Abstract
Globally, surface water, groundwater, soil moisture, snow storage, canopy water, and wet biomass constituents make up water storage, which plays a significant role in the hydrological water balance. Evaluating the variations in water storage anomalies associated with climate forcing and human activities over river basins is crucial for assessing water scarcity and predicting potential pressures on water resources in the future. In this study, we assessed the impacts of climatic and anthropogenic drivers on the change in water storage in the river basins of Ethiopia by using the independent component analysis to examine Gravity Recovery and Climate Experiment with the Global Land Data Assimilate System-based water storage and comparing the independent component analysis with hydro-meteorological data and statistical data related to human activities. It is of great significance for helping people better understand the evaluation of terrestrial water storage anomalies under the combined influence of climatic change and anthropogenic activities and providing information for better protection and utilization of water resources at river basin level. It is crucial to take effective measures to protect these precious land and water resources and prevent their further deterioration. The estimated result will be essential for sustainable water management and protection.
HIGHLIGHTS
Characteristics of water storage variation was analysed over each river basin.
The impact of climatic change and human activities on water storage was evaluated.
Independent component analysis was performed on terrestrial water storage anomalies for each river basin and obtained two leading modes (the trend and annual modes).
The intensity of human activities has increased over each river basins.
Climate change is the dominant factor observed during the study period.
INTRODUCTION
Freshwater resources are critical ecosystems that support human and animal life (Geda et al. 2020). The availability of water is a critical issue that is used in different sectors of the world that serve indirectly or directly throughout the lifetime of human beings. This includes hydroelectric energy production, water supply for drinking, irrigating agriculture, transportation, industrial water supply, and others. Due to that, the demand for water has been rising all over the world (Singh et al. 2014, 2017). For efficient utilization of surface water resources that are available at balanced extents to increase economic, environmental, and social benefits, it is important to understand available water resources. Estimating the water balance means assessing the needs, current condition, and trends of the water resources on the land surface, which strengthens decision-making for water resource management and is said to be pricing. The role of opening an inclusive strategy for optimization and management of water resources to manage demand was to reduce the limit of the water problem by identifying its balance (Belkhiri et al. 2018).
Currently, different scholars are focused on water balance assessment around the world by using different models like Arc Soil and Water Assessment Tool (SWAT) and WEAP (Kumar et al. 2018). According to Gupta et al. (2014), the availability of water resources in large river basins is the current topic for discussion. Assessment of the availability of water storage was very essential for water resource planning and management. They focused on the Narmada River Basin located in India to evaluate its water availability by using an integrated approach of remote sensing and GIS tools and techniques. An evaluation of the water budget at the heart of southwestern Ontario, located in central Canada, including all drainage areas for the Grand River and its contributors, was carried out to develop scenarios for future weather conditions by using global circulation models (Golmohammadi et al. 2017). Estimation of water balance for the Chirchik River Basin in northeastern Uzbekistan was performed by using the integrated Mike SHE models to generate quantitative results of the river (Usmanov et al. 2016). Raghavendra & Cholke (2017) evaluated continental evapotranspiration by using Gravity Recovery and Climate Experiment (GRACE) in the Australian Semi-Arid Basins, which is a recently developed land surface model to assess water balance. Water balance is based on conservation of the law of mass, which shows that any change in the water content of a given soil volume during a specified period must equal the difference between the amount of water added to the soil volume and the amount of water withdrawn from it. It also helps to quantify the relationship between precipitation, evaporation, evapotranspiration, and aquifer draft and provides a framework for the future planning of sustainable exploitation of the available water resources (Kanga 2017). Based on water balance models, the water balance components have been evaluated at diverse scales in recent years. For example, McCabe & Wolock (2013) assess the mechanism of water balance at the global scale, and Herrmann et al. (2015) assess the spatial variation of water balance components at the regional level. Generally, researchers have mostly focused on the estimation of water balance components at basin or watershed levels (White et al. 2011; Gunkel et al. 2015; Uniyal et al. 2015). Different findings have been discussed on water balance through parameters like vegetation patterns, land cover types, and ecosystem level. For example, quantifying the change of every water balance component in a small watershed with diverse land cover types (Jian et al. 2015), Thompson et al. (2011) evaluated the spatial scale dependency of ecohydrologically mediated water balance partitioning for catchment ecohydrology using a simple model of water balance and vegetation coupling on a hierarchical flow path network and reported that the generated spatial dependency in a really averaged hydrological variables, water balance, and parameters describing hydrological partitioning provides a theoretical approach to connect water balance at patch and catchment scales. Jasrotia et al. (2009) estimated the moisture deficit and watershed moisture surplus by using remote sensing and the ArcGIS approach. A spatial semi-distribution of water balance model was developed to simulate the mean monthly hydrological process by using the soil conservation service curve numbers (Jenifa et al. 2010). The water balance of river basins was analyzed and quantified by the Arc SWAT model (Sathian 2009). Stream flow, water balance, and monthly inflow volume were simulated to predict water balance components by Arc SWAT (Pagliero et al. 2014).
Ethiopia is gifted by a natural water resource called the water tower of East Africa, due to water availability in the country, but there are water shortages in the areas of water supply, irrigation, and depending on a rain-fed agricultural system due to lack of proper utilization and exploitation of this natural resource of water. According to Awulachew et al. (2007), 560 irrigation potential sites were identified, and it was estimated that about 3.7 million hectares of land were available for irrigation in Ethiopia by the Ministry of Water Resources. Water availability was essential for irrigation (Bayissa et al. 2017; Tiri et al. 2018; Alemu et al. 2022). The economic growth of Ethiopia is growing at the fastest rate and is facing greater challenges to meet energy demands that come from household consumption, the industrial system, and service sectors. To meet the energy demand, mega hydropower projects are constructed on main river basins (Degefu et al. 2015). The national water resources are estimated to have the potential to generate as much as 30,000 MW of power from economically feasible hydropower projects (World Bank 2013). There are nearly 200 identified economically feasible sites for hydropower development in the country (Bartle 2002). The construction of Sugar Industries is growing. The availability of high-potential water leads us to create a good environment for the future in different directions, not only for hydropower generation but also for generating incomes due to indirect additional functions of water based on fish production, tourism, water supply, irrigation, and others (Degefu et al. 2015).
Due to the second phase of growth and transformation, the economic growth of the country was growing at a high rate because of the fast industrial growth, which led to manufacturing subsectors (Altaseb & Singh 2018). This second phase of growth and transformation is targeted at improving the number and productivity of manufacturing industries, supported by the government, to increase the growth of physical infrastructure through public investment projects in water resource basins and other areas, to feed this increasing population, to meet the high demand for hydropower, to produce maximum production by irrigation agriculture, and to increase industrialization, which leads to high demand for water in the country. According to Abboye (2021), the Ethiopian government's food self-sufficiency policy program, improving the productivity and production of bread wheat as well as facilitating its marketing access, is one of the key parts of the economic growth strategies. Due to that, understanding the current condition of water availability at river basin level was the main issue in the country for optimization and management of water resources to identify different demands related to water and to reduce water crises that were happening because of underutilization of water resources.
Spatiotemporal variation in its water storage capacity has a major impact on the socioeconomic development and livelihood of the country. Frequent natural disasters are mainly represented by large-scale and long-term drought and regional floods, which are generally linked to abnormal terrestrial water dynamics (Thomas et al. 2014; Xu et al. 2019). However, the lack of in situ observation limits our understanding of the water cycle in this region. Fortunately, studies have reported that the terrestrial water storage derived from GRACE satellite data shows both the potential for revealing variation in surface water storage and the ability to support large-scale hydrological modeling (Klees et al. 2007; Luo et al. 2016; Yang et al. 2017; Zhao & Li 2017; Frappart & Ramillien 2018; Yoshe 2023). Studies show that GRACE has displayed satisfactory results for terrestrial water storage (Li et al. 2013, 2018; Ni et al. 2014; Feng et al. 2017; Khaki et al. 2018; Wang et al. 2019). There are no studies on characterizing terrestrial water storage and factors that lead to variation in water storage based on the GRACE dataset and Global Land Data Assimilation System (GLDAS) model on the major river basin of Ethiopia, for example, Awulachew et al. (2007) focused on annual flow of the major river basin, but neglecting the spatial heterogeneity of terrestrial water storage changes. Generally, the spatiotemporal variations of terrestrial water storage and its changes are the result of the combined effect of climate change and human activities (Hu et al. 2018). Precipitation, runoff, and evapotranspiration are dominating processes of change in terrestrial water storage either in space and/or time (Tangdamrongsub et al. 2011; Frappart et al. 2013; Yang et al. 2015). Anthropogenic processes like irrigation, the abstraction of groundwater, and land use influence terrestrial water storage in highly populated areas (De Beurs et al. 2015; Huang et al. 2015; Khandu et al. 2016). Each river basin has its own particular local climates, relief conditions, and human conditions, and some of these factors affect the evaluation of terrestrial water storage. Generally, the water storage estimate provided by GRACE and GLDAS is comprehensively affected by human activities and climatic changes, and climatic change may cause us to underestimate or overestimate the impact of human activities on water storage in different periods (Chen et al. 2016; Humphrey et al. 2017; Humphrey & Gudmundsson 2019; Zhong et al. 2019; Liu et al. 2021). It is difficult to isolate the impact of climate change and human activities on water storage derived from GRACE and GLDAS observations. Many studies have attempted to assess the relative contribution of natural and anthropogenic effects on water storage (Fasullo et al. 2016; Felfelani et al. 2017; Zhong et al. 2019). Felfelani et al. (2017) isolated natural and human-induced changes in water storage by comparing results from different hydrological model simulations and GRACE data over large basins. Yi et al. (2016) analyzed the impact of human activities and climate driving on water storage by establishing a linear relationship between variation in precipitation and water storage, assuming that the impact of human activities on terrestrial water storage is constant over time. Zhong et al. (2019) quantified the contribution of climate-driven water storage and its long-term trends based on the reconstruction of long-term climate-driven water storage variations. These studies were primarily conducted based on hydrological models or established statistical models that simulate natural, dynamically driven changes in water storage to reconstruct climate-driven water storage anomalies. However, due to the uncertainty of state-of-the-art hydrological models and forcing datasets, the impact of human intervention on terrestrial water storage may be underestimated or overestimated (Scanlon et al. 2018). The primary objective of this article is to explore the potential driving factors for the change in terrestrial water storage in Ethiopian river basins. To do so, we conduct a comprehensive analysis of the terrestrial water storage from GLDAS and GRACE in the major river basin of Ethiopia based on the independent component analysis (ICA). First, we investigate the water storage changes in the major river basins of Ethiopia from GRACE and GLDAS, and then we decompose the terrestrial water storage using the ICA technique to illuminate the spatiotemporal patterns of water storage variations in the major river basins of Ethiopia. Finally, we compare the ICA modes with hydrological model data, ground statistics data (coal mining, irrigation, grazing, etc.), and simulated climatic water storage anomalies products derived from the statistical model proposed by Humphrey to explain the meaning of each dominant signal that separates through ICA results (Sun et al. 2020). This will improve our understanding of the various factors affecting basin water storage, including anthropogenic and climate-driven water storage variations. To date, no such investigation has been available in the existing literature for Ethiopia at the river basin level. Therefore, adopted approaches for long-term availability of water storage and its result will be regarded as a new and honest contribution to Ethiopia's river basins.
MATERIALS AND METHODS
Description of the study
Location map of the study area for the major river basins of Ethiopia.
Data collection
There are different satellite data products that are useful for the estimation of the water budget, such as combined Tropical Rainfall Measuring Mission (TRMM) and Global Precipitation Measurement (GPM) precipitation, which was Integrated Multi-Satellite Retrievals for GPM (MERG) as spatial and temporal resolution product at 0.1° × 0.1°, Terra and Aqua MODIS at 500 m, liquid water equivalent thickness from GRACE data product at 1° × 1°, GLDAS data product at 0.25° × 0.25°, Terra-Climate, and others. But, for this study, GRACE and GLDAS data products were used to estimate the long-term water storage of major river basins of Ethiopia. All-important data were collected to accomplish this research and discussed as follows.
Methods for data analysis
Equivalent water thickness by the GRACE satellite
The Mascon method directly estimates the liquid equivalent water thickness based on the mass change in a certain region, mainly through high-precision inter-satellite observation of data from the GRACE satellites. Its surprising feature is that the earth mass change is directly estimated as a parameter, and thus, it is more flexible with respect to the region of interest (Swenson & Wahr 2002; Wang et al. 2019). Watkins et al. (2015) designed 4,551 spherical caps of equal area with uniform mass distribution (Mascon) on a global scale and estimated the liquid equivalent water thickness of each ball cap through the weighted least squares method, using the three GRACE satellite datasets. During satellite data tracing, there was signal loss, which had to be removed to find the gravitational anomaly coefficient, and then, the spherical harmonic coefficient was obtained by the band limit of Stokes coefficient at spherical harmonic degree (Wahr et al. 1998). The damping of the signal amplitude due to filtering causes leakages (Vishwakarma et al. 2017; Chen et al. 2019; Koster & Suarez 2022) have undergone pre-processing, such as de-striping filter, Gaussian smoothing, and glacier isostatic adjustment (Chen et al. 2019). The accuracy of the GRACE product was improved by averaging the three datasets (Landerer & Swenson 2012; Seyoum & Milewski 2017), and the missing data were filled by linear interpolation (Long et al. 2015). To restore the original signal that was lost during data processing, a multiplicative scaling factor was used for the monthly data provided by the GRACE product (Seyoum et al. 2019), and more information about scaling factors can be found in Landerer & Swenson (2012). In this study, the spatiotemporal distribution of the three datasets were collected for each river basin.
Global Land Data Assimilation System
GLDAS is a commonly used land surface modeling, with the objective of producing an optimal field of land surface states and fluxes based on satellite and ground observations, and has been frequently used for the calculation of terrestrial water storage and groundwater storage (Rodell et al. 2004; Syed et al. 2008; Nie et al. 2016; Hsu et al. 2020). Four land surface models (LSMs) were used by GLDAS to provide hydrological data of the area, such as the variable infiltration capacity (VIC) (Liang et al. 1998), the community land model (CLM) (Dai et al. 2013), Mosaic (Koster & Suarez 2022), and Noah (Ek et al. 2003). The average of the four LSMs datasets (VIC, Noah, CLM, and Mosaic) was utilized to estimate the anomaly of SWE, SM, and CWS to reduce any errors or biases (Katpatal et al. 2018). For the estimation of the change in terrestrial water storage from GLDAS, 22 bands were utilized with 1° × 1° spatial resolution (Li et al. 2019). The validation of GLDAS result is confirmed by water resources and climate studies in Iran, where lack of observations inhibits advanced hydro climatological studies and sustainable water resource management (Moghim 2018). In this study, spatiotemporal variations of the GLDAS dataset were collected for each river basin.
Climatic dataset
During this study, climatic datasets like precipitation, temperature, and others were collected from the power project data products, which is the Prediction of Worldwide Energy Resources (POWER) Project, which provides solar and meteorological data from NASA research for support of renewable energy, building energy efficiency, and agricultural needs. In this data product, there are multiple data access options using the Data Access Viewer, which is a responsive web mapping application that provides data subsetting, charting, and visualization tools in an easy way.
Data of human activities
The grazing data, mining data, agricultural data, and other human activities in each river basin were collected from ministry of water and Energy of Ethiopia (https://www.mowe.gov.et). In addition, detailed information about human activities in each river basin was collected from the regional water and energy offices during study durations.
Water storage equation






Water balance equation


Reconstruction of climate-driven water storage anomalies







Independent component analysis


ICA obtains the source signals by optimizing the decomposition matrix W so that the independence between the source signals is strongest. When the best decomposition matrix W is determined, the mathematically defined Y(t) are referred to as independent components (ICs). In this study, ICA can be used to separate the source signals of terrestrial water storage. By comparing the source signals of the terrestrial water storage with the other hydrological products, it is possible to quantify the contributions of different drive factors to the variations in the terrestrial water storage anomalies over the study period.
Correlation analysis
Evaluation of methods of model performance
RESULTS AND DISCUSSION
Liquid water equivalent thickness from GRACE over the major river basins of Ethiopia
Characteristic of average LWE for the major river basins of Ethiopia.
Spatial distribution of liquid water equivalent thickness from GRACE
Spatial variation of average LWE from GRACE for the 12 river basins of Ethiopia from 2002 to 2016. (1) Spatial distribution of LWE for Abbay, (2) spatial distribution of LWE for Tekeze, (3) spatial distribution of LWE for Mereb Gash, (4) spatial distribution of LWE for Denakil, (5) spatial distribution of LWE for Awash, (6) spatial distribution of LWE for Aysha, (7) spatial distribution of LWE for Baro-Akobo, (8) spatial distribution of LWE for Omo-Gibe, (9) spatial distribution of LWE for Rift Valley, (10) spatial distribution of LWE for Genale Dawa, (11) spatial distribution of LWE for Wabishebel, and (12) spatial distribution of LWE for Ogaden.
Spatial variation of average LWE from GRACE for the 12 river basins of Ethiopia from 2002 to 2016. (1) Spatial distribution of LWE for Abbay, (2) spatial distribution of LWE for Tekeze, (3) spatial distribution of LWE for Mereb Gash, (4) spatial distribution of LWE for Denakil, (5) spatial distribution of LWE for Awash, (6) spatial distribution of LWE for Aysha, (7) spatial distribution of LWE for Baro-Akobo, (8) spatial distribution of LWE for Omo-Gibe, (9) spatial distribution of LWE for Rift Valley, (10) spatial distribution of LWE for Genale Dawa, (11) spatial distribution of LWE for Wabishebel, and (12) spatial distribution of LWE for Ogaden.
Estimation of temporal variation of liquid water equivalent thickness from GRACE
Temporal variation of mean annual LWE from GRACE for the major river basins of Ethiopia from 2002 to 2016. (1) Temporal variation of LWE for Ayesha, (2) temporal variation of LWE for Awash, (3) temporal variation of LWE for Denakil, (4) temporal variation of LWE for Baro-Akobo, (5) temporal variation of LWE for Genale Dawa, (6) temporal variation of LWE for Mereb Gash, (7) temporal variation of LWE for Ogaden, (8) temporal variation of LWE for Omo-Gibe, (9) temporal variation of LWE for Rift Valley, (10) temporal variation of LWE for Tekeze, (11) temporal variation of LWE for Wabishebel, and (12) temporal variation of LWE for Abbay.
Temporal variation of mean annual LWE from GRACE for the major river basins of Ethiopia from 2002 to 2016. (1) Temporal variation of LWE for Ayesha, (2) temporal variation of LWE for Awash, (3) temporal variation of LWE for Denakil, (4) temporal variation of LWE for Baro-Akobo, (5) temporal variation of LWE for Genale Dawa, (6) temporal variation of LWE for Mereb Gash, (7) temporal variation of LWE for Ogaden, (8) temporal variation of LWE for Omo-Gibe, (9) temporal variation of LWE for Rift Valley, (10) temporal variation of LWE for Tekeze, (11) temporal variation of LWE for Wabishebel, and (12) temporal variation of LWE for Abbay.
Estimation of terrestrial water storage changes over the major river basins of Ethiopia from GLDAS
Terrestrial water storage estimated from GLDAS data products for the major river basins of Ethiopia.
Terrestrial water storage estimated from GLDAS data products for the major river basins of Ethiopia.
Yearly and seasonal variation of GLDAS terrestrial water storage for the major river basins of Ethiopia
Temporal variation of mean annual TWS from GLDAS for the major river basins of Ethiopia from 2010 to 2020. (1) Temporal variation of TWS for Ayesha, (2) temporal variation of TWS for Abbay, (3) temporal variation of TWS for Tekeze, (4) temporal variation of TWS for Mereb Gash, (5) temporal variation of TWS for Awash, (6) temporal variation of TWS for Denakil, (7) temporal variation of TWS for Ogaden, (8) temporal variation of TWS for Wabishebele, (9) temporal variation of TWS for Genale Dawa, (10) temporal variation of TWS for Rift Valley, (11) temporal variation of LWE for Omo-Gibe, and (12) temporal variation of TWS for Baro-Akobo.
Temporal variation of mean annual TWS from GLDAS for the major river basins of Ethiopia from 2010 to 2020. (1) Temporal variation of TWS for Ayesha, (2) temporal variation of TWS for Abbay, (3) temporal variation of TWS for Tekeze, (4) temporal variation of TWS for Mereb Gash, (5) temporal variation of TWS for Awash, (6) temporal variation of TWS for Denakil, (7) temporal variation of TWS for Ogaden, (8) temporal variation of TWS for Wabishebele, (9) temporal variation of TWS for Genale Dawa, (10) temporal variation of TWS for Rift Valley, (11) temporal variation of LWE for Omo-Gibe, and (12) temporal variation of TWS for Baro-Akobo.
Spatial distribution of terrestrial water storage from GLDAS
Spatial variation of average TWS from GLDAS for the 12 river basins of Ethiopia from 2010 to 2020. (1) Spatial distribution of TWS for Abbay, (2) spatial distribution of TWS for Tekeze, (3) spatial distribution of TWS for Mereb Gash, (4) spatial distribution of TWS for Denakil, (5) spatial distribution of TWS for Awash, (6) spatial distribution of TWS for Aysha, (7) spatial distribution of TWS for Baro-Akobo, (8) spatial distribution of TWS for Omo-Gibe, (9) spatial distribution of TWS for Rift Valley, (10) spatial distribution of TWS for Genale Dawa, (11) spatial distribution of TWS for Wabishebel, and (12) spatial distribution of TWS for Ogaden.
Spatial variation of average TWS from GLDAS for the 12 river basins of Ethiopia from 2010 to 2020. (1) Spatial distribution of TWS for Abbay, (2) spatial distribution of TWS for Tekeze, (3) spatial distribution of TWS for Mereb Gash, (4) spatial distribution of TWS for Denakil, (5) spatial distribution of TWS for Awash, (6) spatial distribution of TWS for Aysha, (7) spatial distribution of TWS for Baro-Akobo, (8) spatial distribution of TWS for Omo-Gibe, (9) spatial distribution of TWS for Rift Valley, (10) spatial distribution of TWS for Genale Dawa, (11) spatial distribution of TWS for Wabishebel, and (12) spatial distribution of TWS for Ogaden.
Estimation of evapotranspiration from water balance equation
Characteristics of precipitation, temperature, evapotranspiration, and LWE for each river basin
Cross-correlation for the climatic variable with TWS and LWE. Pr is precipitation, aet is actual evapotranspiration, EVT is evapotranspiration obtained from GRACE dataset and precipitation, tmx is the maximum temperature, LWE is liquid water equivalent from GRACE dataset, and TWS is terrestrial water storage estimated from precipitation and actual evapotranspiration.
Cross-correlation for the climatic variable with TWS and LWE. Pr is precipitation, aet is actual evapotranspiration, EVT is evapotranspiration obtained from GRACE dataset and precipitation, tmx is the maximum temperature, LWE is liquid water equivalent from GRACE dataset, and TWS is terrestrial water storage estimated from precipitation and actual evapotranspiration.
The result of cross-correlation shown in Figure 10 demonstrates that the cross-correlation for Abbay river (0.96, 0.83, 0.8, −0.76, 0.76, −0.75) between (Pr + TWS), (TWS + EVT), (Pr + EVT), (Pr + aet), and (Pr + tmx), respectively; Ayesha river (1, 0.998, 0.997, −0.87, 0.44, 0.441) between (Pr + aet), (Pr + TWS), (aet + TWS), (LWE + EVT), (TWE + tmx), and (Pr + tmx), respectively; Mereb Gash river (0.92, 0.92, 0.81, 0.80, 0.79, −0.69) between (Pr + aet), (Pr + EVT), (aet + EVT), (Pr + TWS), (TWS + EVT), and (LWE + tmx), respectively; Tekeze river (0.89, 0.88, 0.87, 0.86, −0.72, −0.7) between (Pr + TWS), (Pr + aet), (Pr + EVT), (TWS + EVT), and (LWE + tmx), respectively. Denakil river (0.98, 0.89, 0.78, −0.68, 0.62, 0.56) between (Pr + aet), (Pr + TWS), (aet + TWS), (LWE + EVT), (TWS + EVT), and (Pr + EVT), respectively. The positive cross-correlation shows the variable was a direct relationship, but the negative cross-correlation indicates an indirect relationship between the variables.
Factors affecting the decrease in LWE and TWS for the major river basins
Characteristics of precipitation, liquid water equivalent, evapotranspiration, and maximum temperature: (1) Baro-Akobo River Basin, (2) Abbay River Basin, (3) Ayesha River Basin, (4) Tekeze River Basin, (5) Denakil River Basin, (6) Awash River Basin, (7) Ogaden River Basin, (8) Wabishebele River Basin, (9) Genal Dawa River Basin, (10) Rift Valley River Basin, (11). Omo-Gibe River Basin, and (12) Baro-Akobo River Basin.
Characteristics of precipitation, liquid water equivalent, evapotranspiration, and maximum temperature: (1) Baro-Akobo River Basin, (2) Abbay River Basin, (3) Ayesha River Basin, (4) Tekeze River Basin, (5) Denakil River Basin, (6) Awash River Basin, (7) Ogaden River Basin, (8) Wabishebele River Basin, (9) Genal Dawa River Basin, (10) Rift Valley River Basin, (11). Omo-Gibe River Basin, and (12) Baro-Akobo River Basin.
Advantages and disadvantages of GRACE dataset in terrestrial water storage
Hydrological prediction depends on knowledge of initial hydrological conditions. Traditionally, the analysis and prediction of hydrological extremes have involved statistical assessments of long-term climatic and hydrological datasets to evaluate flood thresholds, severity, recurrence intervals, and drought events. Efforts to improve hydrological prediction and reduce uncertainty can now leverage a wealth of new tools for measurement and estimation of land surface states, namely, satellite remote sensing products and advanced numerical models (Famiglietti et al. 2015; Li et al. 2019). The GRACE satellite mission has been an important step forward in observing terrestrial water storage with global coverage. GRACE provides monthly data on change in the Earth's gravity field, which we assume is correlated with the movement of water through the Earth's system at specific temporal and spatial scales. GRACE has been providing monthly gravity field solutions since April 2002 and has proved an effective tool to observe changes in the mass of ice, snow, groundwater storage, surface water storage, and hydrologic drought characterization, which is confirmed by different research studies (Kim et al. 2009; Houborg et al. 2012; Li et al. 2012; Thomas et al. 2014; Richey et al. 2015). GRACE has also contributed to flood estimation, which is consistent with different findings (Reager & Famiglietti 2009; Reager et al. 2014). GRACE terrestrial water storage anomaly data can contribute to an improved understanding of soil moisture and groundwater contributions to future runoff generation, but the utility of GRACE data for operational analysis or actionable prediction is challenging. Three prevailing limitations are as follows: (1) coarse spatial resolution that shows one arc-degree for the level-three data product (Landerer & Swenson 2012; Famiglietti et al. 2015); (2) the aggregated observation of multiple water storage components as a single integrated value for each grid cell; and (3) latency in data product processing and release (Reager & Famiglietti 2009; Gebere et al. 2015). These obstacles make many applications difficult, as water resource management trends occur at watershed scales and on daily to weekly timescales. Assimilation of GRACE observations enable spatial, vertical, and temporal disaggregation of terrestrial water storage components, including groundwater, surface water, root zone soil moisture, and snow, while preserving the internal consistency of the modeled storages and fluxes and taking into account uncertainties due to model and observational errors. Model uncertainty is caused by errors in surface meteorological forcing, model parameters, and model structural errors. Some of the uncertainty is related to unmodeled processes, most notably human impacts such as pumping groundwater, irrigation water supply, domestic water use, and water management practices, which are not considered by GRACE satellite. Therefore, it is common to rescale the observations prior to data assimilation to address model and observation biases. Such rescaling may discard important signals in the observation and make it difficult to isolate errors caused by unmodeled processes including true observational features for data assimilation, which is also confirmed by different findings (Reichle & Koster 2004; Zaitchik et al. 2008; Ozdogan et al. 2010; Kumar et al. 2015; Zhou et al. 2016). The water crisis is one of the main global risks that has different impacts on society. This study uses available data and tools to track water storage change in the area, where there is a lack of observations and limited hydro climatological studies. The GLDAS and GRACE datasets were globally used in water resources and climate studies, particularly in areas where data are temporally and spatially limited. Therefore, the application of GRACE data and GLDAS provides a new research approach to explore change in water storage throughout the world for areas with scare datasets and also for Ethiopia in particular. Different studies throughout the world evaluated water storage variations for larger watersheds (Zaitchik et al. 2008; Yoshe 2023) and even groundwater storage (Rodell et al. 2007). The result of the study adds high value for the study area to monitor drought events, hydrological extremes, and chemotic changes, and it is also very essential for water resource management and optimization. As agriculture was the backbone of Ethiopian economic development, this finding creates valuable information for water resource planners and decision-makers.
Terrestrial water storage anomalies decomposition using ICA
We performed ICA on terrestrial water storage anomalies for each river basin and obtained two leading modes (the trend and annual modes). Since the ranking of the components obtained by ICA corresponds to a random ordering instead of a reduction in variance, we estimated the average ratio of the contribution of each IC to the observed unfiltered time series in order to reorder ICA components in the ascending order for each river basin, and the spatial portion of each ICA result is normalized by its maximum value. For each river basin, the annual mode, which explains more than 80% of the total variance of the filtered terrestrial water storage anomalies time series, is followed by the trend mode, which explains less than or equal to 20% of the total variance for each river basin. The spatial weight of the IC shows a positive and negative contribution to the corresponding time functions for each river basin based on spatial water variation in each river basin, climatic change, and anthropogenic activities. The negative and positive weights correspond to decreasing and increasing water storage in the study area, which may be affected by distributions of land use, land cover, population density, precipitation, soil type, and different human activities in each river basin.
Comparison of the ICA model with simulated climatic water storage anomalies and Humphrey water storage anomaly products
For each river basin, the annual mode and the trend mode show the seasonal fluctuation in water storage anomalies. Our results are generally consistent with the Humphrey result of terrestrial water storage anomalies Equation (5) (Humphrey et al. 2017), which are affected by climate variability, with an R2 value for each river basin greater than 85%. This shows that the climatic water storage anomalies in the major river basins of Ethiopia can be explained by the annual mode and the trend mode, respectively, and that ICA may provide a new method to isolate the impact of climatic variability and human activities on water storage variation. For each river basin, the maximum seasonal water storage fluctuations and the average amplitude of climatic water storage anomalies from GRACE and the GLDAS model vary from each other because the climatic water storage anomalies derived from GRACE, based on ICA, integrate all surface and subsurface water storages, whereas the GLDAS hydrological model contains soil moisture estimates from the surface to a certain depth, surface runoff, canopy water storage, and others, excluding the contribution of reservoir and groundwater storage. Each river basin has experienced a similar climate with little variation in drying and warming climates, and the impacts of climate change on terrestrial water storage anomalies have good consistency. During the dry season for each river basin, the climatic-derived water storage loss rates are very high, whereas in each river basin, water storage increases were observed during the wet season. This comes in response to precipitation fluctuation, which indicates that precipitation was the dominant driving force for the terrestrial water storage depletion in each river basin. Water storage anomalies with and without human interventions are used for each river basin to further isolate the effects of human intervention on terrestrial water storage anomalies. The result of human-driven water storage anomalies is consistent with the result of the statistical data yearbook. The primary causes of variation in terrestrial water storage are the effects of human interventions, which are mainly based on irrigation data and other activities. Scanlon et al. (2018) also suggest that global models underestimate large decadal declining and rising water storage trends relative to GRACE satellite data, and our result was consistent with these findings.
Each river basin has experienced significant increases in mining, irrigation, grazing, urbanization, hydropower construction, industrial activities, and other human activities. Overgrazing caused grassland degradation and resulted in a decrease in soil moisture retention capacity and groundwater depletion in each river basin, which is also confirmed by similar findings (Tao et al. 2015). The increase in irrigation expansions is highly increasing in each river basin and has caused extensive groundwater withdrawals, which is consistent with similar studies (Tao et al. 2015; Zhong et al. 2019). Coal mining is the most important product for economic development, and these activities were increasing in each river basin, which is an extremely water-intensive industry; each ton of coal mined more than 2.54 m3 of water, which may cutoff rivers and destroy underground aquifers, as confirmed by similar studies (Tao et al. 2015; Liu et al. 2021). Compared to the rapid expansion of farmland due to increasing population growth in each river basin of Ethiopia, the agricultural area was increasing in each river basin, which added stress on water storage. The difference in population density in each river basin leads to a huge difference in the human intervention on water storage variation. In wet seasons, the impact of human activities on land water storage may be offset by climate changes; in dry seasons, intensive human activities may lead to a further decrease in terrestrial water storage in each river basin of Ethiopia, which will exacerbate drought and cause several ecological problems, including a sharp decrease in surface runoff, shrinking lakes and reservoirs, and intensifying desertification. It is crucial to take effective measures to protect these precious land water resources and prevent their further deterioration in each river basin in Ethiopia. This study provides a potential method to quantify the climate- and human-driven water storage variations in the major river basins of Ethiopia. It is of great significance for helping people better understand the evaluation of terrestrial water storage anomalies under the combined influence of climatic change and anthropogenic activities and providing information for better protection and utilization of water resources at river basin level. It will also help to facilitate basin (regional) sustainable water resource management and protection.
CONCLUSION
In this study, we used the GRACE dataset to evaluate LWE for major river basins of Ethiopia from 2002 to 2016. Combining with the GLDAS hydrological model, we evaluated the spatiotemporal variation of terrestrial water storage. The impact of climatic parameters on the terrestrial water storage was evaluated, and the following conclusions were drawn.
From the perspective of interannual variation, the liquid water equivalent shows a continuous increase and decrease, which indicates a distinct seasonal cycle. The liquid water equivalent gets its peak in the rainy season for each river basin, whereas it gets its lowest value during the dry season. The maximum and minimum LWE for each river basin, the monthly variation, and the spatial distribution of LWE for each river basin were evaluated for this study. The spatial distribution of LWE for Abbay ranges from −2.07 to 0.68 cm/month, Tekeze (−0.44 to 1.04 cm/month), Mereb Gash (−0.44 to 0.47 cm/month), Denakil (−0.82 to 1.04 cm/month), Awash (−0.41 to 1.85 cm/month), Ayesha (1.46 cm/month), Baro-Akobo (−2.1 to 2.43 cm/month), Omo-Gibe (−1.42 to 4.41 cm/month), Rift Valley (0.82 to 4.89 cm/month), Genale Dawa (2.78 to 5.6 cm/month), Wabishebele (0.69 to 3.57 cm/month), and Ogaden (1.45 to 2.11 cm/month). In addition, the overall trend in the study area was evaluated for each river basin. The annual water storage for each river basin was fluctuating in the study area. We understand that liquid water equivalents derived from GRACE show both the potential to reveal changes in terrestrial water storage and the ability to support hydrological modeling. In addition, from the GLDAS land state surface variables, water storage change was estimated for all river basins. For estimation of water storage change from the GLDAS dataset, soil moisture (surface soil moisture, profile soil moisture, and root zone soil moisture), canopy water storage, and surface water storage are the variables used. The time series of the estimated water storage for the mean monthly water storage for each river basin evaluated shows continuous rising and falling, showing the seasonal pattern of the study area. The evaluated liquid water storage from GRACE datasets and GLDAS was compared for each river basin. There is a match for the seasonal peak and low of water storage for each river basin. However, interannual variation shows a significant difference in each river basin for some months of study periods. The cross-correlation coefficient between water storage from GRACE and GLDAS for each river basin was evaluated.
Under the background of intensive climate warming and human activities, quantifying water storage dynamic process is a key issue for ecologically fragile areas such as the major river basins of Ethiopia, this work implicated that it is possible to access the impact of climate change and human activities on terrestrial water storage anomaly changes through the ICA method and multi-sources datasets. It can help manage and protect water resources and provide possible solutions for sustainable ecological and socioeconomic development.
The increase in evapotranspiration was also the dominant factor contributing to the decrease in LWE/TWS for each river basin, while the decrease in precipitation had a certain effect on LWE/TWS as well. We observed that an increase in temperature causes a decrease in terrestrial water storage for all river basins. We concluded that climate change and anthropogenic activity influence terrestrial water storage, mainly precipitation, temperature, and evapotranspiration. Finally, we prove that GRACE satellite provides users with valued data even in areas where available data are limited. In addition, the GLDAS dataset also provides the necessary dataset for estimating the water budget of an area. The quantified result demonstrates that understanding the spatiotemporal distribution is important to avoid water crises in future, and conserve natural water resources by monitoring the water storage. Policymakers and researchers also benefited from this finding as baseline information for the study area.
FUNDING
This research did not receive any specific grant from funding agencies in the public, commercial or nonprofit sectors.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The authors declare there is no conflict.