Recent global changes in terrestrial water storage (TWS)
and associated freshwater availability raise major concerns about the sustainability of global water resources. However, our knowledge regarding
the long-term trends in TWS and its components is still not well documented.
In this study, we characterize the spatiotemporal variations in TWS and its
components over the Asian and eastern European regions from April 2002 to June 2017 based on Gravity Recovery and Climate Experiment (GRACE) satellite
observations, land surface model simulations, and precipitation
observations. The connections of TWS and global major teleconnections (TCs)
are also discussed. The results indicate a widespread decline in TWS during
2002–2017, and five hotspots of TWS negative trends were identified with
trends between
Terrestrial water storage (TWS), a fundamental component of the terrestrial hydrological cycle (Tang et al., 2010), represents the total water stored above and below a land surface (Syed et al., 2008). TWS is composed of surface water (SW), including lakes, snow water equivalent, canopy water and glaciers, soil moisture (SM), and groundwater (GW) storage (Ni et al., 2018; Cao et al., 2019). Changes in TWS are strongly affected by climate change, e.g., drought, floods, prolonged high temperatures, and anthropogenic activities, e.g., abstraction-driven groundwater depletion. Recent TWS information has raised worldwide concerns because of its association with freshwater availability and concerns about the sustainability of global water resources (Creutzfeldt et al., 2015; Meng et al., 2019). Accurate monitoring and quantification of TWS are therefore critical for sustainable water resource management.
Gravity Recovery and Climate Experiment (GRACE) satellites measured global TWS changes from April 2002 to June 2017 (Reager et al., 2009), which provided hydrologists with practical insights at regional and global scales in comparison to in situ measurements (Zhang et al., 2015; Cao et al., 2019). With GRACE data, the previous literature has mostly focused on the TWS changes at the basin (Zhang et al., 2015; Shamsudduha et al., 2017; Yang et al., 2017), regional (Rodell et al., 2009; Long et al., 2013; Creutzfeldt et al., 2015; Ndehedehe et al., 2017) or continental scale (Syed et al., 2008; Rakovec et al., 2016; Yi et al., 2016; Ni et al., 2018). For instance, Rakovec et al. (2016) analyzed the TWS anomaly using GRACE in 400 European river basins. GRACE data also contributed to the exploration of hydrological storage changes, e.g., glacial mass loss (Jacob et al., 2012; Yi et al., 2014; Brun et al., 2017; Huss et al., 2018), lake level and extent changes (Zhang et al., 2013, 2017), and groundwater depletion (Rodell et al., 2009; Wada et al., 2010; Döll et al., 2014; Long et al., 2016; Feng et al., 2018; Tangdamrongsub et al., 2018). However, few studies have focused on the contributions of hydrological components to TWS variability at a large scale, particularly in water-limited and densely populated regions (Tapley et al., 2019). Two recent global-scale studies substantially improved our knowledge by identifying 34 hotspots of TWS changes during 2002–2016 (Rodell et al., 2018) and the changes in global endorheic basin water storage (Wang et al., 2018).
The Asian and eastern European regions, home to half of the world's population and 50 % of its arid/semiarid climate areas, are undergoing
intensive water exploitation for agriculture and domestic water needs (Huang et al., 2016) (Fig. 1). Most of the countries located within the borders
of Asia and eastern Europe are experiencing water resource shortages caused by low annual precipitation (less than 400 mm yr
The large-scale mode of teleconnection (TC) is an overwhelming factor in regional water resources, modulating the location and strength of storm tracks and fluxes of heat, moisture, and momentum. For example, prominent teleconnection patterns such as El Niño–Southern Oscillation (ENSO) show that El Niño years are related to reduced precipitation, continental freshwater discharge, and evapotranspiration over many land areas; therefore, TWS variability occurs over many land areas (Gonsamo et al., 2016). Many studies have attempted to address the possible causes of TWS changes by connecting TWS with TC (Phillips et al., 2012; Ndehedehe et al., 2017; Ni et al., 2018; Forootan et al., 2019). However, these studies focused primarily on the effect of ENSO on TWS. Notably, many other global and regional climate TCs have also influenced the changes in TWS; these TCs, however, have been less extensively documented, which consequently limits our understanding of a comprehensive TWS-TC correlation. Therefore, knowledge of the influence of multiple TCs on TWS is critical for improving our understanding and proper management of water resources (Phillips et al., 2012; Ndehedehe et al., 2017).
In this study, we conducted a comprehensive analysis of the spatiotemporal variations in TWS across the Asian and eastern European regions and addressed the contributions of each hydrological component and connection with TCs using multisource data. First, we calculated the de-seasonalized trend and analyzed the spatiotemporal variations in TWS across Asia and eastern Europe. Then, we partitioned the components of TWS into SW, SM, and groundwater by using GRACE, the Noah land surface model, and lakes and glacial observation data. Finally, we calculated the cross-correlation coefficients between TCs and the detrended and de-seasonalized TWS time series. We aimed to explore (1) the spatial pattern of long-term trends in TWS, (2) the contributions of water components to TWS variations among regions, and (3) the role of TCs in the changes in TWS and its components within the Asian and eastern European regions.
The Asian and eastern European regions, with arid and semiarid land comprising 54 % of their total area, are located between latitudes 6
Boundaries of the Asian and eastern European regions. Panel
The GRACE satellite measures the vertical terrestrial water storage from the land surface to the deepest aquifers and can be used to monitor
spatiotemporal variability in terrestrial water storage anomalies (Scanlon
et al., 2016). The advanced mass concentration (mascon) approach contains a
much higher signal-to-noise ratio in TWS retrieval than the traditional
global spherical harmonics (SH) technique because of reduced leakage errors
(Scanlon et al., 2018, 2019). Notably, the GRACE mascon solutions derived
from the Jet Propulsion Laboratory (hereafter JPL-M,
The Global Land Data Assimilation System (GLDAS) data between April 2002 and
June 2017 were used to partition the GRACE-observed TWS changes into SW (snow water equivalent, canopy water, lakes and glaciers), SM and groundwater. The
monthly data products from the GLDAS version 2.1 Noah model contain 36
variables, including canopy water storage, snow water equivalent and SM
data. Noah has a total of four layers of SM thickness: 0–10, 10–40, 40–100, and 100–200 cm. To compute the GLDAS TWS, the SM in all layers, the snow
water equivalent, and canopy SW are summed. The summed GLDAS TWS is
comparable to GRACE TWS over land (Rodell et al., 2004), and, notably, the
GLDAS version used here does not include groundwater and separate SW
components (such as rivers and lakes). Therefore, deviations from the GRACE
total water storage changes can be expected. A comparison between GRACE and
GLDAS is shown in Fig. S1 in the Supplement. The native spatial resolution of the GLDAS
dataset is 0.25
The term teleconnection may refer to patterns arising from the internal
variability of the atmosphere as well as from the coupling between the air
and the ocean (Zhu et al., 2017). In this study, we analyze the TCs that
dominate climate variability in the Northern Hemisphere, including the
Arctic Oscillation (AO), North Atlantic Oscillation (NAO), East Atlantic
(EA), East Atlantic/Western Russia (EAWR), Scandinavia (SCAND),
Polar/Eurasia (polarEA), West Pacific (WP), and Pacific/North America (PNA); we also analyze four important atmosphere–ocean coupled variability patterns that influence global climate, including the Indian Ocean Dipole (IOD), the
Atlantic Multidecadal Oscillation (AMO), the Pacific Decadal Oscillation
(PDO), and ENSO (Zhu et al., 2017). The first eight indices refer to Northern Hemisphere atmospheric circulation patterns. These eight indices were calculated for 20–90
Descriptions of datasets used in this study.
The original GRACE TWS signal is decomposed into long-term trends,
seasonality signals, and residual components by implementing the seasonal decomposition of time series by loess (STL) approach. The STL method is a
robust method for time series decomposition, and the equation is as follows
(Scanlon et al., 2016):
The de-seasonalized time series was used to calculate the linear trend of
TWS and precipitation for the Asian and eastern European regions from April 2002 to June 2017 using the Theil–Sen trend method. The advantage of the Theil–Sen trend analysis is that it is nonsensitive to outliers and
therefore can be more accurate than a simple linear regression for skewed
and heteroscedastic data (Sen, 1968). This method compares strongly against
the least squares method, even for normally distributed data. The TWS trend,
Contemporaneous weather conditions impact TWS residuals and often show
evident time lags. Therefore, in this study, we employ the cross-correlation
method to explore the relationship between the TWS residual and
teleconnection indices. The cross-correlation measures the similarity of the
two time series datasets as a function of the displacement of one set
relative to the other (Oppenheim et al., 2009). The cross-correlation is
defined as follows:
Methodology flow diagram of data processing in this study.
Both GRACE-based solutions (JPL-M and CSR-M) show a similar spatiotemporal pattern of changes in TWS (Figs. 3 and S3). Since the JPL-M solution
has a lack of correlation between neighboring mascon elements in the
retrieval, in this study we use JPL-M for trend analysis and mapping. JPL-M
indicates that the Asian and eastern European regions experienced widespread declines in TWS during 2002–2017 (Fig. 3a). Noticeably, the spatial
regime of the TWS variation matches that of the precipitation trend, except
for northwestern India, areas north and east of the Caspian Sea, and the area north of Xinjiang in China (Fig. 3b), thereby suggesting that variations
in TWS in these regions are intertwined with human impacts. The North China
Plain (region 1), a vast agricultural region in China, has undergone a continuously negative trend in TWS (
There are also several regions with increased TWS over the mid–high latitude, i.e., most regions of Russia and northeastern China, coinciding with an increase in precipitation in these regions during the study period (Fig. 3). Other hotspots with increased TWS in China during 2002–2017 are located in South China and the hinterlands of the Tibetan Plateau. In contrast to the sharp decline in TWS over northwestern India, TWS in central and southern India exhibits an increasing trend during the GRACE era. The variability in southern monsoons and the associated increase in rainfall likely account for the positive trend in TWS (Rodell, 2018).
Spatiotemporal changes in TWS as obtained from GRACE
Figure 4 shows the spatial distribution of the cross-correlation coefficients, illustrating the possible relationship of TCs with interannual variability in TWS. The results indicate that ENSO, AO, and NAO have a significant area of influence on TWS variability. Spatially, the pattern of correlation coefficients between TWS and ENSO is heterogeneous, with positive correlations occurring mostly in Southeast China and boreal regions and negative correlations occurring in Southeast Asia, India, and eastern boreal regions. The second and third most dominant teleconnection modes are AO and NAO, respectively. AO mainly affects TWS variations across high-latitude regions through its impact on temperature variability, and NAO has a wider footprint that is scattered across the whole study area. Following the three dominant TCs, the positive effects of IOD are scattered throughout northwestern India, southern Arabia, the European boreal region, Northwest China, and the Yellow River basin, whereas the negative effects of IOD are mainly located in Southeast Asia. Other teleconnection modes typically have a smaller impact on TWS dynamics over the study area.
Proportions of time lags for different TCs are shown in Fig. 4d. Nearly half of the area (49.14 %) lags behind the TCs by up to 3 months, while the proportions of TWS variations lagging behind the TCs at 4–6 months and at 7–9 months are 20.27 % and 12.28 %, respectively. These time lags are mainly scattered in the mid–high-latitude region and the Yangtze River basin in China. Longer lags (10–18 months), accounting for 18.31 %, are observed in parts of the Tibetan Plateau, the Mongolia Plateau, and the Middle East region. Notably, the spatial pattern of the dominant TC has only a limited extent with respect to their influence on climate conditions. The heterogeneous pattern highlighted the importance of focusing on the effect of multiple TCs on TWS rather than one teleconnection index.
Spatial distribution of cross-correlation analysis between TWS and teleconnection indices.
The changes in TWS aggregate the contributions of different water storage
components (Fig. 5). Groundwater depletion (
Contributions of different hydrological storages to TWS changes in five hotspots. Uncertainties represent 95 % confidence intervals.
Our results indicate that the water storage components are simultaneously influenced by several teleconnections (Table S2). For instance, SM in region 2 significantly correlates with NAO, AO, EAWR, PNA, ENSO, IOD, EA, AMO, polarEA and PDO, with negative correlations for some indices and positive correlations with others. Moreover, the dominant teleconnection varies for different water storage components among the separate regions (Table S3). The changes in TWS and groundwater are generally less sensitive to TC signals compared to the surface and SM counterparts. A possible explanation may be that TWS is a synthesis signal; i.e., its trend will be offset by its components in different ways. The groundwater component intensively interferes with anthropogenic activities such as irrigation and domestic needs and groundwater withdrawal, which indicates a lower correlation with TCs.
Further seasonal analysis indicates that the response of water storage to
TCs is seasonally different from one region to another (Fig. 6). For
example, TC signals have a dominant control on TWS and component variability
in spring and summer for region 3 and region 1, respectively, whereas the
signals control most of the changes in SM in region 5 in autumn and winter.
Notably, although it has been thoroughly documented that the dramatic
decline in TWS in northwestern India can be attributed to the overexploitation of groundwater (Rodell et al., 2009), our seasonal response of water
components to TCs suggests that the SM in this region is highly correlated
with spring ENSO signals (Fig. 4,
The residual time series of spring soil moisture and associated ENSO in region 3 during 2002–2017.
We investigated the spatiotemporal trend of TWS and its components over Asia
and eastern Europe during 2002–2017. The spatial patterns and trends of TWS throughout the study area are consistent with those of previous studies
(Humphrey et al., 2016; Scanlon et al., 2016). Our estimated trend
(
Periodic variability in the climate system can strongly influence regional
meteorological patterns and their associated TWS. Unlike a single
meteorological variable, teleconnection patterns control heat, moisture, and
momentum balances through their effects on temperature, precipitation, and
solar radiation reaching the Earth's surface (Zhu et al., 2017; Ni et al.,
2018). Therefore, the inherent mechanisms of the TCs' influence on TWS
variations are related to the combined simultaneous effects of TCs on
regional climate factors (precipitation, temperature, and radiation); the
changes in climate factors will substantially affect the recharge
(precipitation) and loss (evapotranspiration) of regional water resources,
which eventually influence the changes in TWS. We have identified several
dominant TCs that influence the variability in TWS and its components.
Spatially, ENSO mainly controls the TWS variation over Southeast Asia,
Southeast China, and India. During positive ENSO phases, warmer and drier
conditions can easily occur over these regions. Higher temperatures and
lower precipitation are both associated with an eventual decrease in TWS in
these areas (Ni et al., 2018). IOD is similar to ENSO and often co-occurs
with ENSO (Du et al., 2009). During a positive IOD phase, anomalous cool
(warm) waters appear in the eastern (western) Indian Ocean in association
with large-scale circulation changes that bring anomalous dry conditions to
Southeast Asia, i.e., Indonesia, while eastern Africa experiences above-average rainfall (Webster et al., 1999). The IOD may exert a negative impact on TWS
due to the decrease in precipitation over Southeast Asia. Similarly, AO primarily dominates TWS variations in high-latitude areas and the
surroundings of the Black Sea regions. When the AO index is positive, and
the vortex is intense, the winds tighten like a noose around the North Pole,
locking cold air in place and contributing to unusual warmth over the
Northern Hemisphere land masses (Zhou et al., 2001). This unusual warmth
could lead to an increase in water loss through the evapotranspiration
process, thereby contributing to a negative impact on TWS. The positive
phase of the NAO, which is highly correlated with AO (
Our results indicate that climate variability could explain the variability in TWS in most remote and sparsely populated regions. To a certain extent, climate variability may also indirectly explain glacial melt-induced changes in TWS, such as warming-induced glacial retreat. However, climate variability is influenced by human activities, such as groundwater abstraction in regions with intensive human activities. Although we obtained the contributions of different water storage components (SW, SM, and groundwater) through TWS partitioning, each component also influenced both the climate variability and human activities, which makes determinations of the influences of climate change and other processes extremely complicated. Thus, well-designed experiments and coupled human–natural system models are still needed to clarify the quantitative contributions of each influencing factor on TWS and its components' variability (Samaniego et al., 2010; Zhang et al., 2017; Rakovec et al., 2019). Several uncertainties also exist in understanding the changes in TWS and its components over the Asian and eastern European regions. These may include the unaccounted reservoirs and rivers in surface water storage, which may induce uncertainties in a certain area in groundwater estimation by eliminating the surface water and soil moisture from TWS. The glacier data used in this study were obtained during 2000–2016, which was inconsistent with our study period (2002–2017); this incongruity may also have caused uncertainties in separating the water components from TWS. Additionally, the interdependencies among multiple teleconnections may further cause uncertainties in quantifying the relationship between TWS and TCs (Runge et al., 2019). Nevertheless, our study provides a new view of teleconnections that can enable a more thorough understanding of the changes in TWS. Moreover, our study focused primarily on a water storage deficit hotspot analysis because a basin-based evaluation may experience bias in calculating the basin-averaged TWS when a given basin simultaneously experiences drying and wetting trends in different sub-basins (Sun et al., 2018). A multiscale hydrological model with high spatial and temporal resolutions may help in understanding the effects of climate variability on the hydrological response across the globe. We infer that climate variability-induced extremes, such as drought and heatwaves, will exacerbate the TWS loss; this occurs through increased consumption of water resources from groundwater for irrigation and human water demand in these hotspots, rather than the climate variability alone being the sole cause of the observed TWS loss.
There are several recommendations for future hydrological studies: (1)
withdrawal of freshwater from groundwater in water-limited regions is
important for the sustainable development of water resources and food
supplies (Rodell et al., 2018). However, groundwater drought is a distinct
phenomenon resulting from a decrease in groundwater storage (Thomas et al.,
2017). Understanding groundwater drought is important in water-limited
regions where the interplay between groundwater recharge and abstraction
results in variable groundwater stress conditions. GRACE has the unique
potential to obtain data on groundwater storage by introducing subsidiary
datasets. (2) Glacier mass loss in mountainous areas can relieve drought
stress in drought years (Huss et al., 2018), but it can additionally result
in hydroclimatic extremes, e.g., floods. Neither of these phenomena can be
detected using only precipitation datasets, such as those commonly used in
monitoring drought and flood events (Sherwood et al., 2014); this highlights
the importance of TWS-related hydroclimatic extremes. With the release of
the GRACE follow-up satellite (Famiglietti et al., 2013), consecutive
prolonged data records could provide a valuable solution for evaluating
hydrological conditions from a long-term perspective and would lead to
considerable improvements in our knowledge of TWS-related hydroclimatic
extremes (Famiglietti et al., 2013). (3) A recent study found that the
In this study, we characterize the spatiotemporal variations in TWS as well
as its components and connect these variations with TCs over the Asian and
eastern European regions from April 2002 to June 2017 using multiple data sources. The results indicate a widespread decline in TWS during 2002–2017,
and five hotspots of TWS negative trends were identified with trends ranging
between
The data and code generated in this study are available from the authors upon request (liuxianfeng7987@163.com).
The supplement related to this article is available online at:
XL and XF conceived and designed the research, XL conducted the experiments and analyzed the results, and XL wrote the manuscript with contributions from XF, PC and BF.
The authors declare that they have no conflict of interest.
The authors would like to thank the Goddard Earth Sciences Data and
Information Services Center (
This work was sponsored by the National Key Research and Development Program of China (2017YFA0604700), the National Natural Science Foundation (grant nos. 41991230, 41722104 and 41801333), the Chinese Academy of Sciences (QYZDY-SSW-DQC025), the China Postdoctoral Science Foundation (2019M650859 and 2019T120142), and the Fundamental Research Funds for the Central Universities (GK201901009, GK202003068).
This paper was edited by Luis Samaniego and reviewed by three anonymous referees.