Water storage variations and their relation to climate factors over Central Asia and surrounding areas over 30 years

Continental or regional water storage variations (WSVs) are crucial to regional economic development and human society and play an important role in coping with global change. Water scarcity is currently an especially key issue in Central Asia (CA), and therefore the study of WSVs can aid in the adoption of measures for mitigating pressures from contemporary environmental changes and economic development in CA. Based on Gravity Recovery and Climate Experiment (GRACE), Global Land Data Assimilation System (GLDAS), and CRU meteorological datasets and a proposed combined filter strategy, WSVs in Central Asia and its surrounding areas over 30 years are investigated in this paper. The results indicate that the WSVs derived from GRACE and GLDAS over CA generally show a decreasing tendency. CRU data demonstrated that CA has been undergoing a warming trend. The water loss in CA may be caused by warming, which will lead to the loss of soil moisture. Moreover, the water mass in the Tibetan Plateau and Tarim basin increases, which may be caused by glacier melting in the Pamirs and Himalaya. The precipitation contributed little to changes in water storage, but at the basin scale, the precipitation anomalies were very similar to the GRACE and GLDAS data, which can be viewed as an indicator of WSVs. This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/). doi: 10.2166/ws.2017.206 om http://iwaponline.com/ws/article-pdf/18/5/1564/251188/ws018051564.pdf 2021 Xinwu Li Yuting Chang (corresponding author) Dapeng Mu Zhongchang Sun Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing, China E-mail: cyt-shz@outlook.com Xizhang Gao State Key Lab of Resources and Environmental Information System, Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China Yuting Chang Xinjiang Environmental Monitoring Center, Urumqi Xinjiang 83011, China Dapeng Mu Jinyun Guo College of Geodesy and Geomatics, Shandong University of Science and Technology, Qingdao, China Hailong Liu Water Resources and Architectural Engineering College, Shihezi University, Shihezi, China


INTRODUCTION
Central Asia (CA) is one of most seriously ecologically and environmentally deteriorated regions in the world and now faces vegetation degradation, land desertification, soil salinization, shortages of water resources, discontinuous river flows and disappearing lakes. Water scarcity is currently an especially key issue in this region and has severely restricted economic and societal development.
Water storage is a key state variable in the global water cycle and climate change and is an extremely important resource related to economic and societal development.
Water scarcity is currently a key issue, and given that the main rivers are common to state boundaries, conflicts over water use arise regularly on national and local scales in Central Asia (CA) (Horsman ; Bernauer & Siegfried ).
Millions of people in the geopolitically important region of Central Asia depend on water from international rivers that are driven by snow and glacial meltwater, especially the Syr Darya and Amu Darya Rivers (Aizen et al. ).
The riparian countries of these rivers have experienced recurring water allocation conflicts since the collapse of the Soviet Union (Siegfried et al. ). At the same time, economic development has increased the demand for water resources, and a large amount of terrestrial water resources have been exploited in CA (Stone ). This exploitation caused friction between CA countries in arid areas and affected regional stability and national security (Qi & Evered ). Climate change is also widely expected to exacerbate water stresses in CA (Siegfried et al. ). The warming trends in CA have increased the melting rates of mountain glaciers and snow over the past 50 years and have changed the water resources of major river basins (Qi & Evered ). Issues surrounding water resources in Water storage is a key state variable of the global water cycle and climate (Famiglietti ), and it is an extremely important resource for economic and societal development (Yang et al. ). The study of water storage variations (WSVs) in CA and its surrounding areas (SA) is necessary for analyses of water recycling and climate change and for estimating water resource changes in the future. The Gravity Recovery and Climate Experiment (GRACE) was launched by NASA and the German Aerospace Center in March 2002 to observe the Earth's gravitational field and energy cycle, and has provided new approaches for WSV measurements of unprecedented accuracy (Tapley et al. a). Over 10 years, GRACE has collected large amounts of spatial data that can aid studies of variations in global and regional water storage, hydrology and water resources (Ramillien et al. ). There has recently been considerable progress in the use of GRACE datasets for monitoring WSVs at mid-latitudes in Asia. Xu et al. used

Study area
The study area included five countries in Central Asia This dataset is provided in 0.5-degree grids every month and is available at http://www.cru.uea.ac.uk/data. Figure 2 shows the methodology for analyzing the WSVs in the study area using GRACE and GLDAS. Three aspects are included: (1) preprocessing the GRACE, GLDAS and GRACE measurements are particularly sensitive to PGR (Velicogna & Wahr a, b). Therefore, this effect should be removed when GRACE is used to invert the mass change of the Earth, especially in North America, North Europe, Greenland and Antarctica, whose change rates due to PGR reach 5-11 mm/yr. The PGR signal in the study area was investigated using the ICE-5G model, which gives global uplift rates. From Figure 3, the PGR change rate in CA and its SA increased from west to east and was less than 0.5 mm/yr, which amounted to less than a 2 mm EWH correction for GRACE. Considering that the geoid accuracy determined by GRACE is approximately 10 mm, the PGR effect can be ignored in this study area.

WSV estimation based on least squares spectral analysis
The time series of mass anomalies were obtained on a 1 × 1 grid from the GRACE and GLDAS datasets. These mass anomalies mainly indicate a hydrological signal that has significant annual, semi-annual and linear changes. The least squares spectral analysis was applied to perform the deterministic modeling of the mass anomalies (Vanícěk ; Wu et al. ; Craymer ). Thus, the fitting of each grid can be conducted as follows: where H represents the equivalent water height derived from GRACE and GLDAS, t is time, A 0 is a constant, B 0 is the rate, and f 1 and f 2 are the annual and semi-annual frequencies. The annual amplitude S 1 , semi-annual amplitude S 2 and corresponding phase φ can be obtained by The amplitude of the fitting result indicates a strong or weak degree of periodic change, and a linear trend indicates increasing or decreasing water mass during a period (Rodell et al. a).

RESULTS VALIDATION AND ANALYSIS
The cross validation of WSVs derived from GRACE and

GLDAS datasets
Due to the lack of in situ measurements in Central Asia, cross validations of the WSVs derived from GRACE and GLDAS were conducted to demonstrate the effectiveness of the results. The WSV linear trends for most of the areas showed decreasing tendencies, except for areas such as the Pamir region, Tarim basin, Tibet plateau, Balkhash Basin and parts of Russia. The difference can be mainly attributed to GRACE and GLDAS measurement differences, which will be explained later in this paper.
The correlation coefficients of the WSVs derived from GRACE and GLDAS were used to quantify the effectiveness of the WSVs. Figure 5 shows the spatial variations in the correlation coefficients between the monthly GRACE-based WSVs and the average GLDAS-based WSV.
As shown in Figure 5, the GRACE-born WSV time series which could be attributed to the GLDAS model's inability to simulate ice sheet flow and mass balance, which was possibly related to groundwater changes. In these areas, the correlation coefficient value was low; it was reduced to 0. Interannual variability of water storage from

GRACE-and GLDAS-based estimates
The monthly EWH from the GRACE and GLDAS datasets was obtained using Equations (1) and (2). The trend of  EWH is calculated on 1-degree grids by using the GRACE and GLDAS datasets. In order to investigate the WSV in a wide range, we assigned the mean value of trends in the grids of the study area as the WSVs. Figure 6 shows the results of using the GRACE data (red) and GLDAS data (blue).
The series (Figure 6)  October and November. However, the loss rates indicated by the GLDAS data were greater than those indicated by the GRACE data, especially during 1999-2012 ( Figure 6).
As shown in Figure 6, there was a notable shift in the GLDAS WSV. A possible reason may be that the land cover map derived from MODIS, which has a higher resolution, was used to generate the GLDAS datasets, and more accurate GLDAS datasets were therefore obtained (Rodell & Houser ).

Spatial variabilities of the various estimates determined from GRACE and GLDAS
The least squares spectral analysis was used to capture the linear trends and amplitudes in the WSV. Figure

DISCUSSION
There were notable spatial differences in the WSVs variations in CA from the above analysis. In this section, we thus focus on the discussion of why such spatial differences exist and what the main impact factors are.

Effects of glacier on WSVs in typical regions and basins
Substantial changes in the cryosphere of Central Asia and its  Hydrological signals are stronger at the basin scale. The time series of the WSV in the four basins (see Figure 1) are shown in Figure 10, and the least squares spectral analysis results are summarized in Table 1 As shown in Table 1, the GRACE and GLDAS datasets varied in their annual amplitudes, with the exception of the linear trends. In general, the GRACE data's annual amplitudes were much greater than those of the GLDAS data, except for the annual amplitude for the Tarim River Basin.
This is easily understood, because the total water mass signal based on the GRACE data was stronger than the soil moisture data from the GLDAS dataset. For example, the annual amplitude for the Aral drainage based on the GRACE data was almost 2.5 times higher than that based on the GLDAS data (see Table 1). This can be explained by runoff and lake water, which were not included in the GLDAS simulations.

Effects of climate change on WSVs
According to high resolution regional climate models

Effects of temperature on WSVs
Similar to the meteorological station and reanalysis data, the CRU temperature data for 1979-2012 also demonstrated that the five countries in CA experienced warming trends (see the red line in Figure 11). In Figure 11, the temperature represents the average of grids in the five countries in CA, and the temperature data for 2004 meet the Mann-Kendall test for trends, which indicates that the warming trend was significant. The time series based on the GLDAS data for the five countries in CA in 2002 meets the Mann-Kendall test for trends (see Figure 11). The significance statistic for temperature in 2002 was 1.94, which approximated the significance level (see Figure 11). Moreover, except some years (1984)(1985)(1986)(1987), the statistical values of the temperature have been greater than zero since 1980 (see Figure 11), which indicate that the temperature has an obvious rising trend. By contrast, the statistical values of water reserve change have been less than zero except in 1988, which indicates the water reserve has been reducing all the time. This relationship between temperature and the GLDAS data implied that warming trends had a potential influence on WSVs.
The changes in the glaciers in CA were influenced more by climatic factors, especially temperature (Hagg Note: A is the amplitude, and P is the phase. The rates in different periods were estimated using data in the corresponding periods. The linear trends of the four basins varied in different periods (see Table 1). It can be inferred from Figure 10 and Table 1  These two selected regions are covered by ice and snow, which supply water resources to the SA. The CRU data confirmed that the temperatures of R01 and R02 increased at rates of 0.46 C and 0.39 C/decade, respectively (see Figure 12). Although the warming trend was notable, the highest annual mean temperature during 1979-2012 was below the freezing point. This warming trend may not accelerate glacial melting. Actually, the shrinkage of glaciers is highly related to the trend of increasing temperatures in the summer (Kriegel et al. ). The CRU also revealed that R01 and R02 have experienced increasing trends in June, July and August at rates of 0.49 C and 0.29 C/decade, respectively (see Figure 12).

Effects of precipitation on WSVs
Precipitation has an important effect on WSVs (Chen et al.

).
As indicated by the CRU data, the precipitation over the five countries in CA was stable during the period of the study (see Figure 11). No abnormal change was detected by the Mann-Kendall test. However, we cannot conclude that precipitation did not contribute to the WSVs over the study area.
The precipitation anomalies were obtained by subtract- Basin is more complicated than the environments in other basins because the large Taklimakan Desert is located in this basin. It is noteworthy that the magnitude of the precipitation anomaly was lower than the magnitudes indicated by the GRACE data and the GLDAS data. In addition, other contributions from runoff, lake water and groundwater cannot be neglected. Generally, at the basin scale, the precipitation anomalies were very similar to the GRACE and GLDAS data, which can be viewed as an indicator of WSVs.

CONCLUSIONS
Based on GRACE, GLDAS and CRU meteorological datasets, variations in water storage in Central Asia and its SA over 30 years were investigated in this paper. The result indicates the following. (1) The water storage values obtained from GRACE and GLDAS over CA generally exhibited a decreasing tendency, and the results from GRACE over CA were spatially more heterogeneous than those from GLDAS.
(2) For CA's SA, the water storage presented a different change tendency due to the effects of natural factors such as snow or glacier melting.
(3) The water losses in the five CA countries may be caused by warming, which will lead to the loss of soil moisture, and furthermore, the water increases in the TP and Tarim basin may be caused by glacier melting around the Pamirs and Himalaya. At the basin scale, the precipitation anomalies were very similar to the GRACE and GLDAS data, which can be viewed as an indicator of WSVs.
Due to a lack of in situ measurements related to water resource changes, a full quantitative analysis and assessment of the change tendency of WSVs was not conducted in this paper. In addition, because human and social economic data were not included in this study, the mechanisms of changes in water storage were not fully explained and understood. In the future, to further investigate water storage changes under a background of global change, an observational network needs to be constructed, and more datasets need to be collected to fully understand tendencies in water storage changes and their mechanisms and how these changes affect the ecological environment of Central Asia and its surroundings.