The global consensus on the gradual withdrawal of all fossil fuels has made the issue of coal phase-out a focal point. However, effectively carrying out this phase-out and addressing its feasibility present considerable challenges, particularly in major coal-consuming countries1. Despite the global focus on coal phase-out, coal consumption has actually increased in 2022, not only in developing nations but also in European countries2. Prior research has underscored the significant social justice issues that arise from coal phase-out3,4,5. These issues encompass complex challenges, including the livelihoods of workers in the industry6,7, the management of stranded assets8,9, and the transformation of economic structures10. These challenges are particularly pronounced in coal-dependent developing countries such as China, India, and Indonesia. To navigate these challenges effectively, it is essential to gain a comprehensive understanding of the potential benefits that can arise from coal phase-out, including those derived from coal mine closures. Examining these benefits can contribute to advancing the coal transformation process and establishing effective strategies for a successful transition.

Previous studies have predominantly focused on the air pollution benefits of coal phase-out, particularly from the perspective of coal use11,12,13,14,15. However, there exists a significant research gap in understanding the benefits of coal phase-out in mining areas. In China, the central and western regions are key coal production areas and face water scarcity challenges due to arid or semi-arid climates. This presents notable challenges to the extraction and utilization of coal in these regions. Throughout the coal development process, water-related issues, such as depletion and pollution, are prevalent and have detrimental impacts on water resources16,17,18. Similar constraints and adverse impacts on water resources are observed in other coal-producing countries, including the United States, where states like Wyoming face water scarcity issues. Although the closure of coal mines has been acknowledged for its impacts on various aspects, including the reduction of air pollution19,20, biodiversity21, local employment6, and livelihoods7, the consideration of water resources has often been overlooked.

This paper examines the effect of coal mine closure in China on water availability, which is a crucial aspect of the environmental and ecological systems on Earth. The closure of coal mines can impact terrestrial water storage through various mechanisms. During coal mining operations, significant water usage is required for activities like coal washing and dust suppression22,23. Moreover, the dewatering process employed to keep the mines dry involves pumping out groundwater, leading to a lowered water table and reduced water storage in aquifers17,24. As a result, the mining of coal per ton generally leads to water depletion by more than one cubic meter25. With the closure of the mine, the demand for water decreases and the dewatering process ceases. The abandoned mining pits may even act as reservoirs, allowing for the accumulation of groundwater.

In addition to these direct effects, mine closure can have economic implications at the local level, indirectly influencing water usage in industrial and agricultural activities26,27. The reduction in mining-related activities may precipitate a decrease in water demand from these sectors, thereby influencing local water storage patterns. Furthermore, the closure of coal mines can contribute to the restoration of ecological systems, resulting in alterations in terrestrial water storage28,29. Specifically, the increased vegetation cover resulting from mine closure can influence water storage dynamics. A greater presence of vegetation can enhance water absorption and retention in soils, thus potentially increasing terrestrial water storage. However, increased vegetation can also lead to higher water consumption due to the greater water needs of additional plant growth30.

In this study, we analyze the co-benefits of coal mine closure on water resources by combining GRACE data on terrestrial water storage (TWS) with a comprehensive dataset on coal mining operations in China. Using a staggered difference-in-differences approach, we quantify the impact of coal mine closure on TWS. We also examine potential mechanisms for the change in TWS by analyzing data on land use classification, vegetation index, cropland and irrigation areas, business registration of industrial enterprises, and groundwater storage. Additionally, we compare the short-term effects of mine closure on TWS to the predicted long-run changes due to climate change mitigation in the future, using hydrological model simulation results.

Results

Coal mining activities of China and their TWS trends

Since 2010, the Chinese government has implemented coal phase-out policies, initially targeting low-efficient coal production capacity and subsequently focusing on high-emission and excessive capacity in the coal industry. As a result, approximately 12,000 mines were closed between 2010 and 2019. This closure process is expected to accelerate further as part of China’s commitment to achieving its net-zero emissions target in the future.

Figure 1 displays the geological distribution of the 6347 major coal mines of China between 2014 and 2019 in our sample. These mines are concentrated in North China (mainly in Inner Mongolian and Shanxi), Northeast China (Heilongjiang, Jilin, and Liaoning), and a few southern provinces (particularly Chongqing, Guizhou, Yunnan, and Hunan). The time trends of the average TWS anomaly for the non-closed and closed mines are shown in Fig. 1b. For non-closed mines, their average TWS exhibits a steadily decreasing trend over time, from −22 cm in 2010 to −112 cm in 2020. This substantial reduction aligns with the substantial water usage and depletion effects commonly associated with coal mining. Conversely, in regions devoid of mining activities, the average TWS also demonstrates a declining trend during this period, but to a much lesser extent than the TWS observed around non-closed mines.

Fig. 1: Coal mine distributions of China and their TWS trends.
figure 1

Geographic distribution of coal mines of China between 2014 and 2019 (a). Terrestrial water storage of closed mines, non-closed mines, and GRACE grids without coal mining activities within a 25 km radius (b).

Full size image

Conversely, there is a reversed trend in the TWS of the closed mines. Between 2014 and 2016, when the majority of mine closures in our sample took place, their TWS experiences a notable uptick. This reversal indicates that the closure of coal mines not only stops the negative impact of mining operations on water storage but also has the potential to aid in water table restoration. However, there is a decline in TWS in 2017, potentially due to drought conditions in Northeast China and North China during that year. Additionally, it is important to note that some of the closed mines are in close proximity to the non-closed mines, which may have a negative effect on the TWS of the closed mines. Therefore, the actual reverse in the TWS trend could be even more pronounced than what is depicted in Fig. 1.

Effects of coal mine closure on TWS

While Fig. 1 provides descriptive evidence for the effects of coal mine closure on TWS, TWS is influenced by multiple factors such as meteorological, hydrological, and geological conditions, which may correlate with mining activities. To identify the causal effect of coal mine closure, we employ a staggered difference-in-differences approach, leveraging the timing of mine closures. We leverage mine-level information concerning the years of production cessation within our dataset, which encompasses both operational and closed coal mines spanning the period from 2014 to 2019. Using this information, we designate coal mines that ceased operations by 2019 as the treatment group, while those that remained operational are assigned to the control group. We intentionally choose not to compare the treatment group to locations with no mining activities for two main reasons. Firstly, mine locations could be endogenously correlated with TWS, which may introduce omitted-variable biases. Secondly, as depicted in Fig. 1b, TWS around non-closed mines exhibits a more pronounced declining trend compared to locations without mining activities. This disparity underscores the negative impacts of mining activities on water storage. Therefore, utilizing locations without mining activities as the control group would lead to an underestimation of the effect of mine closure, as it disregards the detrimental influence of production cessation.

To better assess the parallel trend assumption, which serves as a critical identification assumption for the difference-in-differences design, we extend our sample period from 2004 (the earliest year of mining activities for the 6347 coal mines in our sample) to 2021 for estimation. This extended timeframe allows for a more comprehensive examination of temporal trends and facilitates a rigorous evaluation of treatment effects over time. Consistency is maintained throughout all empirical analyses by employing this unified sample period. Figure 2 presents the point estimates and 95% confidence intervals for Eq. (2) using Ordinary Least Squares (OLS) and two robust estimation methods31,32. For all methods, the pre-treatment coefficients preceding mine closure are all centered around zero and are not statistically significant. This suggests that TWS around the closed mines in the years leading up to closure shares a similar trend with the non-closed mines. Meanwhile, the post-treatment coefficients indicate a slight increase in TWS for the closed mines in the first two years after closure, followed by a substantial surge in subsequent years. According to the OLS estimator, TWS of the closed mines has experienced a rapid rise of 12.1 cm by six years post-closure. The two robust estimators provide even larger estimated increases (16.4 cm and 23.0 cm). Although the 12.1 cm increase may seem modest, it represents the average effect of a single mine closure. When considering the number of mass concentration blocks for these closed mines (321) and the approximately 625 km2 grid size, the cumulative effect becomes substantial. The total increase in water availability resulting from these mine closures after 6 years amounts to 24.3 billion m3. This number is considerably higher than the predicted water consumption reduction from China’s coal transition in 2050 for two reasons6. Firstly, the impact of coal mine closure on terrestrial water storage extends beyond reducing water consumption, encompassing multiple channels. Secondly, the effects represent TWS increases resulting from the closure of 4355 coal mines, while the number of remaining mines is only half of that.

Fig. 2: Dynamic effects of coal mine closure on terrestrial water storage.
figure 2

Coefficients on the lags and leads in Eq. (2) are estimated by Ordinary Least Squares (blue circles), Sun and Abraham approach (orange triangles), Borusyak et al. approach (red squares). Their 95% confidence intervals are constructed with mine-level clustered standard errors.

Full size image

Supplementary Table 1 presents the average treatment effects for the closed coal mines as estimated by OLS. After controlling for meteorological conditions, column 4 shows that coal mine closure increases TWS by 18.8 mm each year (P-value < 0.001). To evaluate the extent of omitted variable bias, we utilize the Oster test to examine the relationship between the selection of observed control variables and the selection on unobservable33. Supplementary Table 1 reports the Oster bounds for the estimated effects of mine closure on TWS. Across different sets of control variables, the estimated Oster bounds consistently exclude zero, indicating the robustness of our empirical specification to the potential omission of unobservable factors influencing local terrestrial water storage. Supplementary Fig. 1 also displays the estimated coefficients on temperature bins and precipitation bins, illustrating a negative impact of temperature and a positive impact of precipitation on TWS.

In the Supplementary Information, we discuss the impacts of high spatial correlation in TWS on the empirical estimation and statistical inference. Additionally, we conduct robustness analyses to address and rule out concerns that the observed TWS increase is driven by gravity changes induced by mineral extraction or potential policies implemented concurrently with coal mine closure. We also demonstrate the robustness of our findings to different binning choices for temperature and precipitation.

Underlying mechanisms for the rapid TWS increase

We analyze possible channels for the significant effects of coal mine closure on terrestrial water storage. First, we examine whether the increase in TWS is induced by ecological restoration using data on land cover changes. Figures 3a and b show the effects of coal mine closure on land areas covered by forests and grass. The post-treatment coefficients for both types of areas are quite small and not statistically significant, suggesting that on average, mine closure has negligible effects on expanding forests and grasslands during the sample period. Therefore, the observed drastic TWS increase is unlikely to be driven by the restoration of forests and grasslands. Moreover, we consider the effects of mine closure on the Normalized Difference Vegetation Index as a robustness check. Supplementary Table 2 shows that the estimated coefficients are minimal (0.0003), reaffirming that on average there are no significant changes in vegetation cover around the closed coal mines.

Fig. 3: Potential channels for the observed TWS increase.
figure 3

Dynamic effects of mine closure on forest area (a), grassland area (b), waterbody area (c), farmland area (d), number of industrial enterprises (e), and groundwater storage (f). Figure a–e are estimated with Poisson regression since the dependent variables are count variables. Figure f is estimated with OLS. Circles denote estimated coefficients on leads and lags. Orange lines represent the average effects on the closed mines. Shaded areas denote the 95% confidence intervals constructed from mine-level clustered standard errors.

Full size image

It is important to note that these null effects observed within the six-year period following mine closure should not be interpreted as a complete absence of impact on vegetation restoration. While there is limited evidence of significant changes in forests and grasslands during this timeframe, it does not discount the potential for restoration to occur over a longer period. Ecological restoration is a gradual and long-term process that may take more time to manifest observable changes in vegetation cover.

In addition, in Fig. 3c, we use land cover data to examine whether coal mine closure affects surface water bodies (wetlands and reservoirs) and find negligible impact. This limited impact is likely because the majority of closed mines are underground mines, and their closures do not significantly affect surface water bodies.

Another possible channel is that, in conjunction with coal mine closures, local governments might be implementing other policies that may affect TWS. Given that agriculture accounts for over 60% of water usage in China, it is likely to have a substantial impact on TWS. However, Fig. 3d shows that the effects of coal mine closure on farmland areas are generally negative but small, except for the sixth year after closure. Supplementary Table 3 further shows that the effects of mine closure on wheat and rice cultivation areas, as well as irrigation areas, are all not statistically significant. Therefore, there is no evidence to suggest that changes in agricultural activities around the closed mines contribute to the increase in TWS.

The closure of large coal mines can have a substantial impact on the industrial sector, resulting in the exit of coal-fired power plants and high-energy-consuming manufacturing firms. Furthermore, areas with coal mine closures may also experience a decrease in the entry of new industrial firms compared to non-closed mines. While data on firm exits is not available for our analysis, we examine this potential channel by exploring the effects on the entry of new industrial firms around the coal mines in our sample. Figure 3e demonstrates a negative effect of mine closure on firm entry, albeit somewhat small. Supplementary Fig. 2 further shows the average treatment effect is roughly −0.8%. Additionally, we consider the effects on static land thermal anomalies, which are used as a proxy for industrial production activities34. Supplementary Table 4 shows that mine closure reduces thermal anomalies by about 1.7%. These reductions imply a decrease in industrial activities following mine closure, which further suggests a reduction in industrial water usage. As a result, these changes in industrial activities could contribute to the increase in TWS.

Lastly, the dewatering process and substantial water usage of mining operations lead to groundwater depletion. However, the closure of coal mines not only halts this depletion but also has the potential to increase groundwater levels. This is particularly relevant for underground mines, as their mining pits can be repurposed for water storage. To examine the effects of coal mine closure on groundwater, we control for additional factors such as surface runoff, soil moisture, plant canopy surface water, and snow depth in Eq. (2). Figure 3f demonstrates that coal mine closure substantially increases groundwater storage by more than 10 cm by the sixth year after closure. Additionally, Supplementary Fig. 2 shows that the average increase in groundwater storage amounts to 3.5% per year (P-value < 0.001). Considering the average TWS, the 3.5% increase implies a 17.5-mm increase in groundwater storage. This significant effect suggests that most of the TWS increase is driven by an increase in groundwater storage.

The observed impact on groundwater storage in our study corresponds to the characteristics of predominantly unconfined aquifers in mining regions. This renders it conducive to detection through GRACE satellite observations, owing to their direct surface connectivity and capacity to readily detect water level fluctuations. Specifically, many aquifer systems associated with coal mines, especially in the arid northwest and northeast regions, remain unconfined35. Notably, the northeast region, with its higher incidence of mine closures, offers a robust context for evaluating the impact of mining activities on groundwater dynamics. Furthermore, in order to mitigate the risk of mine flooding, the primary cause of coal mine accidents in China, coal mining operations typically avoid areas characterized by extensive confined aquifer systems36. Meanwhile, coal mining operations inherently exert a more significant influence on water storage within unconfined aquifer systems, primarily due to their proximity to the surface and heightened vulnerability to surface disturbance and subsidence induced by mining activities. Consequently, mining operations can substantially alter groundwater storage dynamics.

Additionally, in the Supplementary Information, we assess the contribution of production cessation to the observed increase in TWS by leveraging the timing of coal mine operations. The analysis reveals that the cessation of mining operations contributes to over 60% of the observed TWS increase. These findings suggest that coal mine closure leads to an increase in groundwater storage, primarily due to the cessation of the detrimental impacts of coal mining operations on groundwater.

Discussion

Figure 2 demonstrates the average impact of one coal mine closure on terrestrial water storage. To gain an overall implication of the potential impact of coal phase-out in China, we consider an extreme scenario where all non-closed coal mines by 2019 were closed. Figure 4a displays the predicted TWS changes after six years following their closures, based on the OLS estimation results from Fig. 2. There are significant TWS increases in North and Central China, regions that generally have lower water availability. Furthermore, certain southern regions with extensive coal mining activities also exhibit substantial TWS increases after mine closure.

Fig. 4: TWS changes due to coal mine closure and climate change mitigation.
figure 4

TWS change due to coal mine closure after six years (a). TWS differences between the low-emission scenario (SSP1-2.6) and high-emission scenario (SSP3-7.0) in the late century (2070-2099), averaged based on simulation results of two hydrological models and six climate drivers (b).

Full size image

Extensive research has demonstrated the influence of climate change on the terrestrial water cycle37,38,39. These studies indicate that future changes in climate are expected to profoundly alter the availability and abundance of global water resources40,41,42,43. The coal phase-out can reduce greenhouse gas (GHG) emissions and counteract the impact of climate change on water availability.

To evaluate the indirect impact of climate change mitigation, we incorporate simulation results from two hydrological models, CWatM and H08, utilizing five different climate drivers. By comparing the simulation results under the low GHG emission scenario (SSP1-2.6) and the high GHG emission scenario (SSP3-7.0), we can assess the differences in TWS as an indicator of the effects of climate change mitigation through GHG emissions reduction on water storage. Our analysis focuses on TWS changes in the late 21st century, specifically from 2070 to 2099. This extended timeframe allows us to capture the potentially larger impacts of climate change and mitigation.

In the late 21st century, Fig. 4b reveals that most southern regions of China are projected to experience increases in TWS due to GHG emission reduction efforts. In contrast, most northern regions, excluding the northeast, are expected to undergo substantial TWS losses compared to the high-emission scenario. These losses are particularly pronounced in North and Central China, which are regions with a substantial number of coal mines. However, when considering the increases in TWS resulting from the closure of coal mines, the relative losses of TWS in these regions could be substantially mitigated. Therefore, while climate change and its mitigation efforts have heterogeneous impacts on TWS, the closure of coal mines in China appears to complement the effects of GHG reduction by increasing water storage in regions with relatively lower water availability.

In addition to the impact on TWS, it is important to acknowledge that coal mining activities have a large effect on water pollution. Acid mine drainage, soil erosion, sedimentation, and contamination from heavy metals and mine discharges are among the contributors to water pollution associated with coal mining. The closure of coal mines plays a crucial role in halting water pollution originating from mining activities and preventing further deterioration of water quality. This aspect of water benefit resulting from coal mine closure deserves further attention and research to fully understand its extent and implications.

In the pursuit of the net-zero emission target, there will be a heightened tradeoff between carbon and water as new technologies, such as dry cooling and carbon capture and storage, are utilized in thermal power generation44. Our analysis underscores the importance of considering the synergy with water resources in the assessment of the feasibility and potential impacts of energy transition.

More broadly, the transition away from coal will have wide-ranging implications, raising concerns about social justice, particularly in regions heavily reliant on coal-related industries. However, our study highlights the substantial benefits that arise from the closure of coal mines in terms of water resources. These benefits, although substantial, often go unnoticed by the general public. Recognizing and understanding the positive impact on water resources can further bolster and encourage the process of phasing out coal, particularly in arid regions where water scarcity is a pressing issue. Additionally, although the immediate effects of coal mine closure on vegetation restoration may be limited, the long-term implications could be significant. The restoration of water resources will provide a foundation for future ecological restoration, contributing to ecosystem management and sustainable development in mining regions.

Methods

Data

To analyze the impact of coal mine closures on water resources, we primarily rely on a dataset from the National Development and Reform Commission pertaining to above-scale coal mines in China, in conjunction with a satellite-based dataset on terrestrial water storage.

The dataset from the National Development and Reform Commission includes all above-scale coal mines in China for the period between 2014 and 2019. This dataset contains mine-specific information, including the name, start year, the year of closure (if applicable), and the geographical location of each mine. After excluding mines with missing key variables, except for the year of closure, we are left with a total of 6347 mines. The earliest of these mines began operations in 2004. By the end of 2019, 4355 of these mines had ceased operations. Approximately 8%, 19%, and 35% of these closures occurred in 2014, 2015, and 2016 respectively. Furthermore, around 10% closed in each subsequent year from 2017 through 2019.

We utilize the satellite-based product from the Gravity Recovery and Climate Experiment (GRACE) to measure terrestrial water storage. Launched in March 2002, the GRACE mission consists of two satellites, GRACE-1 and GRACE-2, that orbit the Earth in tandem, maintaining an approximate distance of 220 km from each other. Changes in Earth’s gravity, caused by shifts in mass such as water movement, affect the relative velocities of the two satellites and the distance between them. Water distribution across the planet can be inferred by tracking these variations over time. We use the CSR GRACE/GRACE-FO product with corrections applied45. This product offers a resolution of 0.25 degrees, derived from the original GRACE resolution of approximately 300 km, for global monthly terrestrial water storage since April 2002. These corrections account for measurement errors resulting from instrumental noise, atmospheric and oceanic effects, and spatial leakages.

For each mine in our dataset, we match it to the closest grid in the GRACE product and utilize the Terrestrial Water Storage of that grid as the TWS for that particular mine. Given that we lack information regarding the exact month of mine entry and closure, we aggregate the monthly TWS data to an annual level.

In order to account for the effects of weather on TWS, we consider a comprehensive set of meteorological variables, including temperature, precipitation, humidity, air pressure, wind speed, and evaporation. These variables are obtained from the Global Land Data Assimilation System (GLDAS), which provides global meteorological variables at a 0.25-degree resolution. In addition, to isolate changes in groundwater, we further consider other factors that directly contribute to water balance, including surface runoff, soil moisture, plant canopy surface water, and snow depth (in terms of water equivalent). These variables are also sourced from GLDAS.

To understand underlying mechanisms of the effects of coal mine closure on TWS, we consider changes in vegetation cover. We utilize land type classification data and the Normalized Difference Vegetation Index from the Moderate Resolution Imaging Spectroradiometer (MODIS). We also consider changes in wheat and rice cultivation areas46 and irrigation areas47, to understand the contribution of agricultural water usage. To explore the role of industrial firms in the increase in TWS, we analyze the effects of coal mine closure on thermal anomaly at non-vegetation static land sources, which are considered a proxy for industrial activities, by using the high-resolution (375 m) thermal anomaly data from VIIRS S-NPP for this purpose. Additionally, we consider changes in the entry of new manufacturing firms by using business registration data from the State Administration for Industry and Commerce. To ensure consistency with the spatial resolution of TWS, we calculate mine-level outcomes for other variables using a 25-km radius around each mine.

We utilize simulation results from two global hydrological models (CWatM and H08) and five different climate drivers (MRI-ESM2-0, IPSL-CM6A-LR, MPI-ESM1-2-HR, UKESM1-0-LL, and GFDL-ESM4) under two emission pathways (SSP1-2.6 and SSP3-7.0) to measure the impacts of reducing greenhouse gas emissions on terrestrial water storage. These model simulation results are obtained from the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP).

Models

We estimate the impacts of coal mine closure on TWS via a difference-in-differences design, leveraging the staggered closures of the 4355 mines in our dataset.

To capture the total effects of mine closure, it is essential to consider both the effects from production cessation and water restoration. For this purpose, we designate the coal mines that had closed by 2019 as the treatment group, while those that had not closed serve as the control group. We then estimate the effects of mine closure on TWS using data from the commencement of each mine’s operation. This choice of control group not only helps mitigate bias due to endogenous site selection of the coal mines, but also helps establish a parallel pre-treatment trend between the treatment and control groups, enhancing the validity of the empirical design.

We consider the following equation as our main empirical specification:

$${TW}{S}_{{it}}=\alpha +\beta {Exi}{t}_{{it}}+\gamma {X}_{{it}}+{\eta }_{i}+{\tau }_{{lat},t}+{\delta }_{{lng},t}+{\epsilon }_{{it}}$$
(1)

where \({TW}{S}_{{it}}\) represents the terrestrial water storage of coal mine \({{{\rm{i}}}}\) in year \({{{\rm{t}}}}\), \({Exi}{t}_{{it}}\) is dummy variable that equals one if mine \(i\) has closed by year \(t\), \({X}_{{it}}\) is a set of covariates to control for meteorological conditions, including temperature, precipitation, humidity, air pressure, wind speed, and evaporation. To allow for non-linear effects of temperature and precipitation on TWS, we consider 8 temperature bins (binned by 0, 5, 10, 15, 17, 19, and 21 Celsius degrees) and 11 precipitation bins (binned by 100, 200, 300, 500, 800, 1000, 1200, 1400, 1600, and 2000 mm). \({\eta }_{i}\) represents mine-level fixed effects, capturing time-invariant factors that affect TWS, such as topographical and geological conditions. We include latitude- and longitude-specific year fixed effects by leveraging the latitudes and longitudes of matched gridded cells in CSR GRACE/GRACE-FO data (\({\gamma }_{{lat},t}\) and \({\delta }_{{lng},t}\)). This set of fixed effects serves to capture time-varying characteristics specific to geographic bands or at a large scale. In particular, it can capture environmental or climatic changes that affect specific latitudinal belts differently over time. For instance, increasing temperatures over time might have disparate impacts on low-latitude and high-latitude regions48. Additionally, it accommodates variations in economic and environmental policies that may impact eastern and western parts of the country differently. Furthermore, as illustrated in the Supplementary Information, this set of fixed effects aligns with the original resolution of the outcome variable and accounts for its inherent spatial correlation. We use Ordinary Least Squares (OLS) to estimate Eq. (1), with standard errors clustered at the mine level.

One specific concern about the empirical design is that the staggered phase-out of coal mining may suggest a decreasing production trend for the closed mines before their closure. If such anticipation effects exist, the estimated coefficients from our event-study specification will provide a lower bound, and the true effect could be larger. Furthermore, we can explicitly investigate the existence of anticipation effects by estimating the dynamic impacts of mine closure with lags and leads. More specifically, we consider the following specification to examine the parallel trend assumption.

$${TW}{S}_{{it}}= \, \alpha +{\beta }_{\le -10}{Exi}{t}_{{it},\phi \le -10}+{\sum}_{\phi =-9,\phi \ne -1}^{6}{\beta }_{\phi }{Exi}{t}_{{it},\phi }\\ +\gamma {X}_{{it}}+{\eta }_{i}+{\tau }_{{lat},t}+{\delta }_{{lng},t}+{\epsilon }_{{it}}$$
(2)

where \({Exi}{t}_{{it},\phi }\) is a dummy variable that equals one if mine \(i\) in year \(t\) is \(\phi\) years away from closure.

Given that the empirical specification follows a staggered event-study design, the examination of the parallel trend assumption with OLS could potentially encounter pitfalls such as negative weights and contamination from multiple treatment cohorts. Therefore, in addition to OLS, we examine the parallel trend assumption using two robust estimators31,32.

The absence of significant pre-trends could rule out anticipation effects, and also alleviate concerns about other confounding factors that affect terrestrial water storage. However, the timing of coal mine closures might coincide with other policies that also positively affect TWS. Specifically, the pressure for environmental and ecological protection in mining areas might compel local governments to both shut down high-damage mines and enhance regulations on existing mines. Therefore, the increase in TWS around closed mines may also be attributable to improvements in environmental protection practices at nearby operational mines. We investigate this potential concern in the Supplementary Information by utilizing a list of national green mines, which have superior performance in terms of environmental protection.

Additionally, we construct Oster bounds to evaluate the extent of potential omitted variable bias in Supplementary Table 1. The Oster bounds are constructed as \([\widetilde{{{{\rm{\beta }}}}},{{{{\rm{\beta }}}}}^{* }\left({{{{\rm{R}}}}}_{\max },1\right)]\), where \(\widetilde{\beta }\) represents the coefficient on \({Exi}{t}_{{it}}\) with control variables, and \({\beta }^{* }\left({R}_{\max },1\right)=\widetilde{\beta }+(\widetilde{\beta }-{\beta }^{o})\times \left({R}_{\max }-\widetilde{R}\right)/\left(\widetilde{R}-{R}^{o}\right)\). Here, \({\beta }^{o}\) denotes the coefficient without any control variables, \({R}^{o}\) denotes the R-squared without control variables, \(\widetilde{R}\) denotes the R-squared with control variables, and \({R}_{\max }\) denotes the maximum possible R-squared. Adhering to established guidelines, we set \({R}_{\max }\) as \(\min \{1.3\times \widetilde{R},1\}\)33.