Effects of Groundwater Pumping on Ground Surface Temperature: A Regional Modeling Study in the North China Plain
Abstract
Overexploitation of groundwater (GW) in the North China Plain (NCP) since the 1960s has many environmental consequences. However, mechanistic understanding of this perturbation remains limited, particularly at the regional scale. In this study, the coupled ParFlow.CLM model representing subsurface and land surface processes and their interactions was applied in the NCP at high spatiotemporal resolutions. The model was validated using the water and energy fluxes reported in previous studies and from the JRA-55 reanalysis. Numerical experiments were designed to examine the impacts of GW pumping and irrigation on the ground surface temperature (GST). Results show significant effects of GW pumping on GST in the NCP. Generally, the subsurface acts as a buffer to temporal variations in heat fluxes at the land surface, but long-term pumping can gradually weaken this buffer, leading to increases in the spatiotemporal variability of GST, as exemplified by hotter summers and colder winters. Considering that changes of water table depth (WTD) can significantly affect land surface heat fluxes when WTD ranges roughly between 0.01 and 10 m, the 0.5-m/yr increase of WTD simulated by the model due to pumping can continue to increase the regional averaged WTD and hence, perturb GST for about 20 years after GW pumping began in the NCP, before WTD exceeds 10 m. The variations of GST are expected to increase faster initially and gradually slow down due to the nonlinear behaviors of GST with WTD. The findings from this study in the NCP may also have implications for other regions with GW depletion.
Key Points
- Effects of groundwater pumping on GST in the North China Plain are investigated using an integrated hydrologic model
- Groundwater pumping increases the spatiotemporal variability of GST and results in hotter summers and colder winters
- The effect of groundwater pumping on GST may last for 20 years, with a stronger effect in the beginning due to nonlinear response
1 Introduction
Ground surface temperature (GST) and soil temperature (ST) have a significant influence on terrestrial processes such as ecosystem functions, surface-subsurface interactions, and land-atmosphere interactions, with implications for the terrestrial system in response to climate change (H. Zhang et al., 2016). For example, ST controls the growth of vegetation, which is extremely sensitive to temperature, so it is a major factor influencing crop yield in agricultural regions (H. Zhang et al., 2016). The increase in ST has accelerated soil respiration. A 20% increase in soil respiration corresponds to a release of about 14- to 20-Pg C/yr soil carbon, which is 2–3 times the amount of carbon released from fossil fuel and land use change (7 Pg C/yr) (Davidson et al., 2006; Lloyd & Taylor, 1994). It is noted that GST is the ST of the top soil layers (an average of ST in the 0.1-m-thick top soil layer in this modeling study).
Groundwater (GW) is an important source of fresh water in many populated regions such as the North China Plain (NCP) (Figure 1), which is the political, economic, and agricultural center of China. Overexploitation of GW in the NCP since the 1960s, mainly for irrigation, has many environmental consequences, such as river deterioration, land subsidence, and seawater intrusion (C. M. Liu et al., 2001). Water table depth (WTD) in most places of the NCP is now more than tens of meters, much deeper than 0–3 m during the 1960s, with some regions such as Beijing, Shijiazhuang, and Cangzhou showing WTD of more than 100 m (Cao et al., 2013). Meanwhile, in a warming climate (Kang & Eltahir, 2018), the annual average air temperature in the NCP has increased by 0.23 °C per decade, which is slightly higher than the global average during the same period (Dou et al., 2016). The rising air temperature could lead to the changes in GST (Pollack et al., 2005).
Figure 2 shows the variations of GST and WTD in the NCP for the last 40 years. With increasing trends in both GST and WTD, an important question is whether the GW withdrawal may have contributed to the increasing GST, although the latter has generally been attributed to the increasing air temperature (Pollack et al., 2005; H. Zhang et al., 2016; T. Zhang et al., 2001). Recent studies addressing the key role of GW in the subsurface-land surface-atmosphere system (Keune et al., 2016; Maxwell & Condon, 2016; Taylor et al., 2013) noted the important control of GW on soil moisture that governs the land surface energy fluxes, with subsequent influence on local weather and climate through land-atmosphere interactions (Ferguson & Maxwell, 2012). However, few studies have looked at the impact of human activities such as GW pumping or irrigation on GST at the regional scale. During the rapid socioeconomic development over the NCP in the past few decades, GW pumping has been extensively utilized to overcome shortages in fresh water resources (Cao et al., 2013; C. M. Liu et al., 2001).
GW movements and pumping in the NCP has been examined by numerous studies using mostly GW models such as MODFLOW (Cao et al., 2013; Cui et al., 2009; Hu et al., 2010; Jia & Liu, 2002; J. Liu et al., 2008; Wang et al., 2008; Xue et al., 2010; X. Zhang, 2007; X. Zhang et al., 2008). These models were applied at the local scale focusing on the water budget and did not study GW feedbacks to land-energy budget. Though Zou et al. (2015) studied the effects of GW exploitation on land surface processes at the Haihe River Basin located in the NCP, the GW component was conceptualized as a bucket model, and GW pumping was simply added to the water balance calculation. This oversimplification of GW dynamics in land surface models has been recognized in the last decade, highlighting the lack of representations of lateral GW flow and heat transport (Bisht et al., 2018; Fang et al., 2017; Zeng et al., 2018), which may limit our ability to model not only land surface processes but also land-atmosphere interactions.
Recent progress has been made in development of coupled land surface-subsurface models (Alkhaier et al., 2012; Davison et al., 2015; Maxwell et al., 2011; Maxwell & Miller, 2005; Rahman et al., 2015). In addition, high-resolution regional or global maps of hydraulic parameters for soils and deep aquifers have become available (Gleeson et al., 2014; Gleeson et al., 2011; Y. G. Zhang et al., 2018) to facilitate integrated modeling at large scales, such as ParFlow.CLM over most of the continental United States (Maxwell & Condon, 2016) and TerrSysMP over the European CORDEX domain (Keune et al., 2016). Using an integrated model ParFlow.CLM in the medium-sized Little Washita Basin, Ferguson and Maxwell (2011, 2012) studied the effect of irrigation and pumping on land surface water and energy fluxes and compared this effect with that of climate change. Subsequently, in the same basin, Condon and Maxwell (2014a, 2014b) studied the system dynamics with a spatiotemporal framework under managed irrigation by coupling ParFlow.CLM with an additional Water Allocation Module. Most recently, Condon and Maxwell (2019) studied the sensitivity of evapotranspiration and streamflow to GW depletion over most of the continental United States. However, few studies have analyzed the regional GST dynamics under GW pumping using integrated modeling approach.
The overarching goal of this study is to explore the possible effects of GW pumping on land surface processes, especially GST, in the NCP using ParFlow.CLM. The model was first set up over the NCP then used to study the effects of GW pumping and combined pumping and irrigation (P&I) on water and energy cycles, with possible effects on the warming trend in the NCP. In what follows, we introduced the numerical experiments based on ParFlow.CLM and evaluated the baseline scenario by comparing the simulated WTD, GST, sensible heat flux (H), and latent heat flux (LE) with results from previous studies and the JRA-55 reanalysis data products. These variables were selected due to their important roles in land surface-subsurface interactions. Different scenarios were also quantitatively compared to characterize uncertainties in the modeling. Then the effects of P&I on WTD and GST were explored. Finally, implications and limitations of this study were summarized.
2 Integrated Modeling in the NCP
2.1 Model Description: ParFlow.CLM
ParFlow.CLM (Maxwell & Miller, 2005) is an open source, integrated land surface and subsurface model (https://github.com/parflow/parflow) developed by coupling a modified version of the Common Land Model (CLM) (Dai et al., 2003; Ferguson et al., 2016; Jefferson et al., 2017; Jefferson & Maxwell, 2015; Maxwell & Condon, 2016) and an integrated surface-subsurface flow model (ParFlow) (Ashby & Falgout, 1996; Jones & Woodward, 2001; Kollet & Maxwell, 2006). ParFlow represents the variably saturated subsurface flow by solving the three-dimensional Richards's equation and integrates it with the overland flow by solving the two-dimensional kinematic wave equation and a free-surface boundary condition (Kollet & Maxwell, 2006). Replacement of the original one-dimensional vertical flow in CLM by ParFlow overcomes both the shortcoming of the lower flow boundary (e.g., the free drainage) and the limitation in simulating lateral subsurface flow in CLM (Keune et al., 2016; Maxwell & Miller, 2005). At the same time, coupling of ParFlow with CLM improves the simple top boundary used in traditional GW model like ParFlow in which snow, surface runoff, soil heating, and root-zone uptake processes are oversimplified or neglected (Maxwell & Miller, 2005). Therefore, ParFlow.CLM can capture more realistic water-energy interactions in the subsurface-land surface system. The massively parallel computing capability and terrain following grid in ParFlow make it possible for ParFlow.CLM to be applied at large scales (Condon & Maxwell, 2014a). More details on ParFlow.CLM can be found in many previous studies (Kollet & Maxwell, 2008; Maxwell & Miller, 2005). ParFlow.CLM has been applied to more than a dozen watersheds around the world including the Big Thompson (CO), Klamath (OR), Little Washita (OK), Rur (Germany), San Joaquin (CA), Sante Fe (FL), Chesapeake (MD), and Skjern (Denmark) catchments (Maxwell & Condon, 2016), but it has not been applied and evaluated in the NCP before.
(1)
(2)
(3)2.2 Modeling Domain
The modeling domain (34.45–41.00°N, 112.73–117.27°E) in this study is shown in Figures 1 and 3. The total area is about 467,274 km2, covering the Beijing Municipality, part of the Tianjin Municipality, and part of the Hebei, Henan, and Shandong Provinces (Figures 1 and 3a). The elevation in the study domain ranges from about 3,000 m in the northwest Taihang Mountain to near the sea level in the east near the Bohai Sea (Figure 3b, Bohai Sea not shown). The current modeling domain was mostly adopted from the commonly defined NCP region (Cao et al., 2013; Cao et al., 2014; Qin et al., 2013) but slightly modified to simplify the implementation in ParFlow.CLM by using a rectangle to cover the largest area of the previously defined NCP. The most active pumping areas, along the Taihang Mountain from Shijiazhuang to Beijing, are well located in the center of the modeling domain, so the boundary effect on the simulation results should be negligible due to the size of the buffer zone (Keune et al., 2016). Since the boundary conditions are not specified at the subsurface topographic boundaries, a rectangular domain was used to move the boundaries away from the study region of interest. Additionally, meteorological records showed that, from 1980 to 2011, the mean annual precipitation and air temperature in the NCP was 525 mm and 13.1 °C, respectively (Pei et al., 2015).
2.3 Subsurface Grid
The subsurface in the study region was divided into 439 × 799 grid cells in the horizontal direction with a resolution of about 1 km. This resolution is higher than the 2-km resolution used in previous regional modeling of the NCP (Cao et al., 2013; Qin et al., 2013), and such a high resolution is necessary to model the effects of well pumping. The cone of depression with WTD increasing from the edge to the center would be averaged to the same WTD if large grid cells are used, which may influence the modeling of land surface energy variations that are dependent on WTD (Taylor et al., 2013). With the terrain following grid (Maxwell, 2013), the subsurface was divided into five layers in the vertical direction. The thickness of the layers is 0.1, 0.3, 0.6, 1, and 100 m from top to bottom. Thus, the total number of grid cells in the modeling domain is 1,753,805. The setting of vertical thickness is reasonable since the subsurface in the NCP is mainly composed of shallow and deep aquifer groups (Cao et al., 2013), which are relatively independent in terms of hydraulic connections (Gao, 2008). The shallow aquifer group is 160 m deep on average (Cao et al., 2013), coinciding with the total thickness of the five layers used in the current study. Only the shallow aquifer group was modeled in this study since it is heavily disturbed by human activities (i.e., P&I). This configuration is also consistent with those used in previous studies that delivered satisfactory results in the continental United States (Maxwell et al., 2015; Maxwell et al., 2016; Maxwell & Condon, 2016).
2.4 Data Descriptions
Topography in the NCP (Figure 3b) was described by the digital elevation model (DEM) of 90-m resolution from the Shuttle Radar Topography Mission (Rabus et al., 2003). Topographic slopes (Sx and Sy) were calculated based on the DEM and then adjusted by the watershed analysis tool in the GRASS Geographic Information System. The slopes were further modified by the parking lot tests (Bhaskar, 2010) and by the subsequent spin-up processes to ensure connectivity of the streams. Land cover data (Figure 3c) following the classification of the International Geosphere-Biosphere Program from the Global Land Cover Characterization database were downloaded from the USGS website (https://www.usgs.gov/centers/eros). Soil textures (Figures 3d and 3e) in the top four layers were from a newly developed global soil map of 1-km resolution (Y. G. Zhang et al., 2018). From northwest to southeast in the modeling area, soil texture varies from loamy sand or loam in the mountain areas to loam, silty loam, or silty clay in the plains. Permeability of the bottom layer (Figure 3f) was from Gleeson et al. (2014) with higher resolution than that in Gleeson et al. (2011). This deep aquifer is composed of sedimentary rocks (e.g., carbonate rocks) in the mountain areas and Quaternary sediments (e.g., sand, silt, clay, and silty clay) in the plains (Cao et al., 2013). Meteorological forcing data were obtained from the Japanese 55-year Reanalysis project (JRA-55). The forcing data were interpolated to the modeling grid. The 3-hourly forcing data were also linearly interpolated to hourly resolution. All the input data were reprojected to the same Mercator projection coordinate system and resampled, if necessary, to the same 1-km resolution.
2.5 Boundary and Initial Conditions
We used the free-surface overland flow boundary condition for the land surface and no-flow boundary condition for all other boundaries. No-flow assumption for the lateral and bottom boundaries is reasonable since the buffer zone around the pumping area is large enough, and the shallow aquifer group in the NCP is hydraulically independent from the deep aquifers. For the subsurface (i.e., the ParFlow model), a constant infiltration of 1 × 10−4 m/hr equivalent to the annual precipitation in the NCP and an initial condition of water table at 2 m below the land surface were set. The ParFlow model was spun up until the spatial distribution pattern of pressure head was in quasi-equilibrium. Then the coupled ParFlow.CLM model was spun up for another 2 years (Maxwell & Condon, 2016). After the model spin-up, 1-year simulations were performed for Scenarios 1–3, and two more years were simulated for Scenarios 4–5 for the P&I experiments. Forcing data in 1970 were used to represent the predevelopment condition since the extensive GW pumping began in the 1970s. An hourly time step (e.g., Maxwell & Condon, 2016) was used in the 1- and 2-year simulations producing daily output for analysis.
2.6 Scenario Setup
The model configuration used the best publicly available data and serves as the baseline (Scenario 1) in this study. Two additional scenarios (Scenarios 2 and 3) were set up by considering the uncertainty from the subsurface heterogeneity (Table 1). Heterogeneous permeabilities were from Y. G. Zhang et al. (2018) in the top soil and from Gleeson et al. (2014) in the deep aquifer. “Homogeneous” refers to the uniform subsurface with a hydraulic conductivity of 0.6958 m/hr, which is the geometric mean of the hydraulic conductivities in the shallow aquifers of the NCP from Cao et al. (2013). In Scenario 3, anisotropy was also considered using a ratio of horizontal to vertical hydraulic conductivity of 10,000 (Cao et al., 2013).
| Scenario | Permeability | Pumping | Irrigation |
|---|---|---|---|
| 1 (baseline) | Heterogeneous | No | No |
| 2 | Homogeneous | No | No |
| 3 | Heterogeneous and anisotropy | No | No |
| 4 | Heterogeneous | Yes | No |
| 5 | Heterogeneous | Yes | Yes |
After the model was evaluated based on Scenarios 1–3, two more Scenarios 4 and 5 (Table 1) with pumping and P&I were simulated based on the setting of scenario 1. GW pumping information in the NCP for Year 2001 (Li, 2013) was used and summarized in Table 2. This information represents the pumping intensity around Year 2000 in the NCP. All the agricultural water in Table 2 was assumed to be used for irrigation only in this study. In this study, GW P&I were only conducted in the NCP, that is, outside the purple area in Figure 3a. In each administrative region (Figure 3a), the water volumes of total pumping and agricultural water (Table 2) were converted to water depth per unit time (m/hr for pumping while mm/s for agricultural water) by dividing the area of that administrative region. Pumping was conducted by adding negative source terms in grid cells in the bottom layer while irrigation by adding agricultural water to precipitation in the forcing data. The P&I rates are constant with time in this study. The spatial and temporal variations of P&I due to scheduling in the NCP can be considered in future studies. In Scenario 4, additional double and half of the pumping rates were also considered. Increasing rates represent the increasing GW demand in the future, while decreasing rates represent the possible effect of smart water management or hydraulic projects, such as the South-to-North Water Diversion Project in the NCP.
| Administrative regions | Total pumping (108 m3/yr) | Area (104 km2) | Pumping rate (m/hr) | Agricultural water (108 m3/yr) |
|---|---|---|---|---|
| Beijing | 26.44 | 1.04 | 2.89 × 10−5 | 13.38 |
| Tianjin | 1.06 | 1.76 | 6.87 × 10−7 | 0.99 |
| Hebei | 118.53 | 11.41 | 1.19 × 10−5 | 99.81 |
| Henan | 28.04 | 2.63 | 1.22 × 10−5 | 23.36 |
| Shandong | 15.93 | 4.79 | 3.79 × 10−6 | 10.10 |
3 Results and Discussion
3.1 Model Evaluation
3.1.1 WTD
The simulated annual average WTD in 1970 in the baseline is shown in Figure 4a. Simulated WTD is generally 1 m in the plains but reaches more than 60 m in the mountain areas. WTD in the plains of the NCP during predevelopment was estimated at 0–3 m (Cao et al., 2013; Fei et al., 2009). Thus, the simulated WTD is within the range of the predevelopment condition. Compared to Scenario 2 (Figure 4b), WTD is more controlled by the DEM for a homogeneous subsurface, which, however, is adjusted by the subsurface heterogeneity in Scenario 1. Although WTD in mountain areas is generally large, it can be much smaller if the aquifer has low permeability (e.g., area indicated by the red circle in Figure 4a).
Some previous studies considered either an exponential decay of permeabilities with depth (Jiang et al., 2009) or a vertical anisotropy (Cao et al., 2013). Scenario 3 with vertical anisotropy has much smaller WTD (Figure 4c). Scenario 1 with homogeneous permeability in the vertical direction also shows a shift of WTD to 0 m although it ranges between 0 and 3 m, indicating that the parameters in Y. G. Zhang et al. (2018) and/or Gleeson et al. (2014) were likely underestimated. Shallower WTD was also estimated in previous studies (Fan et al., 2013; Maxwell et al., 2015), which suggested that the bias may be mainly due to GW pumping that was not considered in the simulation. Similarly, GW pumping already happened in the 1970s in the NCP, which might be another reason for the shallower WTD simulated in scenario 1 compared to the observations. In general, variations of WTD, decreasing from piedmont plain to coastal plain, are well captured in the simulation.
3.1.2 GST and Surface Heat Fluxes
The simulated spatial distribution of annual average GST in 1970 (Figure 5a) is consistent with that of the JRA-55 reanalysis data (Figure 5d). GST decreases from about 288 K in the south to about 278 K in the northwest. There is a clear difference in GST between the plains and mountain areas due to topography. Seasonal cycle of the simulated GST also matches the trend of JRA-55 (Figure 6a). The capability to capture the temporal and spatial variations of GST supports the use of the model to study the effects of GW pumping on GST in the NCP. Spatial distribution of the simulated sensible (H) and latent (LE) heat fluxes are shown in Figure 5. A narrow band with higher H is found in the transition zones between the mountain areas and plains. LE has a more uniform distribution, with slightly higher values in the northeast. These spatial patterns are comparable with those of JRA-55, which are also shown in Figure 5. Forcing data with higher resolution can be considered to improve the distribution of LE. Additionally, the temporal variability of H and LE is quite consistent to that of JRA-55 (Figure 6).
3.1.3 Variations of GST With WTD
The ~1-m WTD in the plains lies in the critical WTD range (1–10 m) proposed by Kollet and Maxwell (2008) and Ferguson and Maxwell (2011), in which GW dynamics have larger influence on land surface processes. Selected land surface variables for this purpose in previous studies are mostly LE and ET (Condon et al., 2013; Kollet & Maxwell, 2008; Maxwell & Condon, 2016) while GST has been rarely discussed. Hence, variations of GST with WTD during predevelopment (baseline) were plotted in Figure 7 to reveal the sensitivity of GST to WTD in the NCP. For the whole modeling area (Figures 7a–7g), higher variability of GST can be found in areas with smaller WTD (<1 m) relative to that in areas with larger WTD (>10 m). There is an obvious transition of GST within the WTD range of 1–10 m. In addition, in plains (Figures 7e–7h) with WTD of ~1 m, where the P&I will be performed, GST shows increasing trends with WTD for all land cover types. Hence, either the whole modeling area or the plains show the sensitivity of GST to WTD in the NCP. The sensitive WTD range (0.01–10 m) in the NCP is basically consistent with previous studies (e.g., Ferguson & Maxwell, 2011) with a shift to smaller WTD. Following this argument, long-term GW pumping increasing WTD from ~1 m during predevelopment to more than 10 m today, which is exactly across this sensitive range, may have greatly altered the water and energy budgets in the NCP.
3.2 Effects of GW P&I
3.2.1 WTD
In section 3.1, evaluation of the simulated water and energy components and identification of the critical depth range for GST indicate the feasibility and the necessity to conduct GW pumping and P&I using the ParFlow.CLM model. Increase of WTD (ΔWTD) after 1 and 2 years of pumping is shown in Figure 8. The most obvious increase occurs in Beijing, western Hebei, and Henan, while smaller increase in WTD is found in Tianjin and Shandong due to smaller pumping rates (Table 2). However, WTD in northern Hebei increases less than other parts of Hebei (red circles in Figure 8), which could be due to the lateral flow recharge from Tianjin and Shandong nearby with less pumping. Besides, the Taihang Mountain next to this area with higher permeability (Figure 3f) might also induce lateral flow from mountain area toward the east. The spatial distribution of WTD can be confirmed by field observations (Figure S1 in the supporting information) (Li, 2013). In Figure S1, WTD is larger in Beijing and along the Taihang mountain, then decreases toward Shandong and Tianjin in the east, and is relatively smaller in northern Hebei. These consistencies between the modeling results and observations further demonstrate the accuracy of the model.
WTD in the plains increases by less than 0.5 m after 1 year of pumping and by almost 1 m in some places such as Beijing after 2 years of pumping. The rate of WTD increase (~0.5 m/yr) is higher than that reported (~0.3 m/yr) in Cao et al. (2013) because they calculated an average rate for the whole NCP without considering the spatial variability as shown in Figure 8. In addition, ignoring irrigation might have exaggerated the impacts of pumping on WTD in our simulation. However, considering the uneven spatial distribution of the cone of depression, the rate could also be underestimated by our model. For example, the WTD increase for the cone of depression in Baoding from 1975 to 1985 is more than 1.6 m/yr (Li, 2013). With irrigation, WTD increase became slow and the maximum increase of WTD after 2 years of simulation was about 0.5 m, which is more consistent with Cao et al. (2013). In general, the spatial and temporal variations of WTD are reasonable in the P&I experiments.
3.2.2 GST
Change of monthly average GST (ΔGST) after 1 year of pumping is shown in Figure 9 and that after 2 years of pumping is shown in Figure S2. Notably, obvious changes of GST only occurred in the area with pumping (a copy of Figure 9 showing the whole modeling area is in the supporting information as Figure S3). GST in the plains is less disturbed by pumping from December to February. GST increases from April to July mainly in Beijing and northern Hebei, while it decreases from September to November in Beijing, western Hebei, and Henan. Taking May and November as examples for summer and winter time, respectively, statistical analysis of ΔGST (presented in Figure 9) is given in Figures 10 and 11, in which further increase in summer and decrease in winter of GST due to the second year of pumping are also observed. Though such variations are basically consistent in space if comparing figures for the whole modeling area, Beijing, and Hebei (Figures 10 and 11), detailed spatial variations are still observed in frequency distributions across areas. In contrast, there is more significant spatial variability for the change of annual average GST (Figure 12). In Beijing and northern Hebei, it shows larger increase of GST in summer than decrease of GST in winter, leading to an increasing annual average GST (Figures 12a and 12b). Taking Beijing as an example, the spatial average and maximum increase of monthly average GST in summer (May to July) are 0.23 and 1.06 °C respectively; while the spatial average and maximum decrease of monthly average GST in winter (September to November) are 0.11 and 0.69 °C respectively, whereas in southern Hebei along the Taihang mountain, the increase of GST is not significant in summer (Figure 9), which leads to the decreased annual average GST (Figures 12c and 12d). In summary, due to the perturbations of GW pumping, monthly average GST exhibits highly temporal variability while annual average GST shows more significant spatial variability.
H. Zhang et al. (2016) reported an increase in annual average GST of 2.07–4.04 and 0.66–2.21 °C in northern and southern China (1962–2011), respectively. Due to 1 year of pumping, the maximum increase of monthly and spatially averaged GST in summer is over 1 °C, suggesting a potentially significant contribution of pumping to the reported GST change, such as in Beijing and northern Hebei as previously discussed. An increase in GST of about 2 °C for cropland associated with WTD ranging from 2 to 5 m was reported by Kollet and Maxwell (2008). In addition, the increase in GST in a 1-year simulation was 1–3 °C as reported by Maxwell and Kollet (2008) under prescribed hot and dry climatic condition in the future. Therefore, our results are comparable to the GST changes reported in previous studies, which, more importantly, have implications to the possible magnitude of ΔGST once pumping occurred in these areas. It should be noted that the aim of this study is neither to claim that GW pumping can inevitably increase GST nor to rigorously reproduce GST variations reported in previous observations or modeling. Our objective is to reveal the variations of GST under perturbations of pumping and the possible contribution of individual pumping to variations of GST by comparing the simulated ΔGST to those in previous studies.
3.2.3 Effects of Pumping Rate and Irrigation
In addition, by changing the pumping rates in Scenario 4, nonlinear variations of ΔGST were obtained. ΔGST in this paragraph is calculated from monthly average GST. Taking June in Beijing as an example, ΔGST (spatially averaged in Beijing) with double pumping rate scenario (0.61 °C) is about twice as large as ΔGST with normal pumping rate (0.33 °C). In comparison, when the pumping rate is reduced to half, the spatially average ΔGST becomes about one third (0.10 °C) of that under normal rate (0.33 °C). More importantly, 1 year of pumping with double rate generated higher ΔGST than 2 years of pumping with normal rate, which can be observed in Figure S4, such as the increase of GST in May and the decrease of GST in October. Therefore, moderate pumping rate under proper water management is important for the sustainability of water resources and sustainable development of ecological environment. With irrigation, the increase and decrease of GST are obviously alleviated (Figures 13 and S5) but cannot be completely eliminated. This can be clearly observed in Figures 10g–10i and 11g–11i, in which the peaks of frequency shift to ΔGST of 0 relative to those after 1- or 2-year pumping but are still larger than 0 in summer while smaller than 0 in winter. Applying P&I to crop areas in the Little Washita watershed, Ferguson and Maxwell (2011) found that pumping led to an increase of WTD over 15% of the watershed area while irrigation led to a decrease of WTD in only 1.6% of the watershed, which also indicates the limited compensation of irrigation.
3.3 Mechanisms and Implications
As the top boundary condition in modeling heat transport in the subsurface (equations 1 and 2], G might be a factor inducing variations in GST. The monthly average changes of LE, H, and G after 1 year of pumping are shown in Figures S6, S7, and 14, respectively. With increasing WTD by pumping, LE decreases and H increases, which is consistent with the general subsurface-land surface feedback mechanism as shown by the increase of Bowen ratio (Keune et al., 2016; Pal & Eltahir, 2001). The increase of G in summer from April to July may contribute to the increase of GST at the same time. However, increased G is also simulated while GST decreases in autumn and winter from September to November. Considering that soil moisture decreases with pumping, the volumetric heat capacity decreases (Abu-Hamdeh, 2003), so the heat storage/release capacity of the subsurface decreases. Therefore, in summer, heat flux into the subsurface combined with reduced heat storage capacity increase GST. In winter, reduced heat flux released from the subsurface due to the reduced heat storage in summer leads to the decrease in GST even though G increases. In addition, decreased thermal conductivity due to decreasing soil moisture (Rerak, 2017) can prevent the quick storage of incoming heat in subsurface in summer and the timely release of outcoming heat from subsurface in winter, which further aggravate the hotter summer and colder winter. However, for more complex variations such as that in southern Hebei, further studies with models of high spatial and temporal resolutions are necessary.
Previous studies suggested the critical depth range that has been mentioned in section 3.1.3 (e.g., Kollet & Maxwell, 2008). Often, GW supplies moisture for surface heat fluxes when WTD is shallow. In contrary, when the water table is too deep, GW has little influence on soil moisture that contributes to surface heat fluxes. The thresholds of such a range was well stated by numerous previous case studies (Ferguson & Maxwell, 2010; Kollet & Maxwell, 2008; Maxwell et al., 2007; Szilagyi et al., 2013) and summarized as 1–10 m (Ferguson & Maxwell, 2011). In Condon et al. (2013), possible influencing factors on the critical depth range were systemically studied. Results revealed that the critical range does not vary significantly with subsurface parameters. Though land cover and soil types affect the shape of the relationship between WTD and land surface heat fluxes, the general range of the critical depth stays the same.
In this study, not only the variations of GST with WTD were discussed in section 4.1.3, ΔGST in summer (e.g., May) and winter (e.g., November) due to 1-year pumping and the corresponding WTD were also plotted in Figure 15. The WTDs before and after pumping are also distinguished, which are noted as WTDb and WTDa, respectively. In Figure 15, the most important finding is that the change of GST due to pumping mostly occurred with WTD around 1 m and less than 10 m. This sensitive range of WTD for ΔGST is more concentrated around WTD of 1 m than that for GST (Figure 7). In addition, obvious increase or decrease of GST to about 1 °C occurred in croplands which is the main land cover type in plains. In summer, obvious increase of GST with a maximum of 1 °C can also be found in forests, which are distributed along the transition zones between plains and mountain areas, whereas GST are less sensitive to perturbations of pumping in grasslands, shrublands and also winter forests. Even though, pumping indeed increases the variability of ΔGST in all land cover types with WTD of 0.01–10 m.
Based on previous studies and the analysis in the NCP, the 0.01- to 10-m critical depth range was adopted for the following discussion. Therefore, with the 0.5-m increase of WTD per year estimated due to pumping, WTD may continue to increase from the 1970, that is, prepumping condition, for about 20 years until the average WTD drops below 10 m. Hence, pumping could have influenced GST for about two decades. Considering the uneven WTD distribution of a depressioncone, the pumping influence can last longer time than 20 years regionally. Considering the nonlinear behaviors of GST with pumping discussed in section 4.2, the increase of GST should be faster in the beginning and gradually slow down. This inference is also supported by Figure 15, in which the sensitivity of ΔGST decreases from WTD of 1 to 10 m. Generally, the subsurface acts as a buffer to heat fluxes at the land surface, but long-term pumping can gradually weaken and finally eliminate this buffer when the water table becomes too deep. The weakened buffer can result in higher temporal variability of GST, that is, colder winter and hotter summer. Although irrigation alleviates the impacts of pumping by increasing soil moisture at the land surface, it cannot completely eliminate the pumping effect (Figures 10g–10i and 11g–11i).
3.4 Limitations and Future Work
Our modeling and analyses could be potentially improved in a few directions. More information on the temporal and spatial variations of P&I in the NCP during the past 60 years may be used to further constrain the model setup in future studies. The NCP is also influenced by urbanization, which deserves attention when attributing changes in GST in the region. For example, land use data of the NCP in 2003 (Cao et al., 2014) and 2009 (Pei et al., 2015) illustrate that urban regions scattered among croplands have expanded gradually. Urban expansion can alter the spatial distribution of irrigation regions by reducing their sizes and fragmenting their coverage. Meanwhile, a depression cone can continue to expand from its center located mainly in the irrigation regions to a much broader area. Near the edges of the GW cones of depression, the shallow WTD may stay within the critical zone (WTD of 1–10 m), so the effect of pumping can be particularly significant when irrigation is not regulated. In this study, sprinkler irrigation is applied by adding the irrigation water to precipitation. However, flood irrigation is the most popular way of irrigation in the NCP (Cao et al., 2013). Flood irrigation has lower water utilization efficiency (Kendy et al., 2007; Scanlon et al., 2012) than sprinkler irrigation, which can lead to larger increase in GST than what we obtained in this study. Flood irrigation has not been implemented in ParFlow.CLM so that future work should consider modeling different irrigation methods to allow investigation into their specific impacts (Delos Reyes & Schultz, 2019; Leng et al., 2017).
Future work could also include improved representation of land surface processes. For example, heat transport in CLM is vertically described using a one-dimensional model, while GW flow is modeled using the three-dimensional Richards's equation in ParFlow. This approach is reasonable for modeling energy and water fluxes in regions with limited energy and less human activities. However, with human activities such as GW pumping, horizontal temperature gradient may be induced, so heat transport in horizontal direction should not be neglected. Other limitations of the modeling approach include the insufficient soil depth and the simple lower boundary condition implemented in CLM for heat transport. Davison et al. (2015) investigated the sensitivity of GST simulations to soil depth using aquifers of 2 m and 8 m depth in 100 days of simulation under prolonged drought condition. Although GST shows no difference between two settings, deep ST (>1 m) is higher for the simulation with aquifer depth of 2 m. With long-term simulations, the difference of deep ST becomes greater and propagates to the land surface so GST may be eventually different between two settings after 100 days of simulation. Since GW pumping has been practiced in the NCP for more than 60 years, future studies should also consider increasing the soil depth for heat transport in the model.
4 Conclusions
In this study, integrated land surface-subsurface modeling of the NCP was performed using ParFlow.CLM. Simulated results of water and energy dynamics are highly consistent with those in previous studies and from the JRA-55 data, respectively. Both the spatial and temporal variations of water and energy processes were well captured by the integrated model. Based on a suite of numerical experiments, the effects of GW pumping and combined P&I on water and energy were explored, with a focus on the GST, which was rarely discussed in previous studies. Results show significant effects of GW pumping on GST in the NCP. Generally, the subsurface acts as a buffer to heat fluxes at the land surface, but long-term pumping can gradually weaken and finally eliminate this buffer. This results in higher temporal variability of GST, featuring hotter summer and colder winter. Increased spatial variabilities of GST were also found in the area.
Considering that changes of WTD can significantly affect surface heat fluxes with WTD roughly in the range of 0.01–10 m in the NCP, the 0.5 m/yr increase of WTD can continuously increase/decrease GST for at least 20 years based on the average WTD in 1970. If the uneven WTD distribution of depression cones is also considered, this influence could last until WTD in the whole NCP increases by over 10 m, which may take longer than 20 years. In addition, GST is expected to increase/decrease faster in the beginning and gradually slow down due to the nonlinear variations of GST with WTD. Irrigation alleviates this situation by increasing soil moisture at the land surface but cannot completely eliminate the effect of pumping. Considering the spatial and temporal variations of P&I, urbanization, and the way of irrigation modeled in this study, the effect of irrigation might be overestimated.
This study aims to build a realistic modeling platform to understand the water and energy cycles and their interactions in a subsurface-land surface system. Hence, no efforts were devoted to establishing a calibrated model to fit the historical data or to predict future changes. Unlike previous studies that modeled the surface and subsurface as separated systems in the NCP, the coupled ParFlow.CLM model provides an important tool for more investigations of P&I in the context of climate change in the future. GW pumping has already been a global problem and occurs not only in the NCP but also in northwestern India, Middle East, the U.S. High Plains (Famiglietti, 2014), and other regions. Hence, the results of this study may have implications for other regions with GW depletion and motivate the need to investigate the role of GW pumping in regional climate change.
Acknowledgments
This study was supported by the Strategic Priority Research Program of Chinese Academy of Sciences (Grant XDA20100104), the National Natural Science Foundation of China (Grant 41807198), and the Center for Computational Science and Engineering of Southern University of Science and Technology. H. L., Y. F., and L. R. L. were supported by the U.S. Department of Energy Office of Science Biological and Environmental Research as part of the Earth and Environmental System Modeling program. PNNL is operated for the Department of Energy under Contract DE-AC05-76RL01830. All data used for modeling are publicly accessible, and the sources are cited in the text. Data for all figures using the model generated data and the model can be accessed through this site (https://doi.org/10.6084/m9.figshare.12003315.v1).
