3 and −71.8 mm, respectively. As shown in Table 3, the drought event in the basin is so severe that either two-drought events per year or long-lasting drought events for two consecutive years were detected during the study periods.

The TWS deficit during the annual (2002–2009), the dry season (2002–2005) and all seasons in 2004 may be strongly related to climate change/variability (ENSO and DMI) the delay effect of the earliest major El Niño of 1997 go along with by three La Niña events in 1998, 1999, and 2000, which El Niño reduces the MRS rainfall, and La Niña reduces the mRS rainfall (Getahun et al. 2021). Besides Getahun et al. (2021) indicated that the warming in the Indian Ocean resulted in a reduction of rainfall in the ARB. GRACE-TWS change is a result of seasonal water mass movement that can be affected both by natural variability and human intervention are responsible for drought and identifying their individual influence of climatic variability/change, land use land cover and regional hydrology along with how they affect each other is highly challenging (Vishwakarma 2020). The mRS rainfall contribution to the annual rainfall of the ARB was 42%; the downstream part is highly influenced by the mRS rainfall. Moreover, the years 2002 and 2004 were strong El Niño years that could cause persistent drought in the basin. The mRS TWS deficit in the ARB for the whole study period is due to the excessive water abstraction for agricultural water demand and other uses (Adeba et al. 2015; Mersha et al. 2018). These authors also mentioned that the water withdrawal in the basin from December to April has been excessive and inappropriate, which was not based on proper water resources management. Moreover, the rainfall distribution and high evapotranspiration in the basin could cause the TWS deficit (Xu et al. 2019; Yu et al. 2019).

According to the MK trend test, a significant increasing trend of annual and seasonal TWS was detected in the ARB that in line with a study carried out in the Sahel region (Okavango River basin, Lake Victoria, Lake Tana, the Blue Nile/Nile River basin) (Ahmed et al. 2014; Hasan et al. 2019; Hasan & Tarhule 2019). On the contrary, an insignificant trend of rainfall, Lake Tana level, and TWS were identified in the Upper Blue Nile (Seyoum 2018). Overall, the significantly increasing trend of TWS in the ARB could be explained by global changes (i.e. climate change/variability, land cover change, and other human activities) rather than rainfall patterns alone. The construction of dams as well as reservoirs could increase the TWS in the ARB (Berhe et al. 2013; Ahmed et al. 2014; Xu et al. 2019). The water flux from the neighboring basins and an insignificant increase in rainfall in the MRS may also increase the TWS in the ARB. A study in the Sahel found that the land use/cover change to agricultural land exhibited a rapid increase in land water storage (groundwater recharge) via diffuse recharge (Ibrahim et al. 2014; Werth et al. 2017). They called it the ‘Sahelian paradox’ since it contradicts the fact that the agricultural land expansion due to deforestation reduces the infiltration capacity and increases runoff that reduces recharge. However, different water conservation structures and ponds could enhance groundwater recharge in agricultural lands. The expansion of farmlands in the ARB could increase the TWS regardless of rainfall reduction (Getahun & Van Lanen 2015; Ayele et al. 2018). Reager et al. (2016) also found an increase in terrestrial water storage around the globe, whereas the sea level rise is slowing down (decrease) which could be induced by climate change and variability. This global sea level decrease could also be another reason why the TWS increased in the ARB.

Figures 5 and 6 show the time series of the EDDI, ESI, SPI, and SPEI at a time scale of 1 month as well as WSDI and Nino3.4 SSTA in the ARB over the study period. There were more extreme and severe drought events under SPI than other drought indices; this suggests that extreme droughts are mainly controlled by rainfall than temperature. The overall number of drought events in a 1-month time scale under SPI, ESI, EDDI, and SPEI were 12, 10, 8, and 9, respectively. The El Niño event in 1997 goes along with the three La Niña events in 1998, 1999, and 2000 that ensure 4 successive years of rainfall decrease in the lower lowland parts of the ARB, where the mRS is the foremost contributor, besides the major El Niño in 2002 that reduce rainfall in the MRS. These consecutive significant deficits of rainfall may cause severe water depletion from 2002 to 2009 that correspond to severe drought events in 2002, 2003, and 2004. Furthermore, the year 2004 was also a strong El Niño year that could reduce rainfall in the basin.

The years 2002, 2004, 2009, and 2015 were categorized as El Niño years (warming phase of ENSO), whereas 2007, 2008, 2010, 2011, and 2016 were classified as La Niña years (cooling phase of ENSO). In general, the warming phase of ENSO is known to cause a reduction of the MRS rainfall and increase the mRS rainfall, whereas the opposite is true for the cooling phase of ENSO. As can be seen, the El Niño years were identified as extreme or severe drought events by most of the drought indices. As an illustration, most drought events in the basin are associated with El Niño episodes even if a few La Niña years were also detected as drought years in the ARB.

The temporal variation of 3- and 6-month time scale SPI, ESI, SPEI, and EDDI drought indices were depicted in Figures 7 and 8. In essence, drought indices of the 6-month time scale showed fewer drought events with longer drought duration as compared to the 3-month time scale. The 3-month ESI revealed that 2006, 2009, and 2012 were extreme or severe drought years, and the remaining years showed moderate drought events, except for 2003, 2007, 2010, and 2016. Based on the 3-month SPEI, EDDI, and SPI the severe and extreme drought events occurred in 2002, 2008, and 2009. Additionally, SPEI and EDDI discovered severe and extreme drought events in 2006, whereas the SPI detected extreme as well as severe drought events in 2013 (Figure 7). Like the ESI, the 3-months SPEI, EDDI, and SPI showed moderate drought in most of the remaining years except for 2003, 2007, 2010, and 2016.

The 6-month analysis of SPI, SPEI, ESI, and EDDI showed that severe and extreme drought events mostly occurred in 2002, 2008, 2009, and 2012 (Figure 8). Moreover, two or three drought indices included 2003 and 2006 as extreme and severe drought years, and most of the remaining years experienced moderate drought, except for 2005, 2007, 2010, and 2014. The drought in 2006 was severe under ESI and SPEI, implying that the influence of temperature in this year was higher than the rainfall. Overall, under the same time scale, the drought indices of ESI, SPI, SPEI, and EDDI showed a high degree of similarity. The patterns of SPI and SPEI were much closer to each other than the remaining indices. In line with our results, many other studies indicated that overall, the MRS rainfall in Ethiopia decreases during El Niño event, while La Niña increases the MRS and the reverse occurs during the mRS (Korecha & Barnston 2007; Gleixner et al. 2016; Degefu et al. 2017; Getahun et al. 2021).

Figures 9 and 10 show the drought indices of SPI, SPEI, EDDI, and ESI over 9 and 12-month time scales from 2002 to 2016. Based on SPI, SPEI, EDDI, and ESI drought indices of a 9-month time scale, severe and extreme drought events were detected in 2002, 2003, 2008, and 2009, except that ESI showed moderate drought in 2002 as indicated in Figure 9. Furthermore, SPI and SPEI drought indices of a 9-month time scale showed moderate drought in 2005, whereas indices of SPEI and ESI indicated moderate drought in 2006. This indicates that the moderate drought in 2006 was highly influenced by temperature rather than rainfall. Under a 9-month time scale of SPI, SPEI, EDDI, and ESI drought indices, moderate droughts were also detected in 2011, 2012, 2013, 2015, and 2016.

As indicated in Figure 10, the 12-month drought indices of SPI, SPEI, and EDDI exhibited extreme and severe drought events in 2003 and 2009. A time scale of 12-month ESI identified extreme and severe drought events in 2008, whereas SPEI of the same time scale detected moderate drought in 2008. Using all drought indices of 12-month time scale moderate drought events were detected in 2012 and 2015. As indicated in Figure 10, one or two-drought indices of 12-month time scale detected moderate drought events in 2009, 2010, 2013, or 2016.

The characteristics of drought namely, duration, magnitude, and intensity of ESI, EDDI, and SPEI drought indices were calculated using the same approach as SPI, described in Equations (5) and (6. Regarding the WSDI drought index, drought duration, severity, and other characteristics of drought were computed differently as described in Equations (2)–(4).

• I.

  DM and severity using the WSDI

Considering WSDI, for a drought event to occur the negative values of WSDI should last for three or more successive months, under those twelve drought events were detected in the ARB during 2002–2017 (Figure 12). Based on WSDI, the maximum drought severity of −28.07 occurred from 2004/05 to 2006/03 for the duration of 23 months with a peak deficit of −2.32 in 2004/08. This indicates an extreme shortage of total TWS in the ARB. The next most severe drought occurred from 2003/02 to 2004/03 for the duration of 14 months, with a total severity of −14.83 and a peak deficit of −2.22 in 2003/02. As shown in Figure 12, the most persistent and severe drought events were indicated from 2002 to 2009, but afterward, the drought events were less in number and not severe.

• II.

  DM and intensity using the SPI, ESI, EDDI, and SPEI

Moreover, the drought onset (starting month) varies from one drought indices to another most with one or two months back and forth. The drought onset also differs based on the monthly time scale change as indicated in Figures 12 and 13. Overall, with the increase in the time scale of drought indices, the drought duration increased, while the number of drought events considerably reduced. As indicated in Figures 12 and 13, in most of the cases, EDDI, SPI, and SPEI were capable to detect a supposable drought onset month earlier, which is useful for drought estimation to be used for drought warning. Overall, it was possible to capture the most recent historical drought years in Ethiopia (2002, 2003, 2004, 2005, 2006, 2008, 2009, 2011, 2012, 2015, and 2016), particularly on 3- and 6-month time scales.

Several studies that have been investigating the cause of drought in Ethiopia and concluded that ENSO is the main cause of most severe drought events in Ethiopia (Segele & Lamb 2005; Korecha & Barnston 2007; Fekadu 2015; Gleixner et al. 2016; Degefu et al. 2017), which agrees with the findings in the ARB. Based on the suggestion that one index is not adequate for accurately detecting and characterization of drought events (Zhong et al. 2020), different commonly used (SPI, SPEI, and ESI) and newly proposed (WSDI, EDDI) drought indices were used for the better characterization of drought events in the ARB. Since WSDI, EDDI, SPI, ESI, and SPEI are formulated differently using different variables and distribution functions, some variation among drought indices is expected. However, the results indicated that drought indices exhibit a similar overall pattern of detecting drought events, except for the variation in the onset and duration of droughts. A study in Upper Blue Nile detected early onset of droughts using SPEI and SPI indices that correspond with our findings (Bayissa 2018).

The most severe prolonged droughts of 2002–2003 were reported by several studies in Ethiopia including the ARB which coincides with our results (Suryabhagavan 2017; Mohammed et al. 2017; Bayissa 2018; Yadeta et al. 2020). In general, drought indices of (SPI, ESI, SPEI, and EDDI) are able to detect most of the drought years such as 2002, 2003, 2004, 2005, 2006, 2008, 2009, 2011, 2012, 2015, and 2016, especially on 3- and 6-month time scales (Viste & Sorteberg 2013; Masih et al. 2014; Mays 2014; Suryabhagavan 2017; USAID 2018; MacDonald et al. 2019). Studies across the globe also indicated that the ongoing climate change induces drought is becoming intense, sever and frequent in recent decades (Shamshirband et al. 2020; Sahana et al. 2021; Band et al. 2022). Hence, regular assessment of drought using deferent drought indices is vital to minimize the impact of drought to the most suspectable sector of agriculture. GRACE-based drought index is an excellent tool to characterize hydrological or water resources related droughts: however, comparing/validating WSDI/WSD with other conventional drought indices may not be a good approach as the hydroclimatic element used for each index is different (Vishwakarma 2020). Vishwakarma (2020) stated that the severity of drought is not only dependent on climate anomalies (rainfall deficit and high temperature) but is also on the landmass hydrological process and conditions. The drought indices of SPI and SPEI well capture the drought events in Ethiopia and have been used in many parts of the country (Tefera et al. 2019). Overall, our results show that the severity and the length of the drought period coincides well with the major droughts reported in Ethiopia (Mohammed et al. 2017; World Bank 2017; USAID 2018; MacDonald et al. 2019).

The spatial-temporal TWSA showed a clear seasonality that the MRS TWSA increased across the basin following rainfall patterns, whereas the TWSA of the mRS and dry seasons decreased across the basin in most of the years. The excessive water abstraction for irrigation and other uses was another reason for the depletion of TWS in the mRS and dry seasons besides the seasonality. The terrestrial water storage in the ARB was significantly increased and the increasing rates of the annual and MRS rainfalls were 3.6 and 3.1 mm/year, respectively. The SPI time scale of 9 months was better correlated with TWSA in the ARB, indicating that SPI-9 can be used for assessing groundwater drought in the absence of groundwater measurements. The WSD analysis from 2002 to 2017 showed a significant water shortage before 2009; afterward, the water storage increased over time. The most severe water deficit occurred from 2002 to 2006 that coinciding with the rainfall shortage in the basin. Of thirteen identified drought events using the WSD, the most severe drought was from 2005/01 to 2006/03 lasting for 15 months with a WSD of −411.8 mm and a peak shortage of −46.24 mm in 2005/03. The drought in 2004 was also so severe that drought occurred twice per year from Jan-Mar and May-Dec with a total WSD of −120.9 and −385.1 mm, respectively. Based on the results of the WSDI, severe and moderate drought events occurred from 2002 to 2009 and in 2012 within which 2002, 2003, 2004, and 2008 were relatively severe drought years. Based on the results of the WSDI, the most severe drought of −28.07 shortage occurred from 2004/05 to 2006/03 for a duration of 23 consecutive months with a peak shortage of −2.32 in 2004/08, representing an extreme terrestrial water shortage in the ARB. The second severe drought event occurred from 2003/02 to 2004/03 for a duration of 14 consecutive months, with a total severity of −14.83. The severe deficits from 2002 to 2006 might be explained by the lag effects of the earliest El Niño events in 1997 go along with three La Niña events in 1998, 1999, and 2000 noting that the La Niña event will induce rainfall reductions in the mRS. Furthermore, the El Niño events in 2002 and 2004 years induced reductions in the MRS rainfall that lead to persistent drought over the years.

Regardless of longer or shorter time scales, SPI, ESI, SPEI, and EDDI identified drought events in 2008–2009 and 2012–2013, and the most severe drought in 2002. Drought indices of SPI, ESI, SPEI, and EDDI are capable to capture past drought years in Ethiopia (i.e. 2002, 2003, 2004, 2005, 2006, 2008, 2009, 2011, 2012, 2015, and 2016), particularly on 3- and 6-month time scales. The drought indices of SPI, SPEI, ESI, and EDDI with a time scale of 3 and or 6 months are the best indicators of meteorological/agricultural droughts in the basin, whereas indices of SPI, and WSDI with a time scale of 9 months are best indicators of groundwater drought. WSD severity is well quantified with drought indices of WSDI, while the magnitude of water deficit is quantified with WSD. The correlation among most drought indices was higher on 3- and 6-month time scales that a correlation of greater than 0.9 was observed among ESI vs. SPEI, SPEI vs. SPI, EDDI vs. SPEI, and EDDI vs. ESI. Furthermore, the correlation between ENSO and drought indices was also relatively high at time scales of 3- and 6 months. Thus, drought indices with 3- and 6-month showed great potential to capture agricultural and meteorological droughts in the ARB, and it is possible to use any one of these indices whenever insufficient data is available. Alternatively, correlations between the WSDI and other drought indices were higher on 9- or 12- month time scales. The ARB is highly sensitive to rainfall and a slight variation could cause a substantial influence on the socio-economic activities in the basin. Although the significant increase of TWS over time is advantageous for irrigation and economic activities in the ARB, the society relies on rain-fed agriculture and the drought is becoming more frequent and severe; so, it is vital to characterize/quantify historical droughts to develop future adaptation pathways to minimize the possible impacts of drought in the basin.

We greatly recognize groups producing SST and GRACE datasets such as NOAA/NWS/CPC, NOAA/OAR/ESRL PSD, the Center for Space Research (CSR), the Jet Propulsion Laboratory (JPL), and the GeoForschungsZentrum Potsdam (GFZ). We acknowledge and appreciate the institutions specifically the National Aeronautics and Space Administration (NASA) and the United States Geological Survey (USGS). We also acknowledge the Ethiopian National Meteorology Institute for giving us the observed meteorological datasets.

Y.S.G. carried out the process and did data analysis, data interpretation, and drafted the first version of the manuscript. M.-H.L. advised how to analyse the data, edited the paper, revised the paper, and shaped the manuscript. Y.-Y.C. revised, edited, and amended the manuscript. T.A.Y. revised and amended the manuscript.

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

The authors declare there is no conflict.