Analysis of factors influencing groundwater drought in the Loess zone of China

Summary

Understanding the characteristics and factors influencing groundwater resources is important for regional water resources management. The Gravity Recovery and Climate Experiment (GRACE)-based groundwater conditions were used to analyze the spatiotemporal characteristics of and the factors influencing groundwater storage (GWS) distribution in the Loess zone of the Yellow River Basin. The results revealed that the spatiotemporal distribution of GWS anomalies in the Loess zone of China was best explained by the first three components of the empirical orthogonal function (EOF), representing 85.6% of the total variance. The normalized difference vegetation index (NDVI) was significantly correlated with groundwater drought (p < 0.05). In addition, NDVI and evapotranspiration (ET) were the dominant factors influencing groundwater drought. NDVI was the dominant influencing factor in 67% and 80% of the total study area between 2002–2014 and 2015–2021, respectively. This study provides important guidance for a future ecological restoration plan in the Loess zone.

Graphical abstract

Highlights

• •EOF can identify the principal components of GWSA
• •NDVI and ET are the main factors influencing groundwater drought
• •Factors affecting groundwater drought are constantly changing

Earth sciences; Geology; Hydrology

Introduction

Groundwater resources account for 30% of the total freshwater reserves in the world.¹ They are commonly used for irrigation, drinking water, and industrial purposes.^2,3 Groundwater does not respond rapidly to climate change and generally remains at stable levels, making it a valuable resource for maintaining ecosystem stability and helping people face future challenges driven by climate change.⁴ Groundwater has the security of sustained baseflow, but groundwater droughts do occur because of the very limited reservoir recharge rates.^5,6,7,8,9

In situ observation of wells is one of the methods used to accurately monitor groundwater level dynamics. Besides its high cost, the effectiveness of in situ observation depends on the number and the spatial distribution of observation wells.^10,11 Since its launch in 2002, the Gravity Recovery and Climate Experiment (GRACE) gravity satellites have been providing a systematized understanding of terrestrial water storage (TWS).^12,13,14 In fact, in contrast to the traditional well groundwater monitoring method, GRACE is cost-effective and provides accurate monitoring data that are widely used in estimating regional terrestrial water storage anomaly (TWSA) and groundwater storage anomaly (GWSA), such as in the Tarim River basin,¹⁵ the North China Plain,¹⁶ and the Lower Transboundary Indus Basin.¹⁷ Groundwater storage (GWS) is one of the components of TWS and is typically combined with the Global Land Data Assimilation System (GLDAS) to isolate groundwater anomalies from the GRACE data.¹⁸ Regional groundwater droughts can be accurately identified on a monthly scale using the drought severity index (DSI). Han et al.¹⁹ showed that groundwater drought occurred eight months following meteorological drought. Zhao et al.²⁰ demonstrated that GRACE-based DSI was sensitive to GRACE-based TWS changes. In this study, DSI was used to assess groundwater drought (DSI-g) in the Loess zone of China.

Climatic conditions such as evapotranspiration and precipitation (P) significantly affect groundwater systems.^21,22 P and downward movement of surface water directly influence groundwater recharge due to several factors, including pedological characteristics as well as changes in land use and afforestation.^4,23 Ali et al.¹⁷ revealed a significant impact of precipitation variation on the GWSA variations in the Indus Basin. Several studies have assessed the relationship between the normalized difference vegetation index (NDVI) and groundwater drought. Han et al.¹⁹ showed a significant positive correlation between NDVI and DSI-g in the Pearl River Basin between 2002 and 2015. In addition, Han et al.²⁴ revealed a clear seasonal relationship between DSI-g and NDVI in the Loess Plateau and demonstrated the greatest impact of NDVI on groundwater drought in most forest areas in the area. The main factors contributing to groundwater drought may change, however, as a result of climate change and large-scale afforestation. To the best of our knowledge, the impacts of environmental factors on the temporal variation of groundwater drought in China’s Loess Plateau are not well understood.

In the Loess Plateau, a large area of sloping arable land has been converted into woodland and grassland (Figure 1). The vegetation cover has increased since the implementation of large-scale vegetation construction projects in 1999, such as returning farmland to forest, condensing mountains, and reforesting barren hills. Although rainfall rates have increased by about 5.16 mm year⁻¹, the increase in vegetation cover increased annual potential evaporation by over 1,000 mm.²⁵ Moreover, 86% of the surface area has exhibited increasing evapotranspiration rates, resulting in increasingly severe drought.²⁶ In general, declines in terrestrial water reserves and groundwater levels are indicated by these observations. The restoration of vegetation in this area is close to the sustainable water resource limit.²⁷ Groundwater droughts may become more severe with climate change and land use conversion, altering the water circulation system in the region. Due to limited groundwater observations, long-term drought conditions in the region are underreported.^28,29

Figure 1.

Basic information about the study area

(A) Location of the Loess zone.

(B) Land use of the Loess zone in 2020.

(C) Distribution of wells, provinces, and Loess zone elevations.

The objective of this study was to assess the spatiotemporal patterns of GWSA using GRACE satellite data. Based on the results of the pattern assessment, the factors influencing DSI-g were investigated under the hypothesis that primary influencing factors may vary according to the changes in P, ET, and NDVI.

Results

Interannual variation of GWS in the Loess zone

Figure 2 shows the spatial distribution and time series of water storage anomalies in the Loess zone from 2002 to 2021. Soil moisture anomaly (SMA), canopy water storage anomaly (CWSA), and surface runoff anomaly (QSA) had annual growth rates of 1.53 mm/a (Figure 2C), while TWSA and GWSA decreased by −3.45 mm/a and −6.13 mm/a, respectively (Figures 2A and 2E). SMA, CWSA, and QSA increased mainly in areas where irrigation area increased (Figure 2C), consistent with the finding reported by Lv et al.³⁰ In addition, there were significant decreasing rates in the eastern part of the Loess zone compared to the western part during the 2002–2021 period (Figures 2A, 2C, and 2E), which may be due to the glacial melt in the Yellow River headwaters area that occurred in recent years³¹ and the long-term groundwater deficit in the northeastern part of the Loess zone.⁸ The line graphs in Figures 2B, 2D, and 2F show the temporal variation in TWSA as well as the sum of SMA, CWSA, QSA, and GWSA at the monthly scale, which were consistent with the results reported by Han et al.²⁴ in the same study area, showing that changes in TWSA are strongly influenced by groundwater.

Figure 2.

Status of water reserves

(A–F) Spatial patterns of annual trends (A, C, and E) and temporal changes (B, D, and F) in total water storage anomaly (TWSA), soil moisture anomaly (SMA), and canopy water storage anomaly (CWSA) in the Loess Zone from 2002 to 2021.

The seasonal spatial distributions of GWSA revealed clear variations (Figure 3). The highest GWSA value of 30 mm was observed in the summer (Figure 3B), while the lowest GWSA value of −25 mm was observed in the spring (Figure 3A). Direct manifestations of seasonal meteorological drought included large-scale, long-term, and continuous precipitation deficits. The results showed seasonal variations in the groundwater distribution (Figure 3). Summer GWSA decreased from south to north, while autumn GWSA increased from west to east. This finding may be due to the seasonal variation in precipitation.³²

Figure 3.

Seasonal distribution of GWSA in the Loess Zone

(A–D) Spatial distribution of GWSA in spring (A), summer (B), autumn (C), and winter (D) over the 2002–2021 period.

Changes in NDVI in the Loess zone

Figure 4 shows the annual trend in NDVI in the Loess zone from 2002 to 2021. The majority of areas showed an increasing trend in NDVI, with a maximum increasing rate of 0.041/year (Figure 4A). The southern part of the Loess zone showed the lowest increase in rate; here, large-scale vegetation restoration was not carried out, explaining the dominance of the agricultural land use class in the study area (Figure 1A). While NDVI has increased, it is possible that the region has reached its vegetation cover limit. In the central part of the Loess Plateau, the increase in NDVI was also lower. The change in NDVI was further analyzed (Figure 4B) by classifying the results obtained using the Theil-Sen-Mann-Kendall method (Table 1).

Figure 4.

Changes in vegetation cover

(A and B) NDVI trends (A) and trend characteristics (B) in the Loess zone.

Table 1.

Significance levels of trends in NDVI

β	Z	Trend features
β > 0	2.58 < |Z|	Extremely significant increase
	1.96 < |Z| ≤2.58	Significant increase
	1.65 < |Z| ≤1.96	Slightly significant increase
	|Z| ≤ 1.65	No significant increase
β < 0	|Z| ≤ 1.65	No significant decrease
	|Z| ≤ 1.65	Slightly significant decrease
	1.9 < |Z| ≤2.58	Significant decrease
	2.58 < |Z|	Extremely significant decrease

The results of the Theil-Sen-MK test were significant at a 99% confidence level and revealed extremely significant increases in NDVI over the 2002–2021 period in approximately 89.03% of the area (Table 2; Figure 4B). The areas with significant decreases in NDVI at a 90% confidence level or higher were mainly observed on the border between Shanxi and Henan provinces, as well as in the southern part of Shaanxi province, accounting for 0.7% of the total surface of the study area. These areas included several megacities such as Zhengzhou and Xi'an. He et al.³³ suggested that human activities are the main factors that influenced the changes in NDVI in this area.³³ The areas showing a lack of significant increase or decrease accounted for 3.2% and 1.3% of the study area, respectively.

Table 2.

Distribution ratio of different significance levels of NDVI trends in the Loess zone from 2002 to 2021

Trend characteristics	Percent (%)
Extremely significant increase	89.03
Significant increase	4.27
Slightly significant increase	1.3
No significant increase	3.2
No significant decrease	1.3
Slightly significant decrease	0.19
Significant decrease	0.3
Extremely significant decrease	0.4

Factors influencing groundwater drought in the Loess zone

In some cases, groundwater amounts cannot accurately reflect drought conditions; because of this, DSI-g was used after standardization to reflect drought conditions more accurately.^19,20 Correlations between DSI-g and NDVI, ET, and P were assessed using partial correlation analysis. There was a significant negative correlation of −0.91 between NDVI and DSI-g over a large part of the study area when neglecting the effects of ET and P (Figure 5A); however, when examining only one effect (ET or P), few areas passed the confidence interval. In the eastern part of the Loess zone, DSI-g and ET correlated negatively, which may be related to the high NDVI in the eastern part of the Loess zone; this relationship changed to a positive correlation in the western part of the Loess zone, where the dominant landform types are deserts and plateaus. There may be various reasons for this phenomenon, and, in addition to the influence of landscape topography, grid stacking or offsetting may also have some influence on the displayed results (Figure 5B). The correlation between precipitation and DSI-g was negative mainly in the southern part of the Loess zone, further illustrating the interaction between meteorological factors, vegetation, and groundwater (Figure 5C).

Figure 5.

Factors influencing the DSI-g

(A–C) Correlation of DSI-g with NDVI (A), ET (B), and P (C) over the 2002–2021 period (color blocks represent areas with results significant at the 95% confidence level; the legend is equally spaced).

Over the study period, NDVI played the most significant role in determining DSI-g, accounting for 79.40% of the Loess zone, while ET accounted for 17% (Figure 6).

Figure 6.

Spatial distribution of bias correlation coefficients between DSI-g and NDVI/ET over the 2002–2021 period

Discussion

GWS changes and EOF analysis

Afforestation and its impacts on regional environments remain controversial topics. The results of this study showed a decrease in GWS in the Loess zone from 2002 to 2021 (Figure 2F), which is consistent with the findings reported in previous studies.^34,35,36 Differences in the seasonal distribution of subsurface water storage were observed in the Loess zone (Figure 3). Heavy precipitation events during the wet season are only able to replenish shallow soil water,³⁷ while groundwater supplies vegetation growth during the remaining seasons, leading to the depletion of groundwater since precipitation does not replenish vegetation evapotranspiration rate sufficiently.³⁸ Previous studies have shown that dry soil layers are abundant in the Loess Plateau, with an average thickness of 2.77 m.^39,40 The presence of dry soil layers blocks the water exchange between the soil and the groundwater and between the soil and atmospheric compartments, affecting the water cycle and the sustainability of ecosystems.⁴¹

To further explain the characteristics of subsurface water storage variability in the Loess zone, the spatial (empirical orthogonal function [EOF]) and temporal distribution weights (principal component [PC]) of GWS were estimated using the EOF approach (Figure 7), with the EOF model and its corresponding PC representing raw data excluding the mean. Table 3 and Figure 7 showed that the first three components explained 85.6% of the variability, of which the first mode of EOF accounted for 63% of the total variance; this indicated that the first mode reflects the major pattern of subsurface water storage. Additionally, a long-term change in GWS was revealed by PC1. On the weight of the temporal distribution of the first mode, the value of PC1 turns from positive to negative around 2013, which is consistent with the accelerated decline of GWSA around 2013 in Figure 2F. The spatial pattern of EOF model 1 showed positive values in the central part of the Loess zone and negative values in the northern and southern parts. With the amplitude time series PC1, the variations in GWSA declined during 2002–2021. These results are in agreement with the time series of subsurface water storage obtained using GRACE and GLDAS data (Figure 2F).

Figure 7.

Three EOF models over the 2002–2021 period, EOF GWSA patterns (left), and corresponding principal components (right)

The green line indicates the Yellow River boundary, and the blue dotted line indicates the linear fit of the corresponding principal components.

Table 3.

Variance contributions for GWSA in the Loess zone using the EOF method

Component	Eigenvalues	Variance contribution rate (%)	Cumulative variance contribution rate (%)	characteristic root error range lower limit upper limit
				Lower	Upper
1	76.82	62.95	62.95	762.98	916.63
2	20.02	16.40	79.36	198.81	238.85
3	7.62	6.25	85.61	75.72	90.96
4	4.20	3.44	89.04	41.68	50.08
5	3.44	2.82	91.87	34.19	41.07
6	2.06	1.69	93.55	20.47	24.59

The second mode of EOF (EOF2), reflecting the seasonal variation in GWS, showed negative values in the southeastern part of the Loess zone, while positive values were observed in the northern and southwestern parts, demonstrating a non-uniform GWSA. In addition, PC2 showed a trend similar to a sinusoidal wave with peaks in line with variations in GWS.

The spatial pattern of EOF mode 3 and PC3 represented the monthly variations of GWSA and had lower variances than those of EOF3 and PC3.

Influence of NDVI and evapotranspiration on GWS

NDVI and ET had the most significant effects on groundwater drought. NDVI was the most significant factor affecting DSI-g over the studied period in 79.40% of the total Loess zone area, explained by the higher partial correlation coefficient of NDVI than that of ET (Figure 6). This situation occurred because the Loess zone showed a significant positive relationship between shallow soil moisture (SM) contents and NDVI due to the implementation of large-scale afforestation programs.⁴² Although there was a significant relationship between vegetation and groundwater, some areas exhibited plant roots over 10 m; however, it is uncertain whether the plants uptake groundwater through capillary rise processes during the dry season. In general, even though the deep air zone may block the direct contact between plant roots and groundwater, there are, in fact, many connections between them. Xu et al.⁴³ identified groundwater-dependent ecosystems in 22.5% of the Weihe River basin, and other researchers have assessed the relationship between vegetation ecosystems and groundwater to a depth of about 20 m, demonstrating that vegetation ecosystems can be connected to groundwater to varying degrees.^44,45 Tree age also affects soil water uptake, with older vegetation no longer using the deeper 2–10 m of soil water.⁴⁶

NDVI was strongly correlated with DSI-g compared to ET in the western and northern parts of the Loess zone (Figure 6). The correlation coefficient between NDVI and DSI-g reached 0.94 in the western part of the Loess zone, where grassland and crops are the predominant vegetation types. Cheng et al.⁴⁷ demonstrated through stable isotope analyses that the main modality of groundwater recharge in farms and grasslands was large-scale diffuse recharge. Huang et al.⁴⁸ demonstrated that the soil infiltration rates in these land use areas were faster than those in areas with economic plant species; because of this, the relationship between NDVI and groundwater under these land use types is likely to be more pronounced. In addition, Petheram et al.⁴⁹ showed that annual shallow-rooted vegetation can promote groundwater recharge in Australia more effectively than annual deep-rooted vegetation and that the soil infiltration capacity enhances the ability of deep-rooted vegetation to uptake more groundwater.

ET dominated the DSI-g in 17.00% of the area, showing scatter spatial distribution in the eastern part of the Loess zone (Figure 6). By implementing erosion prevention measures (e.g., revegetation) the soil water-holding capacity in the Loess area was significantly improved, resulting in increases in the evapotranspiration rates and decreases in groundwater recharge.⁵⁰ More than 50% of the ET increase can be attributed to climate change (mainly expressed by P increases), while revegetation (quantified by NDVI increases) played a dominant role in the ET increase, with a mean contribution rate of about 32%.⁵¹ Figure 5C shows a lack of correlation between P and DSI-g since the groundwater recharge in the rainy season is affected by the thick unsaturated zone.⁵² The thickness of the unsaturated zone in the Loess area can typically reach 350 m, with an average value of about 105.7 m.⁵³ In addition, the groundwater in the study area is derived from precipitation infiltration that occurred long ago,⁷ explaining the weak correlation between groundwater and precipitation.

According to in situ observations, the average groundwater table showed a similar trend to DSI-g from 2005 to 2018 (Figure 8), with a correlation over 0.5 (p < 0.01). To determine whether the main influencing factors of groundwater DSI-g in the study area changed between 2002 and 2021 (Figure 8), the entire studied period was divided into two sub-periods: 2002–2014 and 2015–2021 (Figure 9). The reason for this is that the time series of DSI-g from 2002 to 2021 was not horizontal, and, after attempting several segmented linear fits to DSI-g, 2015 was found to be the most appropriate dividing line. The results showed that ET and NDVI dominated 33% and 67% of the pixels, respectively, between 2002 and 2014, which was significantly different from the results between 2015 and 2021, showing an increase and decrease in the dominance of NDVI and ET, respectively. In addition, proportion values were 20% and 80% for ET and NDVI, respectively, more particularly in the core afforestation areas of the Loess Plateau. The value of PC1 changed from positive to negative around 2013, also indicating a change in groundwater during this period. According to Zhu and Zhang,⁵⁴ natural groundwater recharge in the Loess Plateau was not able to alleviate groundwater drought, despite a slight upward trend in the recharge since urban expansion; increased vegetation enhanced the intensity of groundwater drought.

Figure 8.

DSI-g and groundwater anomaly changes in the Loess zone from 2002 to 2021

Figure 9.

Changes in the main influencing factors of DSI-g

Spatial distribution of correlation coefficients between DSI-g and NDVI/ET during 2002–2014 (A) and 2015–2021 (B).

(C) Regional variation in the dominance of NDVI and ET during 2002–2014 and 2015–2021.

The results of this study demonstrated significant effects of climate changes and NDVI on groundwater drought, which are consistent with the findings reported in previous studies.^4,24,54

Influence of other factors on groundwater water storage

Rust et al.²¹ demonstrated that teleconnection factors influenced groundwater in the United Kingdom through numerous processes. Wang et al.⁵⁵ and Zhu and Zhang⁵⁴ revealed that El Niño-Southern Oscillation and Pacific Decadal Oscillation were major factors influencing groundwater drought in the Loess Plateau. In addition, the use of groundwater for irrigation and coal mining can contribute to groundwater loss. Mining in Shanxi Province, located in the Loess Plateau, interferes with the GRACE satellite estimates of the weight of the region and causes groundwater loss.⁵⁶ In fact, 61.69% of the total groundwater extraction between 2005 and 2013 was used for agricultural irrigation in the Yellow River Basin.³⁵ Despite implementing several policies to reduce arable land area (e.g., the Grain for Green project), there has been no significant increase in GWS due to the continuous increase in industrial water demand.

There are some uncertainties in the data and methodology used in this study. A combination of GRACE and GLDAS products was used to estimate GWS in this study. The GLDAS product considers SM data from the 0–200 cm soil profile, not separating deep soil water and groundwater Furthermore, the accuracy of reservoir storage data within the dataset is insufficient. This should be done in future studies. In terms of validation, only monitoring wells in the southern part of the study area were used for validation because no official information on monitoring wells in the thick vadose zone area in the central Loess Plateau has been published; this does not allow for a very accurate assessment of the changes in GWS in the study area and introduces uncertainty. Additionally, although signal recovery was used, RL06 data may mask some of the signals and give inaccurate results. Also, due to the limitation of GRACE accuracy, NDVI and P data were downscaled, possibly reducing the accuracy of the partial correlation analysis. Overall, further studies on GRACE data improvement are required.

Conclusions

In this study, the spatiotemporal variations of GWS in the Loess zone were assessed using GRACE data and the EOF method. In addition, DSI-g was employed to describe groundwater drought and assess the relationship between groundwater drought and NDVI, ET, and P using partial correlation analysis and classify the study area according to their influence on groundwater drought. The spatiotemporal distribution of GWSA in the Loess zone was explained by EOF1 and PC1, accounting for 63.0% of the total variance and representing the annual variation signal of GWSA in the Loess zone. NDVI was more significantly related to groundwater drought at the p < 0.05 level compared to the meteorological factors (ET and P). NDVI and DSI-g were negatively correlated, while ET was negatively and positively correlated with DSI-g, mainly in the eastern and western parts of the Loess zone, respectively. In addition, the southern part of the Loess zone had a negative correlation between DSI-g and P.

NDVI and ET dominated groundwater drought regulation in approximately 79.4% and 17% of the study area, respectively, from 2002 to 2021. The NDVI-dominated area increased to 67% and 80% in 2002–2014 and 2015–2021, respectively.

The regulatory effect of vegetation on groundwater was quantified, exploring the changes in factors affecting groundwater drought from 2002 to 2021. This study is the first report assessing the impacts of environmental factors on groundwater drought in the Loess zone. The results provide useful guidance to support sustainable groundwater development and the future ecological restoration plan in the Loess zone of China.

Limitations of the study

The accuracy of GRACE gravity satellites limits this study to a certain extent. Higher spatial resolution will help us to reveal the relationship between vegetation and groundwater; longer temporal resolution can confirm the status of groundwater in the Loess zone when human disturbances are relatively minor, so as to accurately assess the impact of human activities on groundwater changes, and the aforementioned limitations will be further improved in the future studies.

Resource availability

Lead contact

Further information and requests for resources should be directed to the lead contact, Yunxiao Qiu, 54289782@qq.com.

Materials availability

This study did not generate new unique materials.

Data and code availability

The datasets/code supporting the current study have not been deposited in a public repository because the fund supporting this manuscript has not been completed yet, but are available from the corresponding author on request.

Acknowledgments

This research was funded by the National Natural Science Foundation of China (U2243202, 42230714, 42277062).

Author contributions

Conceptualization, Y.Q.; writing – original draft, Y.Q. and Z.H.; writing – review and editing, Y.W. and X.Y.; supervision, G.J.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
GFZ-SH	the German Research Centre	http://isdc.gfz-potsdam.de/grace-isdc/
CSR-SH	the Centre for Space Research	http://www2.csr.utexas.edu/grace/
GLDAS	NASA	http://disc.sci.gsfc.nasa.gov
MODIS13Q1	NASA	https://lpdaac.usgs.gov/products/mod13q1v006/
CHIRPS	National Oceanic and Atmospheric Administration	https://catalog.data.gov/dataset/chirps-version-2-0-precipitation-global-0-05a-monthly-1981-present
MOD16A2	NASA	https://lpdaac.usgs.gov/products/mod16a2v006/
Groundwater table	the China Groundwater Table Yearbook for the Loess zone (2005-2018)	book
Software and algorithms
ArcGIS10.2	Esri	https://desktop.arcgis.com/zh-cn/desktop/
MATLAB R2016a	Mathworks	https://www.mathworks.com/products/matlab.html

Method details

Study area

The study area was located in the Loess Zone of China; it extends from the Great Wall in the North to the Qinling Mountains in the South, and from the Wulin Mountains in the West to the Taihang Mountains in the East (Figure 1). The study area consisted primarily of Shanxi, north-central Shaanxi, central-eastern Gansu, southern Ningxia, and eastern Qinghai, covering a total area of 360,000 km² and 45.4% of the Yellow River zone. The region is characterized by a continental monsoon climate with decreasing temperature and precipitation from southeast to northwest. The average annual temperature and precipitation are 8.45°C and 442.39 mm, respectively. Loam soil type is omnipresent in the region, explaining the significant erosion in the study area. The region is dominated by grassland, particularly in the western areas. Owing to the lack of vegetation and its specific soil characteristics, the Loess area is characterized by ecological fragility and significant soil erosion. The long-term irrational land use by humans has also disrupted the natural ground cover and affected the soil water infiltration capacity, further aggravating ecological problems in the area.

Data and preprocessing

GRACE data

The GRACE satellite data used in this study were obtained from the German Research Centre (GFZ) and the Centre for Space Research (CSR) and were freely downloaded to obtain TWS gridded data. The data version was Level-2 RL06, with a spatial resolution of 0.25 ° × 0.25 ° and covering the period from 2002 to December 2021. The missing values were estimated using the least-squares interpolation method (the years with missing values are 2002, 2013, 2016, 2017 and 2018, with 5, 3, 3 and 7 values missing respectively). In past studies, this approach has been commonly used regarding the treatment of missing GRACE values, and the accuracy of this method has been recognized.¹⁸ Even though terrestrial water storage anomalies (TWSAs) can also be described by the mass concentration block (mascon), some studies suggested that mascon solutions can reduce leakage errors.^39,55,57 The GFZ, CSR, and the Jet Propulsion Laboratory (JPL) revealed some inconsistencies in the timing and accuracy of the data, explaining why the mascon solution was not used in the current study.

Here, the GRACE data values obtained from GFZ and CSR were averaged and processed to replace the original data with C20 terms obtained from Satellite Laser Ranging (SLR)⁵⁸ to address the GRACE satellite's inability to track the Earth’s center of mass. Gaussian smoothing and destriping filters with a radius of 300 km were used⁵⁹ to remove the north-south band error and weaken the weight of the higher spherical harmonic coefficient, improving accuracy. In addition, the data obtained were subtracted from the average value of TWS observed over the 2002-2021 period.⁶⁰ TWSA was expressed in terms of equivalent water height (mm).

The monthly groundwater table data from 57 monitoring wells were collected from the China Groundwater Table Yearbook for the Loess zone (2005-2018) (Figure 1C).

GLDAS model data

The GLDAS datasets are well-suited for simulating surface hydrological processes in China.⁶¹ GLDAS has four land surface models, including Noah, mosaic, CLM, and VIC; Noah's time series is the most comprehensive and stable.^62,63 To obtain long-term hydrological processing data for the study area, the land surface GLDAS_Noah V2.1 data, with a spatial resolution of 0.25 ° × 0.25 °, were used. The selected output data consisted of soil moisture (SM), canopy water storage (CWS), and surface runoff (QS) from March 2002 to December 2021, with a monthly scale and spatial resolution of 0.25 ° × 0.25 °, consistent with that of the GRACE data. The SM data were derived from four layers of the 0-200 cm soil profile. SM, CWS, and QS grid data were acquired from GLDAS data products and were processed similarly to the GRACE data.

NDVI product

NDVI is an index that provides information about vegetation cover. MODIS-NDVI has been effective in determining changes in vegetation coverage.^42,43,64 In this study, NDVI data with a spatial resolution of 500 m were obtained from MODIS13Q1 from March 2002 to December 2021. The dataset was first obtained from 16-day NDVI pixel maximum values, and then the annual maximum and annual mean NDVI values were processed for trend analysis and biased correlation analysis, respectively.

Precipitation and evapotranspiration data

Long-term precipitation data were obtained from Climate Hazards Group InfraRed Precipitation with Stations (CHIRPS) V 2.0, with a daily scale and spatial resolution of 0.05 ° × 0.05 °. CHIRPS is efficient in estimating global precipitation. A comparison of the performance of four precipitation products in India has concluded that CHIRPS has a reasonably high level of confidence.^65,66,67 The dataset combines satellite imagery and in situ observations, providing gridded precipitation data daily. The CHIRPS Version 2.0 dataset between 2002-2021 was used, with a daily temporal resolution and a spatial resolution of 1000 m. ET data from 2002-2021 were derived from MOD16A2 data with a spatial resolution of 500 m and an 8-day composite based on the Penman-Monteith equation. Uniform resolution of all data was ensured by resampling. This data has been used frequently in relevant studies and can be considered credible.⁴³

Groundwater calculation

TWS in the vertical direction includes QS, SM, glacial snow, and CWS. Following the derivation of the distance level values, SM, CWS, and QS were separated from the GLDAS product to obtain GWSA from TWSA using the following equation⁶⁸:

$$\text{GWSA}=\text{TWSA}-\text{SMA}-CWSA-QSA$$ (Equation 1)

In each month of the study period, the mean values of GWS, TWS, SM, CWS, and QS data were subtracted to obtain GWSA, TWSA, SMA, CWSA, and QSA, respectively. Glacial snow accumulation was neglected as snow accumulation in the study area was negligible.

The groundwater studied refers to the portion of the ground 2 m below the surface, including the saturated and unsaturated portions of the soil.

DSI calculation

The GRACE-based drought index can be used to describe drought phenomena since it is consistent with the Palmer drought severity index (PDSI) and the standardized precipitation evapotranspiration index (SPEI).^19,20 Using monthly changes in GWSA derived from GRACE and GLDAS, the GRACE-based DSI was determined using GWSA data derived from the GRACE and GLDAS products according to the following formula :

$$DS{I}_{i,j}-g=\frac{GWS{A}_{i,j}-\overline{GWS{A}_j}}{{\sigma }_j}$$ (Equation 2)

where $i$ is the year ranging from 2002 to 2021, $j$ is the month ranging of the year, and $\overline{GWS{A}_j}$ and ${\sigma }_j$ are the mean and standard deviation of groundwater storage anomalies in the month, respectively. A DSI-g value of less than -0.8 indicates moderate or higher groundwater drought. The detailed classification was reported by Zhao et al.²⁰

Empirical Orthogonal Function (EOF) method

The spatiotemporal variability of GWSA in the Loess zone was assessed using the EOF method, which is based on the decomposition of the data into a spatial EOF mode and its weighted principal components (PC) with the corresponding temporal distribution. EOF has been widely used in meteorological analysis since it can filter anomalous data to identify prominent spatiotemporal patterns.^69,70 Detailed calculations can be found in Wei.⁷¹

The EOF method has been extensively employed in the analysis of GRACE-derived TWSA and GWSA.^17,25 Previous studies have shown that the EOF method can be used to downscale raw data by separating the average signal without the signals interfering during processing.^72,73

Data analysis

The NDVI trends were determined using the Theil-Sen estimator and the MK test. Long-term data series can be analyzed using the Theil-Sen estimator since this approach is not sensitive to measurement errors and outliers.^74,75,76 The Theil-Sen estimator (β) was used to estimate the NDVI variation trend in the Loess zone from 2002 to 2021. The statistical significances of the variation trends were determined using MK's Z values (Table 1).

The NDVI trend can be evaluated using the Z statistic; an absolute value of Z exceeding 1.65, 1.96, and 2.58 signifies trends at 90%, 95%, and 99% confidence levels, respectively.

Climatic changes can affect groundwater systems directly by replenishing groundwater (e.g., decreasing recharge amounts) and indirectly by affecting groundwater use.⁴ Three factors, ET, P, and NDVI, were selected to assess relationships between climatic factors and DSI-g. In this study, a partial correlation was used first to eliminate the compound effect between the factors at a 95% significance level and then to determine and assess the primary factors affecting DSI-g during 2002-2021, 2002-2014, and 2015-2021. The main influencing factors were identified for each pixel location as the maximum absolute value for the partial correlation coefficient for that pixel to determine the percentage of the study area that was dominated by the primary factors. All data were processed and plotted using Matlab 2016a, ArcGIS 10.2 software.

Published: September 11, 2024