The study area (in Ar Horqin County, Chifeng City, Inner Mongolia) is located in the Horqin area of northern China’s agro-pastoral ecotone (Figure 1A). With the support of center-pivot irrigation (CPI) for the development of livestock grazing, this area has been known as “China’s grass industry center” in recent years, and has shown the fastest NDVI (normalized difference vegetation index) growth rate around the Horqin area (Figure 1B). It has thousands of CPI systems concentrated in an area of approximately 1,000 km² (from 43°22′ N to 43°50′ N and 120°10′ E to 120°55′ E) that lies between the Uljimulun River, the Xilamulun River, and the West Liao River (flow from southwest to northeast, Figure 1C). Depending on the size of the CPI (its radius), the irrigated area of each circular or fan-shaped plot can range from about 4.5 to 75 ha (Figures 1D–F). Several forage species are cultivated in rotation in the irrigated area: Avena species Linn. and Medicago sativa Linn., either single or mixed in each plot (Figure 1G).

The study area has a temperate continental semi-arid monsoonal climate, with a mean annual temperature of 5.5°C (mean monthly temperatures ranging from -12.8°C in January to 23.4°C in August) and total annual precipitation of 368 mm (www.resdc.cn), of which 75% falls during the growing season from June to September. The total annual potential evaporation is 1800–2000 mm (Zhao et al., 2009). The elevation ranges from 259 to 367 m asl (Jarvis et al., 2008).

Geologically, the study area belongs to the subsidence zone from the Greater Khingan Mountains to the western Songliao Plain of China, and the paleosol and aeolian sand form interlayers. The underlying strata are loose alluvial and aeolian sediments of Quaternary, mainly medium and fine sand (diameter of 0.10–0.50 mm) in the various cracks with relatively high-water content or reserves. The thickness of the sedimentary layer varies from about 100 to 200 m (Qiu, 1989). The dominant zonal soils of this area are Arenosols mainly degraded from Kastanozems and Chernozems based on the taxonomy of the World Reference Base for Soil Resources (IUSS Working Group WRB, 2006).

Hydrologically, water resource is relatively rich mainly from the Yanshan Mountains and the Greater Khingan Mountains. The large part of available groundwater resource is shallow phreatic water in the aeolian sediments (Qiu, 1989), and recharged by precipitation, river runoff. The groundwater discharge is mainly to artificial wells for irrigation and daily use (Zhong et al., 2018).

The groundwater level in wells was measured using a water-stage recorder. The recorder is a 30-m steel ruler, wrapped with electric wire in a plastic shell and coiled around a hand wheel. A 10-cm stainless-steel probe is connected at the start of the tick marks on the ruler. The weight of the probe keeps the ruler perpendicular while it is inside the well. When the probe touches the water surface, a buzzer on the hand wheel sounds; when the probe is removed from the water, the buzzer stops immediately. The tick marks on the ruler from the ground surface represents the depth of the groundwater level. However, most of the gaps between the well edge and the pumping pipe of a CPI system are too narrow for the probe to enter or to reach the water surface inside the well, so it was not possible to measure in all wells. Although there are thousands of wells in this area, we were only able to find 42 irrigation wells at representative locations where we could monitor GL spatially from April 2019 to April 2022, including both CPI wells and conventional irrigation wells that without CPI system (Figure 1C). During our monthly monitoring in 2019, we found it difficult to measure all the target wells at normal level in same period (t) by excluding the lag effect of pumping, because there were sharp decreases which corresponded to GL measurements after an irrigation event. Since 2020, we have only monitored GL at the beginning of the irrigation period (in early April) and at the end (in Mid-October). We calculated the changes in GL (ΔGL) during discharge (irrigation) and subsequent recharge as follows: Δ G L discharge = G L t 2 − G L t 1 (1) Δ G L recharge = G L t 3 − G L t 2 (2) where Δ G L > 0 indicates groundwater discharge (irrigation) and decreased GL and ΔGL < 0 indicates groundwater recharge and increased GL; t ₁ and t ₂ represent the beginning and end of the monitoring in a given year, and t ₃ represents the beginning of the following year. Therefore, t ₁ to t ₂ is the groundwater discharge (irrigation) period, t ₂ to t ₃ is the groundwater recharge period. The 42 (N) measuring wells were divided into three groups according to the normal GL: shallower (n1 = 9), <10 m; medium (n2 = 19), >10 m and <20 m; deeper (n3 = 14), >20 m.

The irrigation method in this study area is basically unified, i.e., the automatic sprinkler irrigation through tube wells of CPI systems. Thus, density of irrigation well can serve as a reasonable proxy for irrigation intensity. In addition, the change in cropping intensity (Duarte and Mateos, 2022) from natural to artificial ecosystems can validate the increased irrigation intensity.

Based on the satellite images, Table 1 showed that the implementation of CPI systems went through an exploratory stage from 2005 to 2012, followed by rapid expansion, reaching great peaks in 2013, 2015, and 2017, respectively, with 539, 308, and 573 new wells added, accounting for 88.9% of the total number during the study period. Figure 2 showed the changes in the spatial pattern of CPI wells over four 5-year periods. The land use type was natural grassland with few irrigation facilities before 2005 (Figure 2A and Table 1). In 2005, CPI started at only two locations, with well densities of 0.16 and less than 0.10 km⁻² (Figure 2B). From 2005 to 2010, CPI still developed slowly with the highest well density (0.64 km⁻²) in the middle of the study area (Figure 2C). Then, the density increased rapidly between 2013 and 2015, from 0.64 to 2.52 wells km⁻² and formed multiple hotspots (Table 1 and Figure 2D). From 2015 to 2020, the multiple hotspots combined to form contiguous CPI area in the southern part of the study area, and two large centers (>1.01 and 2.31 wells km⁻²) appeared in the northern part (Figure 2E). Thus, the gaps were shrinking between CPI wells as the area of higher well density expanding.

The GL trends of the three groups were basically the same during the monitoring time from 2019 to 2022 (Supplementary Figure S1). The changes in ΔGL indicated that the discharge of groundwater accelerated as the normal GL decreased, because the averaged ΔGL in the deeper group (0.94 m for GL > 20 m) was significantly higher than those of the shallower group (0.64 m for GL < 10 m) and medium group (0.73 m for 10 m < GL < 20 m) (Figure 3A). During groundwater recharge period, the averaged ΔGL in the deeper (−0.29 m) and medium groups (−0.29 m) were significantly higher than that in the shallower group (−0.12 m) (Figure 3B). Combined the two periods (Figures 3A,B), it revealed that the deeper group showed a significantly greater net decrease (0.63 m yr⁻¹) than the shallower group (0.38 m yr⁻¹) and medium group (0.43 m yr⁻¹) (Figure 3C).

Regression kriging was used to predict the spatial patterns of ΔGL and GL, which were significantly and positively (R ² = 0.447 and 0.429, p < 0.001, Supplementary Table S1) correlated with the well density (kernel density) of CPI system, respectively. It showed a fitted plane function in the 3D coordinate system based on the variables (R ² = 0.655, p < 0.001, Figure 4). The predicted trend of GL was consistent with the observed change, and indicated a significant positive linear relationship between ΔGL and GL (R ² = 0.612, p < 0.001, Supplementary Table S1). Therefore, ΔGL was determined by the both well density and the depth of GL.

Cropping intensity represents the combination of agricultural water use and land use patterns, which can directly associate with irrigation intensity. It is determined by agricultural policies, social-economic factors, and farmers’ subjective wishes (Duarte and Mateos, 2022). As more than 60% of the water area has shrunk in the Horqin area (Tao et al., 2015), the use of well irrigation has expanded rapidly since 2000 and the groundwater-based irrigation is increasing year by year to make up for the lack of rainfall and surface water (Zhong et al., 2018). A 2013 questionnaire survey in the middle of the agro-pastoral ecotone (around the Yinshan Mountains) showed that 31% of the local farmers had turned dry land into irrigated land (Wang et al., 2020).

Stable water availability is considered as the most critical guarantees for crop production (Auci and Vignani, 2021), particularly in agricultural areas with frequent drought (Bryan et al., 2013; Khan et al., 2021). Our kernel density results suggested that the implementation of high-density, high-efficiency, and large-scale CPI systems initiated a new mode of land-use change. Both the regional NDVI and quadrat-based survey indicated that the cropping intensity was significantly enhanced (p < 0.01) after implementation of CPI systems, as shown between natural grassland and cultivated pastures (Figures 5A,B and Table 2). The NDVI _CP was significantly higher than the NDVI _NG (p < 0.01), with an increase of 30.6%–59.2%. The aboveground biomass of Avena species (503.6 ± 40.7 g m⁻²) and Medicago sativa (270.0 ± 7.6 g m⁻²) were significantly higher (p < 0.01) than that of natural grassland (182.2 ± 11.1 g m⁻²) by 176.4% and 48.7%, respectively. Due to the difference in cropping intensity, the CPI area was clearly separated from the non-irrigated area, because the NDVI _slope of non-irrigated area was much lower than that of CPI area, even showing a negative trend of slight degradation (Figure 5C).

GL is generally affected by the dynamic balance between the recharge and discharge to aquifer storage, including natural (Liu et al., 2022) and anthropogenic factors (Zencich et al., 2002; Seeyan et al., 2014). The causes of ΔGL are always complex from shallow to deep aquifers, including rainfall and evapotranspiration (Bai et al., 2017), irrigated area and cost (Foster et al., 2015), and effectiveness of water-saving approaches and sustainable water management (Malki et al., 2017; Salem et al., 2017). Sustainable use of irrigation water aims to increase crop yields with less water input by improved water management efficiency (Chartzoulakis and Bertaki, 2015). In our study area, the observed ΔGL was mainly caused by human activities. It was consistent with the trend (1975–2000) that had been reported for the Hebei plains in northern China, where the ΔGL ranged from 0.42 m yr⁻¹ to more than 1.21 m yr⁻¹, respectively, in shallow and deep aquifers (Duan and Xiao, 2003). The ΔGL at different depths suggested that shallow groundwater may be relatively susceptible to replenishment by rainfall and subsurface flow (Xie et al., 2011; Gao et al., 2015), while deep groundwater was more sensitive to water withdraw for irrigation (Russo and Lall, 2017).

Our study is also a typical case that related in water–energy–food nexus. The direct consequence of groundwater depletion is the increase in energy consumption for pumping. As more than 1,500 pumps have different power ratings and working hours, we roughly estimated the energy consumption for irrigation (14,521.7 J on average) in terms of change in gravitational potential energy (ΔGPE) of water (each 100 kg equals to 100 mm water m⁻², Figure 6A). Excessive density of CPI wells (deeper GL) not only led to a dramatic increase in the risk of unsustainable water use, but also increase energy consumption at a rate of over 800 J yr⁻¹ per 100 kg water (Figure 6B). Left unchecked, underpowered submersible pumps will have to be replaced, with irreversible effects on irrigation costs.

According to our results (Figure 4), if the well density remains unchanged, and the depth of GL increase by 10 m, the ΔGL will increase by about 0.19 m yr⁻¹; if GL is maintained at the same depth and the well density increases by 1.0 count km⁻², the ΔGL will increase by about 0.06 m yr⁻¹. However, a limitation of our results is that it can only interpret the ΔGL pattern in recent years; whereas the current status of regional depth of GL should be the result of well density and its long-term changes. Thus, some studies believe that an appropriate water price can prompt users to conserve water and use water-saving technologies (Auci and Vignani, 2021). At the same time, excessively dense wells should be adjusted (e.g., kept within 0.5 count km⁻²). Compared to Avena species, Medicago sativa (perennial) should be recommended to reduce both water consumption and soil tillage. Water use mode in the past should be changed by new irrigation schedule, and further research is needed on water use efficiency of cultivated pasture.