Introduction

Cities worldwide are currently experiencing an unprecedented horizontal expansion, with a staggering 326% increase in built-up areas between 1972 and 2019 (Huang et al. 2022). This rapid urbanization has led to significant conflicts between human activities and the natural environment (Floerl et al. 2021; Ren et al. 2022; van Vliet, 2019). The question of whether such expansion is sustainable is of utmost importance, as it not only provides insights into the future evolution of global city landscapes but also serves as a crucial lever for humanity to actively shape the ongoing urban development process. While numerous studies have made meticulous predictions about city expansion (Seto et al. 2012), they have overlooked the substantial impact of electricity access on shaping urban form. Moreover, the growth rate of built-up areas globally has decreased from the 4% annual rate of the previous century to approximately 2% in recent years (Huang et al. 2022). In contrast, per capita electricity consumption has risen by about 1.6 times from 1971 to 2014. Is there a correlation between these two trends, and to what extent can the increasing power generation capacity and electricity demand explain the future patterns of urban development?

The invention and widespread adoption of electricity marked the onset of the Second Industrial Revolution, driving the development of humanity as a whole. Electricity has played a significant role in fueling economic growth (Allcott et al. 2016), enhancing total factor productivity (Fiszbein et al. 2020; Hulten et al. 2006), and generating employment opportunities (Fiszbein et al. 2020). It has become an indispensable input factor upon which modern societies heavily rely. Moreover, access to electricity has brought about unexpected improvements in welfare, including increased enrollment of children in schools (Squires, 2015), higher participation of women in the labor force (Vidart, 2023), and even a reduction in individual mortality rates in extreme environments (He & Tanaka, 2023). However, current research has neglected to explore the impact of electricity on shaping the physical form of modern cities, despite their heavy dependence on it. Existing studies indicate that, compared to the horizontal expansion of cities, the redevelopment of existing urban areas is a cost-effective strategy (Trubka et al. 2010). Horizontal expansion necessitates the construction of additional supporting infrastructure and political capital, whereas vertical expansion focuses on optimizing the operational capacity of existing infrastructure. Power plants serve as the primary source of electricity production and, as large-scale energy infrastructure, they have the potential to significantly enhance electricity access and capacity within a city, enabling vertical expansion. Furthermore, these large power plants have a notable impact on employment (Bank, 1994), further contributing to population concentration at the city level. Taking into account these two factors alone, it seems evident that the establishment of large-scale power plants can intensify urban agglomeration in response to growing population demands.

In the global context, thermal power plants dominate the electricity supply and are the primary focus of this paper Footnote 1. One distinguishing characteristic of thermal power plants, compared to other types, is their significant pollution emissions, making them one of the largest sources of pollution worldwide (Mei et al. 2021; Pan et al. 2024). Consequently, the establishment of a thermal power plant can impose substantial health costs on nearby residents (Casey et al. 2020). The high pollution levels associated with thermal power plants can directly influence the shape of cities. When faced with pollution, individuals often choose to relocate to areas with relatively cleaner environments(Zhang et al. 2023), exercising their preference through what is known as “voting with their feet” (Tiebout, 1956). This, in turn, diminishes the appeal of the city for its inhabitants and undermines local population agglomeration, creating an effect that runs counter to the previous paragraph.

To summarize, the impact of thermal power plants on the form of cities is complex and requires rigorous empirical analysis to understand its comprehensive effects. In our study, we treat power plant establishment as a natural experiment and utilize multiple sets of high-precision, long-span satellite inversion datasets. Our findings indicate that the operation of power plants has a significant influence on promoting agglomeration-type city development. This is manifested through the relative shrinkage of built-up areas, an increase in the average height of city buildings, and a substantial rise in both population and population density. To ensure the validity of our findings, we conducted various robustness tests, including the parallel trend test and the examination of confounding factors.

Further, we demonstrate a number of potential mechanisms: the operation of power plants significantly increases the level of electricity consumption, employment absorption capacity and pollution in the region. In order to demonstrate the mechanisms more clearly, we also examine the city morphology altering effects of gas-fired power plant operation. Constrained by their size, they are less capable of shaping the horizontal and vertical expansion of the city than thermal power plants; however, due to the loss of the pollution attribute, they are significantly stronger in enhancing the population agglomeration. We also tentatively explored the ability of thermal power plant operation to shape urban morphology within cities. When investigating the impact of thermal power plant operation on population distribution within the city, We observed a clear rapid decline in population density as distance increased. After the power plant’s operation, the population within a two-kilometer radius increased by over 40%, primarily reflecting the employment stimulus effect within close proximity to the power plant. Beyond three kilometers, we observed a rapid decline in the influence of power plants on population density. Our analysis of electricity consumption and pollution levels around the power plants led us to conclude that this diminishing effect may be attributed to the rapid decrease in the supplied electricity level with increasing distance, while the associated air pollution does not dissipate as quickly. Finally, we also examine the moderating roles of transportation infrastructure, economic development level environmental regulation and natural factors, in the shaping effects of thermal power plants, which not only robustly proves the baseline conclusions and mechanisms of this paper, but also brings a new perspective on the subject matter.

The innovations of this paper are as follows: First, compared to other literature on urban expansion (Hu et al. 2024; Mahtta et al. 2022), this study explains the global slowdown in built-up area expansion and the increase in urban height from the perspective of electricity infrastructure, addressing the gap in existing literature regarding the role of electricity infrastructure in shaping urban vertical forms. Second, while existing studies focus on urban form changes at the national level (He & Zhou, 2024; Li et al. 2022; Uhl et al. 2021), there is a lack of a global perspective. This paper provides empirical evidence on the agglomeration effects induced by power plants at the global level, demonstrating that renewable energy power plants have a population agglomeration effect comparable to that of thermal power plants. Finally, this study offers complementary evidence regarding the interaction between geographic conditions, other infrastructure levels, and the capacity of electricity infrastructure to shape urban forms, emphasizing the need to explore the effects of electricity infrastructure based on local endowments.

Data resources

To measure city sprawl, this study utilizes a range of comprehensive datasets. The built-up area, total population, population density, and building height of cities are commonly employed indicators in measuring city sprawl, as demonstrated in various studies (Angel et al. 2011; Duranton & Puga, 2014; Guastella et al. 2019; Henderson et al. 2017; van Vliet, 2019; Wang et al. 2016; Zhong et al. 2023, 2022). These dimensions effectively capture the outward and upward expansion of cities, as well as changes in city density, providing a holistic understanding of city sprawl: Firstly, to measure the city’s built-up area, we adopt the impervious surface area (ISA) provided by Huang et al. (2022) (Huang et al. 2022). ISA, which refers to the surface area covered by human activities, is a crucial indicator of city sprawl and urbanization levels (Angel et al. 2011; Huang et al. 2022; Zhong et al. 2023). The dataset spans from 1972 to 2019, making it the longest dataset available to us, although it lacks continuity for years prior to 1985. This raster dataset has a precision of about 30m*30m, which is among the best in similar global databases. Undoubtedly, this long-span dataset will provide more convincing empirical evidence for our estimation. Secondly, the population data used in this study is a combination of two sources. The first source is the 1KM precision, 2000-2019 population raster data from Landscan, while the second source is the 1970, 1980, and 1990 population raster data from Socioeconomic Data and Applications Center(SEDAC), both of which are spliced together to obtain the longest possible set of population data. We use this population data in conjunction with the ISA dataset mentioned earlier to calculate the total population and population density of the city. Of course, due to the initial mismatch between our population dataset and the ISA dataset in the first few years, the population density indicator becomes available in the subsequent years. This also implies that in our baseline regression, we primarily investigate the impact of power plant operation on population density, focusing on samples from the later periods. However, this is not a significant issue, as the effects of power plants on ISA, total population, and population density can mutually support one another. Thirdly, to measure city vertical development, detailed data such as the housing plot ratio indicator used by Wang et al. (2016) (Wang et al. 2016) is not available at the global level. Therefore, we indirectly calculate the average height of city buildings whenever possible. Specifically, GHSL(Global Human Settlement Layer) provides volumetric raster data for housesFootnote 2, available over a five-year interval, which we combine with the impervious surface dataset mentioned earlier to obtain an indicator of the average height of houses in the city.

The power plant data used in this study is sourced from the Global Power Plant Database (GPPD), which provides information on more than 35,000 power plants worldwide. This database includes details such as the operation year of the plant(The first year of electricity generation), the primary energy source used, and the precise latitude and longitude coordinates of each power plant. In the baseline analysis, the focus is on thermal power plants that primarily utilize coal as their energy source. This choice is made because new energy power plants, such as renewable energy sources, account for only approximately 20% of the total power generation globally. By exploring the operation of thermal power plants, the study aims to provide a more representative analysis of power plant development trends Footnote 3. In addition, two control variables, average skin temperature and rainfall at the city level, are included in this paper to ensure the consistency of the estimation Footnote 4. As for the vector map of administrative districts, this paper uses the database provided by Global Administrative Areas(GADM), which is one of the most authoritative vector maps of administrative districts. Specifically, the third level of the GADM database is defined as “city”, providing a consistent and rigorous definition applicable to most countries or regions. To ensure the accuracy of causal estimates and avoid potential confounding factors, the study excludes samples with an average population of less than 2000 from 1970 to 2019, as determined by the population database.

Finally, we have created a panel dataset containing 38,000 second-level national units, with the longest time span dating back to 1970. It is important to note that the interval years and missing years vary across different indicators. We will later demonstrate that retaining the same sample for different indicators does not affect the conclusions of our baseline regression. At the end of the sample period, more than 1000 second-level national units had thermal power plants. At the end of the sample period, more than 1000 second-level national units had thermal power plants, which is approximately five times more than in 1970.

Figure 1 presents the global distribution of urban built-up area, total population, population density for the year 2019, and building height data provided by GHSL for the year 2018. Within cities, the built-up area, population density, and building height are averaged, while the total population is aggregated within each city. Additionally, we also display the distribution of thermal power plants of various sizes across the world in the figure1. It’s evident that our data coverage is sufficiently extensive. However, there is a significant disparity in the global distribution of power plants, with notable gaps in regions such as South America and Africa.

Fig. 1
figure 1

The distribution of urban morphology metrics and thermal power plant.

Full size image

Method, baseline results and robustness check

Method

The study employs a reduced-form econometric model for estimation, using the following variables. The dependent variable, Expansionit, measures the degree of city expansion for city i in year t and includes variables such as built-up area, total population, population density, and the average height of the city, all in logarithmic values. The variable Plantit is a dummy variable that takes the value of 1 if a power plant begins generating electricity in city i in year t, and 0 otherwise Footnote 5. Control variables, denoted as Zit, include factors like temperature and rainfall at the city level. To address potential omitted variable bias, the study incorporates strict fixed effect, including city fixed effects and country × year fixed effect. These fixed effects help control for time-varying characteristics at the country level and non-time-varying characteristics specific to each city.

$${Expansion}_{it}=\alpha +\,\beta {Plant}_{it}+\,\delta {Z}_{it}+\,{\epsilon }_{i}+\,{\delta }_{c* t}+\,{\varepsilon }_{it}.$$
(1)

Baseline results

Table 1 presents the results of the baseline analysis conducted in this paper. By controlling for strict fixed effects and time-varying control variables, the findings in Columns (1) and (4) reveal that the operation of a power plant is associated with a decrease of approximately 20% in the built-up area of the city, while concurrently, city heights increase significantly by 19%. These results suggest that the presence of a power plant is correlated with a restriction on the horizontal expansion of the city, while facilitating an increase in building height. As cities adapt to increased density, vertical expansion emerges as a more efficient use of land, particularly in urban areas where land is scarce and expensive. Additionally, Columns (3) and (4) show that the operation of the power plant coincides with an average increase of 1.9% in the city’s population and 7.6% in population density. This correlates with the establishment of power plants and the agglomeration of urban populations, fostering vertical rather than horizontal expansion, which in turn elevates population density. The availability of electricity supports the operation of taller buildings and other facilities, enhancing its appeal as a strategy for urban development. This reliance on electricity to facilitate vertical growth is a crucial factor linked to the observed increases in population density. In contrast, horizontal development often requires extensive political approvals or other scarce resources (Wang et al. 2016).

Table 1 Baseline results.
Full size table

Robustness check

Event study

As this paper investigates the effects of power plant operation on urban sprawl within a DID framework, the most crucial test is whether the design meets the parallel trend assumption. This means that if there were no shocks related to power plant operation, the trend of changes in the treatment and control groups should be identical. To estimate the dynamic impact of power plant operation, we use the event study method. The test model is as follows:

$${Expansion}_{it}=\alpha +\mathop{\sum }\limits_{j=\,-20/-10}^{50/25}{\beta }_{j}{{Plant}_{it}}^{j}\,+\,\delta {Z}_{it}+\,{\epsilon }_{i}+\,{\delta }_{c* t}+\,{\varepsilon }_{it}.$$
(2)

In Equation (2), the variable \({{Plant}_{it}}^{j}\) is a binary variable that takes a value of 1 when a city observation is j years away from the launch of the power plant. We use j = −1 as the base year. To ensure the balance of the estimated samples in the different regressions, as suggested by Miller & Douglas (2023) (Miller, 2023), we make the following adjustment: observations of samples with relative treatment time less than -20 or -10 are taken as 1 at the dummy variable characterizing -20 or -10, and observations of samples with relative treatment time greater than 50 or 25 are taken as 1 at the dummy variable characterizing 50 or 25. The results of the event study method are shown in Fig. 2. It can be observed that the treatment effect fluctuates around a value of 0 prior to the operation of the power plant, and there is no clear upward or downward trend in the pre-treatment point estimates. This suggests that the significance of the results of our baseline regressions is not driven by a pre-trend in the city form per se, but rather stems from the treatment effect of the power plant. This test provides evidence of a parallel trend. Another piece of evidence provided by the event study is that the effect of power plant operation on the city form has significantly increased over time.

Fig. 2
figure 2

Event study results.

Full size image

Stable unit treatment value assumption

The other test that this paper focuses on is the one related to spatial spillover effects. One of the key assumptions of the DID model we adopt is the SUTVA assumption, which assumes that the occurrence of a treatment does not affect the state of the control group. In the context of this paper, this means that the operation of a power plant in one city does not affect the morphology of the surrounding cities. However, this is a relatively strong assumption, and we do not have sufficient evidence to confirm that it is met. To address this issue, we use the spillover-robust DID model to test whether our core conclusions still hold when the SUTVA assumption is relaxed (Clarke, 2017). This model is chosen for robustness testing because it does not strictly require the form of existence of spatial spillovers. Instead, it assumes that spillovers can only exist in cities adjacent to the city where the power plant exists, and the area where the spillover effect exists is determined by the optimal bandwidth, avoiding the possibility of artificially setting the form and scope of the spillover effect and thus manipulating the results (Clarke, 2017). The selection process is shown in Fig. 3, where we select the bandwidth corresponding to the regression equation based on the principle of minimizing the root mean squared error (RMSE). The regression results for spillover-robust DID are presented in Table 2. Even after considering the case where the SUTVA assumption does not strictly hold, the core conclusions of this paper still hold under the spillover-robust DID model.

Fig. 3
figure 3

RMSE for each model.

Full size image
Table 2 Spillover-robust DID Model.
Full size table

Additionally, we conducted two further tests detailed in the appendix. Firstly, we excluded cities located in close proximity to power plants to account for any immediate geographical influences. Secondly, we removed the three most populous cities from each country from our dataset to mitigate the potential distorting effects of population agglomeration. These steps help ensure that our results are robust against such biases.

Policy confounding effects

A potential concern is that the designation of ‘power plant cities’ could be influenced by regional industrial revitalization policies, which might lead our estimates to reflect impacts not solely attributable to the power plants themselves, but to other confounding factors. However, our empirical estimation strategy substantially mitigates this concern. We define the onset of a ‘power plant city’ based on the earliest year a thermal power plant commenced operation, rather than the construction start year. Upon meticulous review, we observed that the discrepancy between the construction and operational commencement dates of power plants appears random and not systematically fixed. Moreover, our event study analysis supports this by showing no significant pre-trends prior to the plants becoming operational. This strengthens the credibility of attributing observed effects specifically to the operational impact of the power plants. For detailed information and supporting data, please refer to the appendix.

Leave one out estimate

To ensure that our findings are not biased by the influence of any specific country, we implemented a leave-one-out estimation strategy. This involves sequentially excluding each country from the analysis and recalculating the estimates using the remaining countries’ data. The results, displayed in Fig. 4a–d, show no evidence that any individual country disproportionately affects our overall findings.

Fig. 4: Leave-one-out Estimate.
figure 4

a Built-up Area. b Population. c Population Density. d Height.

Full size image

Estimate in same sample

To avoid discrepancies in our estimates due to variations across different samples or to ensure that we are measuring the effects over the same time period, we have filtered a consistent sample across all four indicators for further analysis. The estimation results are presented in Table 3.

Table 3 Estimations with same sample.
Full size table

Delightfully, despite a significant reduction in sample size to ensure a balanced representation across indicators, the estimation results remain highly consistent with the baseline regression outcomes.

Other robustness check

This paper also conducts a series of other robustness tests. Firstly, due to the existence of a small number of missing data situations in various types of raster datasets, observable samples may enter or exit non-randomly, which can significantly affect the causal estimation of this paper. To address this issue, we re-estimate using strongly balanced panels. Secondly, we address the issue of clustering standard errors by clustering one more dimension and adjusting it to the provincial level. Additionally, although we have put the dependent variable in the natural logarithmic form to mitigate the problem of outliers manipulating the results, we winorize the top and bottom 1% of the dependent variables for robustness. Finally, to prove that there is no deliberate screening of samples in this paper, we rejoin the excluded samples into the regression. After conducting these robustness tests, the core conclusions of this paper remain unchanged, as shown in Tables 4 and 5.

Table 4 Robustness check: balanced sample and other clustering strategy.
Full size table
Table 5 Robustness check: winsorize and unselected sample.
Full size table

Then we relaxed the fixed effects, opting to include only city and year bidirectional fixed effects. The estimation results are presented in Table 1.

Mechanism

Direct mechanism

To further explore the mechanisms by which the operation of power infrastructure affects city form and to more robustly justify the above conclusions, we investigate several mechanisms. The effect of power plant operation can be disentangled into three effects: the pollution effect unique to thermal power plants, the electricity supply effect, and the employment and economic development promotion effect. We use PM2.5 data from the raster data provided by Hammer et al. (2022) (Hammer et al. 2022), electricity generation, and GDP data from Chen et al. (2022) (Chen et al. 2022). The results are shown in Table 6.

Table 6 Direct mechanism.
Full size table

Column (1) indicates that with the operation of thermal power plants, the PM2.5 level of the city was increased by approximately 1.8%, and the direct effect of pollution may result in population loss, particularly around the power plant. Conversely, the estimate in Column (2) shows that with the operation of thermal power plants, the electricity supply of the city was increased by around 4.5%, enabling city planners to develop the city vertically instead of resorting to more expensive horizontal development. Column (3) aims to examine the level of economic development, showing that the GDP of the city increased by approximately 3.8% with the operation of the thermal power plant. These finding highlights that with the operation of the thermal power plants, the local economic has a relatively improvement. This, in turn, can create a large number of related jobs and help explain the increase in the total population and population density of the city to some extent. Indeed, this discovery underscores that at the city level, the role of power plants as infrastructure in attracting population outweighs their adverse impact as a significant pollution source on resident concentration. Therefore, more serious air pollution, easier access to electricity, and the improvement of economic development bring the shape of power plant cities more inclined to vertical expansion than horizontal expansion.

To provide further clarity on this mechanism, in the subsequent further analyses, we will present the distribution of urban residents within cities resulting from the operation of thermal power plants, which will distinctly illustrate the trade-offs made by residents regarding these attributes and interactions.

Affirming mechanisms: gas-fired power plants

To provide a more robust and convincing interpretation of the above findings, it may be beneficial to estimate the impacts of gas-fired power plants as a control group for thermal power plants. Compared to thermal power plants, the pollution effect of gas-fired power plants is negligible. However, due to their smaller generating capacity, gas-fired power plants lag behind thermal power plants in their ability to contribute to the economy and employment. Therefore, if our estimates of the effects of gas-fired power plants align with theoretical expectations, it would lend support to our analytical framework for thermal power plants. The second significance of this part of the estimation is that while traditional energy power plants such as thermal power plants have long been the primary source of electricity generation, the share of electricity generated by renewable and clean energy power plants has grown significantly in recent years. If the operation of thermal power plants can explain, to some extent, the changes in the city’s form in the past, then this part of the estimation can provide predictive and empirical evidence for future changes in urban morphology.

All the data in this section comes from the dataset presented above, and the data sources for gas-fired power plants are the same as those for thermal power plants. The results are shown in Table 7. Columns (1) and (4) indicate that gas-fired power plants have a similar impact to thermal power plants on both horizontal and vertical city expansion. The establishment of a gas-fired power plant results in an approximately 11% reduction in the city’s built-up area, while simultaneously contributing to a significant increase of 11% in the height of urban structures. However, due to their smaller size, gas-fired power plants have a significantly weaker ability to shape the city’s form compared to large-scale thermal power plants. On the other hand, the results in Columns (2) and (3) estimate how pollution-free electricity infrastructure affects population agglomeration. The results in both columns show significantly stronger economic significance than the estimates based on thermal power plants in Table 1, suggesting that the operation of power plants is more attractive to city agglomeration after pollution attributes are removed. The findings articulated above demonstrate that the establishment of thermal power plants and electrical facilities actively encourages vertical urban development, fostering an increase in population agglomeration and density. This phenomenon not only signifies a marked shift away from horizontal expansion but also underscores the emergence of a compact and upward-oriented urban morphology. As cities embrace this strategy of vertical growth, they are able to maximize limited land resources while enhancing the functionality and livability of urban environments. Ultimately, this transformation promotes a more sustainable urban landscape, where the emphasis lies on building upward rather than outward, thereby mitigating the pressures of sprawl in the surrounding area.

Table 7 Estimations for Gas-red Power Plant.
Full size table

Of course, such comparisons involve an issue: cities with gas power plants and those with coal power plants may have intrinsic initial characteristic differences. In the appendix, we constructed appropriate control groups for both gas and coal power plants separately and found that the fundamental results did not exhibit significant fluctuations. Furthermore, we have conducted a similar set of channeling analyses based on the operation of gas-fired power plants.

We expect this part of the analysis to provide more robust evidence on how electricity infrastructure shapes the city’s form and the heterogeneous impacts of power plants with different energy attributes on the city’s form. The results are shown in Table 8. Columns (1) once again demonstrate that gas-fired power plants do not have a catalytic effect on pollution emissions, but even a weak inhibitory effect. This may be due to the cleaner effect of gas-fired power plants on the energy used by city economic activities. Specifically, the presence of gas-fired power plants induces the use of cleaner and more readily available renewable electricity by activity units, leading to the cleaner use of related industries around the power plants. This, in turn, induces cleaner technology innovation and investment in related industries around the plant, generating technology spillovers. Furthermore, Column (3) provides evidence that with the operation of gas-fired power plants, the regional economic development was increased by 1.1%, which has a similar economic impact to thermal power plants in most respects, except regional pollutant emissions to some extent. This estimation directly supports how gas-fired power plants affect the city’s form, as shown in Table 3, alongside evidence for the effects of thermal power plants on the city’s form. Furthermore, Column (3) provides evidence that gas-fired power plants contribute to regional economic development. In summary, gas-fired power plants have similar economic impacts to thermal power plants in most respects, but with the significant difference that gas-fired power plants help mitigate regional pollutant emissions to some extent. This estimation directly supports how gas-fired power plants affect the city’s form, as shown in Table 2, alongside evidence for the effects of thermal power plants on the city’s form.

Table 8 Estimations for Gas-fired Power Plant in Mechanisms.
Full size table

By employing a similar estimation framework to that of thermal power plants, our empirical findings further corroborate the proposed mechanisms. Considering the extensive deployment of new energy systems, this research contributes valuable insights for predicting future urban morphology.

Impacts on population within cities

Power plants facilitate urban agglomeration primarily through two synergistic mechanisms: industrial clustering and the provision of electricity. Large-scale factories and industrial infrastructures significantly contribute to industrial agglomeration (Ellison et al. 2010), with labor concentration being a notably prominent effect (Rosenthal & Strange, 2001). This aggregation may stem from the ‘pollution haven’ effect, where enterprises cluster around large polluters to escape stringent environmental regulations, thereby transferring pollution (Zeng & Zhao, 2009). It may also arise from reduced transaction and transportation costs (Ellison et al. 2010), particularly relevant in industries such as ore transport near power plants. The second mechanism is the direct availability of electricity from power plants. Power companies often supply electricity to nearby enterprises via dedicated lines, offering energy at rates lower than market prices and with greater stability. Furthermore, the construction of new power plants typically targets alleviating regional power shortages. While some urban power requirements are met through the grid, additional sources are necessitated by factors like voltage stability and the cost of circuit installations, promoting regional power development.

The establishment of large factories also significantly alleviates employment issues. A primary source is the unemployed population around the power plant, who can choose to participate in work nearby (Monte et al. 2018), especially in resource-intensive, labor-intensive industries like power plants that do not have high technical demands or barriers. Furthermore, there is the migration for job opportunities. When commuting times exceed a certain threshold, local residents may choose not to commute but instead move closer to their places of work to minimize commuting costs (Huang et al. 2018).

The interplay between industrial and electricity agglomeration, compounded by the pollution from thermal power plants, renders the impact of power plants on urban agglomeration multifaceted and intriguing.

The previous discussion primarily focused on the average treatment effects of power plant operation at the city level. In other words, when explaining urban morphology, we have qualitatively attributed how thermal power plants affect urban form to several attributes with opposite effects. If we could further analyze the impact of power plant operation within cities, it would directly allow us to examine how individuals respond to the various dimensions of changes brought about by power plant operation, providing us with spatial evidence of several power plant attributes. This, in turn, helps us gain a more intuitive understanding of the mechanisms through which power plants influence urban morphology. Furthermore, this is a direct response to the topic, as studying the population distribution within cities is equally important as studying the overall urban form at the city level. This section will specifically focus on examining the impact of thermal power plant operation on population distribution across different distances.

To investigate this, we consider the latitude/longitude location of the power plant as the center of a circle and calculate the average total population within the circle using adjacent circles with different radii as basic population observation units. We will then examine how the operation of the power plant affects the population size within circles of varying radii. The regression model used for this analysis will be the same as the baseline model, as follows:

$${Population}_{rpt}=\alpha +\,\beta {Plant}_{pt}+\,\delta {Z}_{ct}+\,{\epsilon }_{p}+\,{\delta }_{c* t}+\,{\varepsilon }_{rpt}.$$
(3)

where the dependent variable of interest, Populationrpt, is the population density in the circle centered on the completed power plant p in year t, with r radii. The reason we use circles of different radii rather than annuli (ring-shaped regions) as the observation units is that if we were to use annuli as the observation units, it is very likely that the population within adjacent two-kilometer circles may not change. In this case, the population within the annulus would be zero, making it impossible to take the logarithm again, resulting in a large loss of samples and biased sample selection. In fact, by using circles as the observation units, we can fully understand the impact of power plants on the population density at different distances around them through the marginal differences in the effect of power plants between adjacent radii. Plantpt describes whether a power plant was established at p in year t, and is taken to be 1 if it was, and 0 otherwise. For a more visual representation of the results, we plot the estimates corresponding to the different circles in Fig. 5.

Fig. 5
figure 5

The impact of power plants on surrounding population density.

Full size image

From Fig. 5, it is evident that the operation of power plants has a clear impact on the surrounding population density, decreasing almost monotonically with increasing distance. Specifically, within the closest two-kilometer range, the operation of power plants results in a population density increase of over 40%. We believe that this reflects primarily the employment absorption effect of power plants as infrastructure within this small range, which does not fully capture the characteristics of thermal power plants. Beyond three kilometers, the promoting effect of power plants on population density decreases almost linearly with distance. If we consider a monitoring area of 20 kilometers, the impact of power plant operation on population growth becomes considerably weaker. The heterogeneity in the impact of power plant operation on the surrounding population at various distances precisely reflects the varying benefits derived from such operation for different population groups at different distances.

Certainly, to further substantiate the aforementioned perspective, we examined the trends in the variation of power plant attributes with distance. Specifically, we selected two representative attributes of the power plant for validation, namely, generating capacity and air pollution. Using the electricity consumption data and PM2.5 data mentioned earlier, we constructed average electricity consumption intensity and average pollution levels within the different radius. The specific model aligns with Eq. (3).

From Fig. 6, we observe that both electricity consumption levels and pollution levels decrease as the distance from the power plant increases. However, this alone does not provide us with a relative change in power plant attributes at a distance level. Therefore, we normalized the estimated coefficients of the power plant’s influence on electricity enhancement levels by the estimated coefficients of its influence on pollution enhancement levels within different concentric circles. This allows us to represent the relative strength of these two defining attributes.

Fig. 6
figure 6

The impact of power plants on surrounding electricity consumption and pollution levels.

Full size image

After calculating the ratio of the coefficients for these two attributes in Fig. 7, we gained a more direct understanding of why the power plant influences the surrounding population distribution as described above. In summary, the pollution levels caused by the power plant decrease at a slow rate as the distance increases, which aligns with our existing knowledge. Current literature even suggests that air pollution from one country can affect the living conditions of residents in neighboring countries (Heo et al. 2023), indicating the inherent difficulty of pollution dissipation. Conversely, the increase in electricity consumption brought about by the power plant sharply decreases with increasing distance. Within the closest two kilometers of the power plant, as we mentioned earlier, the main driving factor for the population increase in this area is closely related to employment associated with the power plant, such as its workforce. Therefore, the population changes in this small area cannot be fully explained by the power supply and pollution effects of the power plant alone. On the other hand, in areas beyond three kilometers, the relative coefficients of power supply improvement and pollution levels from the power plant, compared to the coefficient for population increase, show a consistent trend. This reflects that individuals weigh the positive and negative effects of the power plant based on their own preferences. It greatly helps us understand how power plants influence the form of both individual urban areas and the entire city as a whole.

Fig. 7
figure 7

The relative impact of power plants on surrounding electricity consumption and pollution levels.

Full size image

Heterogeneity analysis

The next step involves conducting a multidimensional heterogeneity analysis to further explore the impact of electricity infrastructure on city sprawl across different regions globally. We analyze sequentially across four dimensions as follows: first is the shaping capacity of urban morphology, closely associated with transportation infrastructure highly relevant to electricity infrastructure; second is the economic development level; third, the government’s environmental regulatory stringency closely related to the pollution capacity of coal-fired power plants; and finally, the natural constraints that are hard for human intervention. This analysis enables us to gain a deeper understanding of the effects of power plant operation and provides additional insights to corroborate the robustness of the aforementioned discussion from a different perspective.

Transportation infrastructure

The first set of heterogeneity analyses conducted in this paper focuses on examining the complementary role of electricity infrastructure and transportation infrastructure. Previous studies have highlighted that transportation facilities, driven by high commuting costs, are the primary driver of horizontal city sprawl (Batty et al. 2003). Therefore, it is essential to understand the heterogeneity in the impact of compact electricity infrastructure on city form in regions with varying levels of transportation infrastructure. Addressing this question not only complements the existing literature on electricity infrastructure and transportation infrastructure but also provides crucial evidence on how these two factors can work together to shape city form. To measure the level of regional transportation infrastructure, the paper employs road network density as a proxy variable. The global-level road vector map is utilized to obtain the necessary data, and the road length-area ratio is calculated for each region (Center for International Earth Science Information Network, 2013). The regions are then divided into quintiles, namely Groups 1-5, based on the density of the road network, ranging from low to high. The subsequent analysis follows the same approach, and the baseline regression equations are estimated for each group.

Figure 8 illustrates the moderating effect of transportation infrastructure on the impact of power infrastructure. Firstly, it shows that the inhibitory effect of power plant operation on horizontal development is significantly reduced as road network density increases. This reduction can be attributed to the diminishing agglomeration effect caused by power plants when commuting costs are lowered. Secondly, the figure demonstrates that the development of city agglomeration resulting from power plant operation, whether measured by total population, population density, or building height, is observed primarily in areas with lower road density.

Fig. 8
figure 8

Heterogeneity analysis: transportation infrastructure level.

Full size image

Next, we consider a specific form of electrified railway: the subway. If road network density emphasizes overall accessibility within a region, then subways focus on the convenience of intra-urban transport. After all, the operation of subways must consider cost-benefit analysis, and more densely populated areas are more likely to have subways built.

In terms of data search, city-level subway data is exceedingly rare, often dependent on non-public administrative data provided by various governments. Another concern is the endogeneity between subway operation and power plant operation; that is, the operation of power plants provides strong infrastructure support, which in turn leads to more developed subway networks. To avoid this potential endogeneity issue, we employed the ITT (Intent-to-Treat) analysis technique. Specifically, we used search engines to gather data on when each country built its first subway line, with primary data sources from https://web.archive.org/web/20070607170931/http://www.cityrailtransit.com/list.htm and https://web.archive.org/web/20181227084909/http://www.urbanrail.net/. This means that once a country has its first subway, it inherently possesses the technology and conditions to build subways. Our ITT results estimate what is essentially the lower bound of the treatment effect. This itself is akin to a triple difference estimation framework, covering two levels of impacts: power plant impacts and subway opening impacts. Therefore, we included an interaction term between power plant cities and subway countries to estimate the average treatment effect. The estimated results are shown in Table 9.

Table 9 Heterogeneity Analysis: Metro Line.
Full size table

We have identified a clear fact regarding subways, an infrastructure explicitly aimed at facilitating convenience for urban residents. Our ITT estimation results suggest that once a country has the capability to construct subways, the operation of power plants in that country subsequently has a stronger influence on urban agglomeration. This demonstrates the synergistic effect of advanced infrastructure development on enhancing urban concentration and development.

Economic development

The second heterogeneity analysis explores the moderating role of economic development levels. In regions with lower economic development, prioritizing growth often overshadows environmental concerns, promoting polluting industries such as thermal power plants, which significantly impact local employment and development. Consequently, these plants are linked to increased local employment demands and population agglomeration. However, less economically developed areas usually depend on horizontal expansion due to the lack of infrastructure necessary for vertical growth. Since vertical development is closely linked to economic levels, regions with insufficient development often cannot support extensive vertical expansion, focusing instead on horizontal growth to accommodate increases in population.

To test the hypotheses and analysis mentioned above, we utilize the average real GDP data provided by Chen et al. (2022) (Chen et al. 2022) at the city level from 1992 to 2019 as a measure of economic development. This data serves as the basis for conducting group regression analysis.

The regression results presented in Fig. 9 confirm two key findings. Firstly, the operation of thermal power plants exhibits a significant population agglomeration effect in economically backward regions. In contrast, this effect is much lower in more developed regions. This discrepancy can be attributed to the regulatory environment surrounding polluting industries in countries with lower economic levels. In such contexts, government regulations on environmentally harmful activities tend to be less stringent. Consequently, large-scale projects like power plant operations are utilized to stimulate employment and drive local economic development. Secondly, the restraint on city expansion resulting from thermal power plant operation is not evident in economically underdeveloped regions. This is primarily due to the fact that city expansion in these regions still heavily relies on horizontal expansion strategies, lacking the necessary infrastructure and capacity for significant vertical expansion.

Fig. 9
figure 9

Heterogeneity analysis: economic development level.

Full size image

Environmental regulation

The third group of heterogeneity analysis focuses on the effects of power plant operation in countries with varying levels of environmental regulation. The government’s attitude towards pollution plays a crucial role in determining the power generation capacity and pollution emission levels of conventional thermal power plants. To capture the regional environmental regulation level, we utilize the latest national Environmental Protection Index (EPI) as a proxy variable (Wolf et al. 2022). This index encompasses dimensions such as Climate Change, Environmental Health, Ecosystem Vitality, and several minor dimensions, which comprehensively assess both the strength of government environmental regulations and the pollution levels within a region. It’s worth noting that pollution levels are inherently linked to the regulatory environment, leading some research to employ pollution levels as an indirect reflection of government regulatory stringency in a particular dimension (Domazlicky & Weber, 2004). Therefore, EPI is a suitable indicator to gauge the regional government’s commitment to environmental governance and the effectiveness of their efforts. Based on the environmental protection index, we divided the samples into five groups, ranging from low to high environmental performance. The results of this analysis are presented in Fig. 10, highlighting the variations observed across the different groups.

Fig. 10
figure 10

Heterogeneity analysis: environmental regulation level.

Full size image

Our analysis yields two key findings. First, the impact of power plant operation on total population and population density follows an inverted U-shaped trend with the EPI. At both the lowest and highest ends of environmental performance, the effect on population is negligible. In countries with low environmental performance, the presence of polluting thermal power plants without adequate pollution controls, such as desulfurization equipment, drives residents away due to health concerns. Conversely, in countries with high environmental performance, thermal power plants are heavily regulated, limiting the job opportunities they generate as these countries favor advanced technology over labor-intensive practices (Fiszbein et al. 2020). The positive impact of power plants on population in these countries is economically and statistically minimal. Moreover, the influence of thermal power plants on both horizontal and vertical urban expansion is significantly reduced in regions with excellent environmental performance, due to strict controls on their operation and scale. Thus, the ability of power plants to influence urban form is diminished in areas with the highest environmental standards.

Geographical factors

In the third aspect of heterogeneity, we focus on the moderating role of geography. Geography plays a crucial role in shaping the form of cities, as cities are often built on topography, and the cost of expansion in rugged areas is typically high (Burchfield et al. 2006). Therefore, in this study, we examine the influence of geographic factors by utilizing terrain ruggedness as a starting point. To assess this, we utilize terrain ruggedness raster data and divide the data into five groups based on terrain ruggedness (Nunn & Puga, 2012), ranging from low to high. By examining the regression results presented in Fig. 11, we can observe that the operation of power plants has a particularly significant impact on compact city development in areas with high ruggedness. This finding reflects the fact that rugged terrain increases the cost of horizontal expansion, leading to a greater emphasis on vertical expansion strategies in these areas. Consequently, the operation of power plants provides the necessary infrastructure support, such as electricity, enabling cities to adopt a more pronounced vertical expansion approach, resulting in higher population densities.

Fig. 11
figure 11

Heterogeneity analysis: ruggedness.

Full size image

We explore the moderating role of geographic factors by examining the likelihood of earthquakes, using historical data from the US Geological Survey (USGS) on global earthquakes of magnitude 5 or greater from 1900 to 1970. This period is chosen to precede our benchmark regression period, avoiding potential endogeneity, and to provide a clearer picture of historical earthquake risks. Based on this data, we categorize regions by historical earthquake occurrence and present subgroup regression results in Fig. 12. Our findings indicate that regions with higher seismic risks show a notable decrease in labor absorption capacity for power plant projects, suggesting that high earthquake risks deter labor attraction. However, we find no varying effects of power plant operations on the urban expansion patterns in regions with different seismic risks, implying that cities prone to earthquakes might employ adaptive strategies, like using seismic-resistant materials. There is also no geological evidence to support that high-rise buildings are less earthquake-resilient.

Fig. 12
figure 12

Heterogeneity analysis: earthquake risk.

Full size image

Conclusion

We conducted a comprehensive study to assess the impact of thermal power plants on urban sprawl. Our findings reveal that cities with large thermal power plants exhibit a 20.2% reduction in the built-up area compared to cities without such plants. Additionally, the total population and population density experience a respective increase of 1.9% and 7.6%, while the average building height rises by 18.7%. These results indicate that the operation of power plants significantly contributes to urban densification. To support our conclusions, we conducted a series of robustness checks.

Mechanism analysis revealed that power plant operation has a significant influence on pollution levels, electricity supply, and economic development in cities. As a non-polluting control group, gas-fired power plants demonstrate a greater ability to attract residents. Overall, our study provides valuable insights into the effects of thermal power plants on urban morphology, highlighting the role they play in shaping cities and their associated dynamics. Besides, we observed a distinct linear relationship between the population density residing near the thermal power plant and the distance from it. This finding provides further evidence supporting the existence of a pollution mechanism associated with the power plant.

To reinforce our conclusions and mechanisms, we conducted a series of heterogeneity analyses across four dimensions: transportation infrastructure level, economic development level, environmental performance, and natural factors. These analyses provided additional support for our findings. In conclusion, our research suggests that as power infrastructure continues to expand and improve, power access rates and capacity rise, and environmentally friendly power sources flourish, the trend of horizontal city expansion in the future will significantly slow down or even reverse. Instead, we anticipate a shift towards vertical city development and population agglomeration.

The primary conclusion of this study suggests that with the continuous improvement of electricity infrastructure, the trend of horizontal urban expansion will gradually slow down and transform into vertical expansion. Firstly, electricity serves as the key enabler of population agglomeration and vertical urban development. Under a given budget constraint, the siting of large-scale government electricity infrastructure can prioritize cities that aim to foster vertical agglomeration. Secondly, although renewable energy facilities are smaller in scale, their capacity to shape urban form is not inferior to that of thermal power plants. With the future reduction in renewable energy costs and its pollution-free advantages, renewable energy plants can also become a priority for governments in shaping urban form. Lastly, due to the horizontal expansion of global cities, which leads to the loss of arable land and other issues, there exists a significant conflict between human activities and the natural environment. In this study, the operation of power plants has been shown to reduce the trend of urban expansion and promote vertical development. Therefore, governments can use the establishment of power plants to curb horizontal urban expansion, while enhancing vertical expansion to mitigate the conflicts between human activities and ecological concerns arising from horizontal expansion.

This paper acknowledges several shortcomings: First, the analysis of urban morphology is largely limited to the dichotomy between flat and vertical forms, thereby overlooking the potential impact of alternative spatial configurations, which could provide further insights into urban development patterns. Furthermore, the challenges associated with obtaining global grid data related to urban dimensions—specifically in terms of industry compositions and investment flows—have constrained the investigation into more granular mechanisms. Consequently, future research could profit from a more in-depth exploration of these micro-level dynamics, particularly with respect to the intricate relationships between industrial structures and investment strategies within varied urban contexts.