Both meteorological droughts and human activities modulated groundwater variations in the northern Yellow River Basin

Groundwater is a critical source of freshwater (Zhang et al. 2019; Scanlon et al. 2023), providing approximately half of the water used for global irrigation and the majority of the drinking water for domestic needs (Jasechko & Perrone 2021; Noori et al. 2023). With population growth and increased water demand, groundwater withdrawal has increased eightfold to meet human needs (Hanasaki et al. 2018; Wada et al. 2014; Bierkens & Wada 2019). Simultaneously, the expansion of irrigated agriculture has intensified groundwater withdrawal, particularly in semiarid regions where groundwater may be the only reliable source of water due to the absence of permanent surface water (Wada et al. 2012; Ashraf et al. 2021). Frappart & Ramillien (2018) found that global groundwater storage has been declining at an increasing rate of depletion. Continuous decreases in groundwater levels can lead to serious issues, such as land subsidence, seawater intrusion, and a decline in agricultural and industrial productivity (Lin et al. 2022). Therefore, it is crucial to understand the influential factors contributing to the groundwater decline.

Increases in drought frequency, duration, and intensity have greatly exacerbated groundwater depletion (Zhao et al. 2022). Given the prevalence of frequent drought events, examining how groundwater depth (GWD) may change in response to meteorological droughts is crucial. Meteorological droughts exert both direct and indirect impacts on groundwater systems (Thomas et al. 2017). Direct impacts involve climate variables, such as precipitation, influencing groundwater recharge and causing fluctuations in groundwater levels (Kubicz et al. 2019). Indirect impacts encompass certain human activities aimed at alleviating droughts (Shah et al. 2021), which intensify the consumption of groundwater induced by meteorological droughts (Schober et al. 2018; Yang et al. 2020). Ojha et al. (2018) found that severe droughts resulted in approximately a 2% loss of total aquifer system storage in California's Central Valley. Agarwal et al. (2023) discovered that the rate of groundwater water storage depletion is faster during the drought periods in Central Valley. Long et al. (2013) observed a significant depletion of water storage, monitored with the GRACE satellite, induced by the 2011 drought in Texas. Previous studies have revealed the influence of droughts on groundwater, elucidating that the process of propagation from meteorological droughts to groundwater is also crucial.

Numerous studies have focused on the propagation of meteorological droughts to groundwater in natural states or monitoring wells without human elements (Mishra et al. 2018; Aadhar & Mishra 2020). These studies reveal the mechanism by which natural elements, such as soil moisture (Zhang et al. 2021a), aquifer characteristics (Bloomfield & Marchant 2013), and vegetation (Liu et al. 2023), influence the propagation of meteorological droughts to groundwater systems (Apurv et al. 2017). Furthermore, these studies contribute to a deeper understanding of how meteorological droughts influence groundwater systems. However, human activities complicate the propagation from meteorological droughts to groundwater systems, and this complexity may exhibit spatial heterogeneity (Zhang et al. 2022b, 2022c). For instance, Yang et al. (2020) found that human activities reduce the severity of hydrological droughts in the south of China but exacerbate hydrological droughts in the north. While some studies have revealed the intervention of human activities in the influence of meteorological droughts on streamflow (Tijdeman et al. 2018; Zhang et al. 2022b), research on how the human activities intervene in the propagation from meteorological droughts to groundwater systems is rare. Given the significant role of groundwater in water resources and the increasing frequency of drought events due to climate change, it is crucial to uncover the mechanisms by which meteorological droughts propagate to groundwater systems in regions with high levels of human activities.

The Ordos, located in the northern Yellow River Basin, is ecologically fragile and frequently disturbed by human activities (Ma et al. 2019). Serving as an essential ecological barrier in northern China, Ordos has witnessed the implementation of various ecological restoration projections, including the Natural Forest Conservation Program (NFCP) and the Grain for Green Program (GFGP) (Feng et al. 2016; Chen et al. 2019). The Mu Us sandy land (MUSL) within Ordos stands as one of the most successful examples of ecological restoration and desertification reversal (Tian et al. 2015; Zhang & Huisingh 2018). Numerous studies confirm the increasing vegetation coverage in MUSL (Zhang & Wu 2020; Zhao et al. 2020; Sun et al. 2021), and this vegetation restoration has led to heightened groundwater consumption (Zhang & Wu 2020; Luan et al. 2023). In addition, farmlands, primarily distributed in the north, rely heavily on groundwater irrigation. Some studies indicate that groundwater can buffer the impact of drought events on vegetation (Deng et al. 2022) and delay the propagation from meteorological droughts to agricultural droughts (Fawen et al. 2023).

However, it remains unknown whether the impact of meteorological droughts on groundwater will change under the influence of human activities, particularly in the semiarid regions with fragile ecosystems. As such, major objectives of this study are (1) to identify the spatiotemporal variation characteristics of groundwater levels in Ordos; (2) to reveal the intervention of human activities in the propagation from meteorological droughts to groundwater systems; and (3) to clarify the lag time of groundwater system's response to meteorological droughts. The grand aim is to support the local government's efforts to manage human activities, thus implementing measures to cope with droughts.

We used the standardized precipitation index (SPI) (McKee et al. 1993; Xu et al. 2015; Eini et al. 2023) to represent the meteorological drought, which could be derived from precipitation (see Section 2.3). Here, the monthly precipitation data from 1970 to 2020 were obtained from the National Tibetan Plateau Data Center Third Pole Environment Data Center (https://data.tpdc.ac.cn/zh-hans/data/faae7605-a0f2-4d18-b28f-5cee413766a2) at a spatial resolution of 1 km (Shouzhang 2020). The daily GWD data for 67 monitoring wells in the Ordos, covering the period from 2018 to 2022, were sourced from the local hydrology departments. We used the daily GWD data to cluster the variation of the groundwater dynamics. We aggregated the daily GWD data into a monthly scale by averaging them for a better examination of its correlation with SPI.

Considering the cumulative effect of meteorological droughts, we calculated the SPI at time scales varying from 1 to 48 months.

We employed a two-step approach for the GWD clustering. The first step is to realign the GWD time series and measure the similarity among groundwater time series. In the first step, we used the dynamic time warping (DTW) algorithm. DTW can find optimal global alignment between sequences (Petitjean et al. 2011) and is widely used to quantify the similarity between sequences (Kruskall & Liberman 1983; Aach & Church 2001; Bar-Joseph et al. 2002; Gilleland & Roux 2015; Mantas et al. 2015). In this method, each point of the first sequence is compared with any point of the second sequence. Finally, sequences with similar patterns that occurred in different periods are considered similar (Izakian et al. 2015). In the second step, we used the K-means algorithm to cluster rescaled time series. K-means is one of the most widely used algorithms for clustering (Fejes Tóth 1959; Ball & Hall 1965; MacQueen 1967; Lloyd 1982), because of its simplicity, efficiency, and empirical success (Jain 2010). In this study, we wrote a program in the Python language and performed the two successive steps: DTW algorithm and K-means clustering.

In this study, we calculated the PCC between GWD and SPI at time scales ranging from 1 to 48 months to identify the time scale of SPI with the most significant impact on GWD. If the SPI has impacts on GWD, the PCC between SPI and GWD is negative. The time scale of SPI with the minimum PCC is considered to have the greatest impact on GWD. We categorize drivers based on the absolute value of the PCC and the p-value into three categories. When |PCC| > 0.5 and p < 0.05, SPI has a strong and significant impact on GWD, and the primary driver influencing the GWD is SPI alone. When |PCC| < 0.5 and p < 0.05, SPI has a weak but significant impact on GWD, with its influence disturbed by human activities related to meteorological droughts, such as irrigation and ecological restoration (Udovičić et al. 2007; Feng et al. 2020). When p > 0.05, the impact of SPI on GWD is insignificant, and the primary drivers influencing GWD are human activities unrelated to meteorological droughts, such as managed groundwater recharge projects.

Typically, the variation in GWD lags behind the variation in SPI. To determine the lag time, we calculated the PCC between GWD and SPI for the preceding 1–48 months. Similarly, the month with the minimum PCC between GWD and SPI is considered the lag time for GWD's response to meteorological droughts, which has been widely used in drought propagation studies (Zhang & Jia 2013; Wang et al. 2017; Wan et al. 2023).

The spatial distribution of all monitoring wells clustered to the four patterns is also shown in Figure 2(e). Initially, we hypothesized that the variation in the groundwater table might be influenced by land use classification, and the monitoring wells in each category would follow distinct spatial rules. However, the spatial distribution rules for each category are not evident. This means the land use classification is not the direct factor influencing the groundwater dynamics. Since meteorological factors and human activities are common factors influencing groundwater dynamics, we further use statistical methods to explore the driving factors of groundwater dynamics.

The groundwater dynamics are divided into four patterns in Ordos. The first pattern is continuously increasing. This may be caused by irrigation-driven deep percolation (Huo et al. 2011). It can also be attributed to the geological structure. For instance, Guangcai et al. (2008) found that the NID belongs to the Cretaceous Aquifer System, where precipitation is the major recharge source and groundwater flow is controlled by topography. Due to the higher topography than surrounding areas, there may be no lateral recharge from the surrounding groundwater systems. Chen et al. (2024) found a similar increasing pattern in southeastern Australia during the millennial drought period and believe that the increasing trends may be related to lateral recharge from mountainous areas. The second pattern is jagged decreasing, which may be caused by frequent anthropogenic abstraction for industry. The third and fourth patterns are both seasonal fluctuations, while the third one is annually flat and the fourth one decreases annually. The crop growth season in Ordos is from May to October. According to Figure 2(c) and 2(d), the groundwater table decreases during the crop growth season and then increases after harvesting. Therefore, we infer that the third and fourth patterns are caused by irrigation extraction (Ju et al. 2012). If the extraction is similar in magnitude to the recharge, the annual groundwater table is relatively stable and belongs to the third pattern. If the groundwater abstraction is larger than the recharge, the water table decreases annually, belonging to the fourth pattern.

Groundwater recharge occurs through precipitation and various human activities, such as irrigation recharging (Jasechko et al. 2014; Qi et al. 2023), while discharge is facilitated by evapotranspiration and pumping (Doble et al. 2006). In semiarid regions, precipitation plays a crucial role in the variability of GWD (Luan et al. 2023). To elucidate the impacts of meteorological droughts on GWD, we calculated the SPI as the meteorological drought indicator and analyzed the correlation relationship between SPI and GWD. The strong and significant correlations between SPI and GWD in more than 60% of monitoring wells in Ordos suggest that the variability of GWD is primarily influenced by meteorological conditions. Less precipitation affects groundwater resources by decreasing soil water infiltration (Han et al. 2021).

A weak but still significant correlation in some monitoring wells indicates that human activities disturb the propagation from meteorological droughts to GWD, particularly activities related to meteorological droughts, such as irrigation in NID and ecological restoration in MUSL. In the dry season, deficit precipitation increases water demand, leading to more groundwater extraction to alleviate water stress for crops, consequently lowering the groundwater table (Zhang et al. 2021a). However, in wet seasons, sufficient irrigation may not raise the groundwater table due to reduced water availability for recharge (Zomlot et al. 2015; Lorenzo-Lacruz et al. 2017). In ecology, densely populated artificial revegetation increases groundwater extraction and may intensify the groundwater drought in the dry season (Lu et al. 2018; Han et al. 2020) because vegetation roots extract more groundwater for transpiration, especially in dry years (Wu et al. 2016). Some studies demonstrate that revegetation can improve soil conditions and enhance infiltration, but groundwater is not recharged due to the transpiration with additional soil moisture in semiarid regions (Bartley et al. 2006; Teuling et al. 2013; Gao et al. 2015). In general, vegetation responds to water stress by partially closing the stomata to reduce transpiration, which limits the carbon uptake by photosynthesis (Peters et al. 2018). Over places with deficit and deep groundwater, drought results in the reduction of gross primary production (Zhang et al. 2022a). Shallow and sufficient groundwater can maintain the stomatal conductance opening and buffer the effects of drought on vegetation (Meinzer et al. 2016; Deng et al. 2022). Therefore, overexploitation of groundwater for irrigation or over-revegetation will decrease groundwater availability, which aggravates the response of vegetation and agriculture to drought (Wu et al. 2017). In addition, water in reservoirs can continuously recharge groundwater, disrupting the propagation from meteorological droughts to groundwater systems (Apurv et al. 2017).

There is still a part of monitoring wells where GWD has an insignificant correlation relationship with SPI, likely due to human activities unrelated to meteorological droughts, such as managed aquifer recharge projects. In the pursuit of sustainable development in Ordos, the local government artificially replenishes groundwater to strike a balance between exploitation and replenishment. Coal mining is also an anthropometric intervention for groundwater dynamics. Ordos is rich in coal resources, and Xie et al. (2018) found that large-scale mining may damage the stratum structure of the mining area and lead to groundwater loss. The lack of correlation between meteorological droughts and groundwater tables is also observed in some studies. For instance, Lorenzo-Lacruz et al. (2017) found that tourism pressure exacerbates groundwater exploitation during summer, leading to an insignificant correlation relationship between SPI and the standardized groundwater index. Wendt et al. (2021) found that managed aquifer recharge significantly reduced groundwater in both duration and magnitude. Another possible reason for the insignificant impacts of meteorological droughts on GWD may be the continuous overuse of groundwater (Wendt et al. 2021), where groundwater abstraction surpasses precipitation recharge, resulting in an increase in precipitation but a decline in the groundwater table. The intervention of human activities in groundwater is various. Irrigation and revegetation reduce the groundwater, coal mining changes the geological structure, and reservoirs and managed aquifer recharge projects change the natural process of groundwater recharge. These human activities complicate the groundwater dynamics, which enhances the difficulty of water resource prediction and management.

The lag time represents the speed of propagation of meteorological drought to groundwater systems (Schuler et al. 2022). The length of lag time that GWD responds to SPI is related to the depth of groundwater level. Throughout the region, the average GWD with a lag time of 0–3 months is much smaller than the average GWD with a lag time larger than 12 months because the surface water needs more time to recharge groundwater when the groundwater table is deeper (Schreiner-McGraw & Ajami 2021). Some monitoring wells exhibit abnormal GWD with a lag time of 0–3 months, closely resembling the average GWD with a lag time larger than 12 months. This abnormality may be caused by human activities related to meteorological droughts, such as irrigation. During meteorological droughts, farmers tend to extract more groundwater to alleviate water stress on crops. During such times, the GWD has no impact on the lag time of GWD responding to meteorological droughts. Similarly, ecological restoration in semiarid regions may also intervene in the influence of GWD on the lag time. In general, vegetation species used for ecological restoration are drought-tolerant, with well-developed root systems (Liu et al. 2021). The drier climate can trigger plants to extend their roots to abstract deeper groundwater and reduce water stress (Liu et al. 2021). Consequently, the lag time of GWD to meteorological droughts may be shortened.

Both SPI and the standardized precipitation evapotranspiration index (SPEI) serve as crucial indicators for meteorological droughts (McKee et al. 1993; Vicente-Serrano et al. 2010; Yao et al. 2018). Given the relatively small size of Ordos, it necessitates data with fine spatial resolution. The calculation of SPEI relies on the key variable of PET (Vicente-Serrano et al. 2010). There are three common equations for calculating PET, namely, the Penman–Monteith equation, the Thornthwaite equation (Thornthwaite 1948), and the Hargreaves equation (Hargreaves & Samani 1985). Among these, Thornthwaite and Hargreaves methods focus solely on temperature, while the Penman–Monteith equation considers multiple meteorological factors influencing PET, including wind speed, relative humidity, radiation, and air pressure (Adnan et al. 2021; Cao et al. 2022). Due to the lack of high-precision products of wind speed and relative humidity, the Penman–Monteith equation is not adopted in this study. Some studies have challenged the conventional belief that temperature is the primary meteorological driver influencing PET in semiarid regions. For instance, Liu et al. (2024) found that relative humidity is the most sensitive factor for PET in semiarid regions. Yin et al. (2021) found that the PET in the Chinese northwest arid regions is more sensitive to radiation, wind speed, and relative humidity than average temperature during the growth season. Therefore, relying solely on temperature for calculating PET could introduce uncertainties. These uncertainties may further accumulate into SPEI, potentially impacting the final analysis.

This study explores the variation of GWD in Ordos, the northern Yellow River Basin, and distinguishes the impacts of meteorological droughts and human activities on the GWD. There are four patterns of groundwater dynamics in Ordos, including continuous increased pattern, jagged declined pattern, fluctuated seasonally but annual stable pattern, and fluctuated seasonally but declined overall. We found that the GWD in Ordos is influenced by both meteorological droughts and human activities, and the GWD is influenced by meteorological factors at the majority of monitoring wells. The lag time of GWD response to meteorological droughts is related to the absolute magnitude of GWD, and most of the lag times are less than 3 months. Our results are helpful for local governments to manage human activities and implement measures to cope with droughts. For instance, predicting meteorological factors accurately at seasonal to annual scales can increase the prediction skill for groundwater dynamics and prepare drought mitigations in advance. Furthermore, promoting water-saving irrigated agriculture and revegetating moderately is a practical way to avoid overexploitation of groundwater. Nevertheless, our study is limited by data availability, making quantification of human activities challenging. Furthermore, GWD may be influenced by multiple human activities, and how to isolate the impact of each human activity on GWD deserves future exploration.

This study was financially supported by the Inner Mongolia Autonomous Region Water Conservancy Research Special Project (Grant No. NSK202301), the Ordos Major Science and Technology Projects (Grant No. ZD20232302), and the National Natural Science Foundation of China (Grant No. 42361144709). We appreciate the valuable contributions of anonymous reviewers, whose constructive and critical comments and suggestions greatly improved the paper's quality.

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

The authors declare there is no conflict.