Regional Urban Shrinkage Can Enhance Ecosystem Services—Evidence from China’s Rust Belt

Abstract

Rapid urbanization is universally acknowledged to degrade ecosystem services, posing significant threats to human well-being. However, the effects of urban shrinkage, a global phenomenon and a counterpart to urbanization, on ecosystem services (ESs) remain unclear. This study focuses on China’s Rust Belt during the period from 2000 to 2020, constructing a comprehensive analytical framework based on long-term remote sensing data to reveal the temporal and spatial patterns of ESs and their associations with cities experiencing varying degrees of shrinkage. It employs a random forest (RF) model and a Shapley additive explanation (SHAP) model to measure and visualize the significance and thresholds of socioeconomic factors influencing changes in ESs. Our findings highlight the following: (1) Since 2010, the three provinces of Northeast China (TPNC) have begun to shrink comprehensively, with the degree of shrinkage intensifying over time. Resource-based cities have all experienced contraction. (2) Regional urban shrinkage has been found to enhance the overall provision capacity of ESs, with the most significant improvements in cities undergoing continuous shrinkage. (3) The impact of the same socioeconomic drivers varies across cities with different levels of shrinkage; increasing green-space ratios and investing more in public welfare have been identified as effective measures to enhance ESs. (4) Threshold analysis indicates that the stability of the tertiary sector’s proportion is critically important for enhancing ESs in cities undergoing intermittent shrinkage. An increase of 10% to 15% in this sector can allow continuously shrinking cities to balance urban development with ecological improvements. This research highlights the positive aspects of urban shrinkage, demonstrating its ability to enhance the provision capacity of ESs. It offers new insights into the protection and management of regional ecosystems and the urban transformation of the three eastern provinces.

1. Introduction

The concept of “urban shrinkage” was characterized primarily by population loss and economic decline, accompanied by negative externalities such as housing vacancies and environmental degradation [1]. Today, over a quarter of cities worldwide are shrinking [2]. Notably, in China, which has the largest scale of urbanization globally, as many as 52% of its districts and counties are experiencing varying degrees of shrinkage [3], forming distinct regional patterns. This intensifying phenomenon poses challenges to achieving United Nations Sustainable Development Goals (SDGs) 8, 11, and 15 [4]. Scholars from disciplines such as sociology, economics, geography, and urban planning are increasingly turning their attention to this emerging phenomenon, which diverges from traditional urban succession trajectories. These scholars have engaged in multi-scalar analyses encompassing conceptual delineation, classification and identification, causative factors, and strategic responses to urban shrinkage [5].Additionally, they have formed research collectives like the Shrinking Cities International Research Network (SCiRN) to facilitate a more coordinated exploration of these issues. The persistent progression of urban shrinkage continues to exert profound influences on socioeconomic conditions and urban environmental development. Consequently, scholarly attention has progressively shifted from merely documenting the objective manifestations of shrinkage to examining the well-being of residents within these shrinking cities. This includes investigations into the impacts of green-space utilization and housing-vacancy durations on health disparities among urban populations [6], social isolation and spatial segregation due to social fragmentation [7], and the effectiveness of social infrastructure in bolstering social resilience [8]. Bernt posits that, irrespective of its causes and outcomes, the transformation of urban life brought about via shrinkage garners more attention [9]. In the current paradigm of human-centered development, a deeper understanding of urban shrinkage through the lens of resident well-being is imperative.

ESs are benefits derived from the natural environment [10], and their high-quality provision is crucial for ensuring urban livability. Rapid urbanization has encroached extensively on ecological spaces, significantly impairing the provision capacity of ESs. Over the past half-century, approximately 60% of ecosystems on Earth have experienced significant degradation [11]. Scholars have conducted extensive research on the impact of urbanization on ESs at various scales, from global [12] to drainage basins [13]. Urban shrinkage, as an aspect of urbanization, also leads to changes in land-use types, patterns, and intensities [14], disrupting the material and energy cycles of existing ecosystems. Recent studies increasingly address the impacts of urban shrinkage on ecological environments, primarily addressing specific topics such as carbon emissions, air pollution, and urban heat island effects. Research by Xiao categorized cities in Northeast China and the Yangtze River Delta into shrinking and expanding types to examine carbon-emission discrepancies. They noted that, while urban shrinkage reduces overall carbon emissions, energy efficiency declines due to population losses, land vacancy, and decreased urban compactness [15]. Peng analyzed the urban heat island effect in 180 shrinking Chinese cities, demonstrating that population reduction mitigates this effect [16]. In contrast, Rao and colleagues, by tracking changes in PM2.5 concentrations over a decade in 174 prefecture-level cities in China, found that urban shrinkage exacerbates air pollution, with this phenomenon being particularly pronounced in the Northeast part of China [17]. Sun and colleagues discovered that urban shrinkage negatively impacts ecological efficiency, while upgrading the industrial structure mitigates the extent of this impact [18]. However, currently, few studies have explicitly delineated the relationship between urban shrinkage and ESs, as well as the factors that drive the emergence and development of this relationship, creating a stark contrast with the increasingly intensifying phenomenon of shrinkage. In previous studies, Haase and colleagues only conceptually discussed the tremendous potential ecological value of the idle spaces created by urban shrinkage [19]. If these spaces are redeveloped into green infrastructure, they can offer urban benefits such as climate regulation and enhanced biodiversity. Lauf, through simulating future land-use developments in Berlin under scenarios of growth and shrinkage, found significant improvements in ESs in the shrinkage scenario [20]. In quantitative research, scholars primarily explore the effects of varying levels of human activity on ESs. These studies have highlighted areas with significant population distribution disparities, such as the regions on either side of the “Hu Huanyong Line”, and nature reserves where a population decline is evident [21,22]. The results confirm a significant negative correlation between anthropogenic pressure and habitat quality [23]. However, these studies tend to focus exclusively on habitat quality as a singular service without adequately considering the diversity of ESs. Theoretically, urban shrinkage has both positive and negative impacts on ecosystem services. Scientifically quantifying and identifying these positive and negative effects, along with their interaction mechanisms, are crucial for enhancing the refined management of ecological environmental systems in shrinking cities and promoting their sustainable development.

Accurately identifying and classifying shrinking cities is essential for quantifying the relationship between urban shrinkage and the provision capacity of ESs. Population change is widely used as a critical indicator for assessing urban shrinkage [24]. However, urban shrinkage is an exceedingly complex phenomenon, and a single demographic indicator may not fully elucidate all aspects of urban shrinkage [9,25]. Urban development is a long-term dynamic process. The variability in the degree, scale, and temporal trajectory of shrinkage across different periods contributes to the diversity of urban shrinkage [26]. Therefore, constructing a multidimensional identification system that spans various times and spaces is crucial. Nighttime light data (NTL), which are highly correlated with population and socioeconomic development, offer a novel perspective on human activities. Their high precision and long-term characteristics can reduce potential research errors associated with cross-sectional data and statistical data, which arise from changes in administrative divisions or data missing, making them highly suitable for the long-term monitoring of urban development dynamics [23,27,28]. While existing research has significantly deepened the scholarly understanding of this topic, such as Zhou’s analyzing the shrinkage pattern of Yichun City, China, from 2012 to 2019 [29] and Niu and Wang’s identification of shrinking cities in the Yellow River Basin and nationwide from 2013 to 2018 [30], there is still room for improvement in fully leveraging the time-series characteristics of the data. Furthermore, some scholars argue that using this data to study urban shrinkage may introduce biases, as its variations are more closely linked to economic activities, rather than aligning consistently with population changes [31]. In regional-scale research, it is advisable to integrate additional data to form composite indicators and analyze urban shrinkage from a developmental-trend perspective.

Research on ESs should address two primary issues: the quantification of ESs and the exploration of their driving forces. Methods such as ESV [32] and the InVEST model [33,34] are widely used to understand the strengths and weaknesses of ES provision capacity. Changes in ESs are driven by a complex set of factors, and identifying the core drivers is crucial for their sustainable utilization. Numerous studies have demonstrated that natural environmental factors such as climate change and topography [35] have significant impacts on ES changes. Increasingly frequent human activities also play an increasingly important role in altering ecosystem services, with potential threshold effects in these changes [36]. Currently, various methods have been employed to investigate the specific factors causing changes in services, such as generalized additive models and piecewise linear regression, which are based on prior assumptions to fit relationships between variables. However, these methods often overlook the complexity of ESs, and their development and evolutionary processes are difficult to accurately predict through linear models [37]. Machine learning models like random forests (RFs) [38] and gradient-boosting decision trees (GBDTs) [39], which are widely used to analyze nonlinear relationships and threshold effects between variables, are rarely applied in studies concerning ecosystem services. This gap provides a strong foundation for this paper’s exploration of the nonlinear and threshold relationships between various factors and ESs. Additionally, the academic skepticism regarding machine learning models as “black box” algorithms has hindered their effective utilization. Rendering their learning processes into model-agnostic, visual interpretations that transform these models into “glass boxes” could enhance their application in urban ecosystem services research. The diversity in urban shrinkage also dictates variations in influencing factors. Clarifying the differences in driving factors among various types of cities can assist planners in devising more suitable strategies. A multi-dimensional, dynamic perspective is crucial for an accurate and comprehensive understanding of the impact of regional urban shrinkage on ecosystem services.

The three provinces of Eastern China (TPNC), often referred to as China’s “Rust Belt”, are a typical region of shrinking cities [40]. During the planned economy era, these provinces experienced rapid development fueled by heavy industry, with mining and manufacturing sectors providing substantial employment for local residents. Due to the depletion of natural resources, national economic reforms, and the shift in the global economic center [41], the TPNC’s industrial structure, which is heavily centered on heavy industry, has gradually become imbalanced and increasingly sluggish. The decline in these dominant industries has triggered a series of “domino effects”: severe economic recession, continuously rising unemployment rates, and ongoing population outmigration, resulting in abandoned housing and urban brownfields. Simultaneously, reform and opening-up policies have substantially enhanced the capacity of cities in southern China to attract development resources, enabling them to emerge as new growth poles and attracting significant population inflows. In contrast, the TPNC, already weakened due to declining developmental momentum, have experienced continued deterioration under the combined pressures of internal push factors and external pull factors. Moreover, the economic development of old industrial areas is often characterized by “high pollution, high emissions, and high energy consumption”. Severe industrial pollution has exerted immense pressure on the ecosystems of heavy industrial regions [42], resulting in a significant imbalance between the supply and demand of ESs in the TPNC and increasing regional ecological risks. The urban development and transformation in these areas face the dual challenges of economic decline and ecological degradation. In view of the foregoing discussion, we hypothesized that regional urban shrinkage could enhance ESs. To test this hypothesis, this study focused on the TPNC, starting with a classification of the study area through the analysis of long-term remote sensing data. It then used an RF model to explore the importance of various factors in the process of changes in ESs. Subsequently, the SHAP model was employed to reveal the nonlinear relationships and threshold effects of these factors. Ultimately, a comprehensive analytical model was constructed from a dynamic perspective, quantifying the relationship between urban shrinkage and the provision capacity of ESs. Our main objectives were as follows: (1) to identify the specific circumstances of urban shrinkage within the study area and categorize the cities; (2) to quantitatively analyze the spatiotemporal evolution characteristics of ES provision; (3) to explore the impact of various dimensional factors on ES provision in cities at different stages of shrinkage and determine optimal thresholds; (4) to view the phenomenon of regional urban shrinkage from a positive perspective. The results of this study provide new insights into research related to urban shrinkage and changes in ESs, and they offer planning guidance for the sustainable development of regionally shrinking cities.

2. Study Area and Data Sources

2.1. Study Area

Known as the “cradle of industrial China”, the three northeastern provinces (TPNC), including Heilongjiang (HLJ), Jilin (JL), and Liaoning (LN), have made significant contributions to the country’s economic development with their rich coal, oil, and natural gas reserves. Cities like Anshan and Benxi laid the foundation for China’s steel industry, while Fuxin, Fushun, and Hegang primarily drove their economies with coal mining. Yichun and the Greater Khingan Range became important timber bases due to their abundant forests. However, comparing data from the sixth and seventh national censuses, the population of this region decreased by 10.9982 million people over ten years, accounting for 15.28% of China’s total population increase. The GDP share of these provinces also declined from 6.94% in 2010 to 5.03% in 2020. The “2020 Key Tasks for New Urbanization Construction and Integrated Urban-Rural Development”, along with various scholarly studies, clearly indicate that regional urban shrinkage in Northeast China has evolved into an objective geographical reality [43]. Additionally, fertile soil makes the TPNC an important grain base for China and Northeast Asia. The northeastern forest belt is also a crucial component of China’s ecological security strategy, and the importance of its ESs cannot be overlooked. By investigating the relationship between ESs and regional urban shrinkage in the “Chinese Rust Belt”, we can derive representative conclusions. This paper focuses on 34 prefecture-level cities with complete data (excluding the Greater Khingan Range and Yanbian Korean Autonomous Prefecture, Figure 1), covering the period from 2000 to 2020.

2.2. Data Sources and Processing

This study utilized two types of data: natural environment data to quantify ecosystem services and socioeconomic data to explore driving factors (

Table 1. Data sources and descriptions.

Data Type	Data Name	Data Resource	Resolution
Natural Environment Data	Land-Use Data (GLC_FCS30 Dataset)	http://aircas.ac.cn/ (Accessed on 10 March 2024)	30 m × 30 m
	Digital Elevation Model (DEM)	https://www.resdc.cn/ (Accessed on 12 March 2024)	30 m × 30 m
	Precipitation Data	https://www.geodata.cn/ (Accessed on 15 March 2024)	1 km × 1 km
	Evapotranspiration Data	https://www.geodata.cn/ (Accessed on 15 March 2024)	1 km × 1 km
	Normalized Difference Vegetation Index (NDVI)	https://www.nesdc.org.cn/ (Accessed on 11 March 2024)	30 m × 30 m
	Net Primary Productivity (NPP)	https://www.usgs.gov/ (Accessed on 11 March 2024)	500 m × 500 m
	Soil Attribute Data	https://gaez.fao.org/pages/hwsd (Accessed on 20 March 2024)	1 km × 1 km
	Watershed Data	https://hydrosheds.org/ (Accessed on 15 March 2024)	Vector file
Socioeconomic Data	Administrative Division Data	https://www.resdc.cn/ (Accessed on 5 March 2024)	Vector file
	Population Density Data	https://www.worldpop.org/ (Accessed on 10 March 2024)	1 km × 1 km
	Other Statistical Data	①China urban statistical yearbooks (2000–2021) ②The statistical yearbooks of Heilongjiang, Jilin, and Liaoning (2000–2021) ③The sixth and seventh population census bulletins of China (http://www.stats.gov.cn/) (Accessed on 3 March 2024)	Non-spatial data; the Arcgis10.2 was used for spatialization
	Nighttime Light Data (NTL)	A Prolonged Artificial Nighttime-light Dataset of China (1984–2020) (http://www.geodata.cn) (Accessed on 13 March 2024)	500 m × 500 m

). The land-use data were derived from the GLC_FCS30 dataset, created by Zhang et al. using time-series Landsat 8 imagery and high-quality training data from the GSPECLib (Global Spatial Temporal Spectra Library) on the GEE (Google Earth Engine) platform [44]. This dataset includes 29 land-cover categories with an overall accuracy of 82.5% and a Kappa coefficient of 0.784, demonstrating strong performance when compared with different datasets. Based on the natural geographic characteristics of the study area and the research objectives, the land cover data were reclassified into six categories: cropland, forestland, grassland, water and soil, built-up land, and idle land. Soil data, including soil texture, sand content, plants’ available water capacity (PAWC) range, organic carbon content, and root-restricting layer depth, were sourced from the China Soil Dataset (v1.1), which is based on the Harmonized World Soil Database (HWSD).

For socioeconomic data, population density and nighttime light data were used to identify shrinking cities, while driving factors were derived from statistical data. After all the data were spatialized, to eliminate errors arising from calculations under different coordinate systems, all data were reprojected to the Albers_Conic_Equal_Area coordinate system using ArcGIS 10.2, and all raster data were resampled to a 1 km × 1 km resolution to facilitate subsequent data analysis.

3. Research Methods

The research process and framework are depicted in Figure 2. The methodology primarily encompassed four components: collecting research data, coding types of urban shrinkage, quantifying ESs, and constructing research models (OLS, RF, and XGBoost). Additionally, the learning outcomes were interpreted using the SHAP (Shapley additive explanations) model.

3.1. Coding Urban Shrinkage Types

We divided the study period into T₁ (2000–2010) and T₂ (2010–2020) periods. Considering the dynamics and persistence of urban shrinkage, this study quantified its extent through ongoing population decline and economic downturn. We categorized and coded cities using a comprehensive shrinkage index, which was constructed through trend analysis of long-term NTL and population density data.

3.1.1. Quantifying Urban Shrinkage Trends

Theil–Sen median (Sen) analysis and Mann–Kendall (M-K) trend tests do not require assumptions of linearity or normal data distribution, and they are insensitive to missing and outlier values in time series. Due to their robustness and non-parametric nature, these methods have been widely used in trend determination for long-term time series data, such as in analyses of climate change [45] and vegetation cover [46]. Scholars have also applied these methods to measure phenomena such as urban “ghost towns” and human development patterns [47], making them suitable for identifying processes of urban shrinkage. Detailed formulas are available in Supplementary Material S1.

First, Sen-MK trend analysis was applied to segment and conduct the long-term monitoring of the trends in NTL and population density data from 2000 to 2020 within the study area. These analyses identified the distinct shrinkage processes and intensities for each city. Subsequently, values were assigned to determine the shrinkage trends D_i in various aspects for each grid cell, following the specific rules outlined in

Table 2. Assignment principle and explanation.

β	Z	Value	Classification Meaning
β > 0	2.58 < Z	4	Extremely significant increase
	1.96 < Z ≤ 2.58	3	Significant increase
	1.65 < Z ≤ 1.96	2	Slightly significant increase
	Z ≤ 1.65	1	Not significantly increased
β = 0	Z = 0	0	No changes
β < 0	Z ≤ 1.65	−1	Not significantly reduced
	1.65 < Z ≤ 1.96	−2	Slightly significant reduction
	1.96 < Z ≤ 2.58	−3	Significant reduction
	2.58 < Z	−4	Extremely significant reduction

.

3.1.2. Definition of the Comprehensive Urban Shrinkage Index

Considering the varying extents of urban areas and intensities of human activities, this study utilized scale-independent mean values to reflect average developmental trends. Accordingly, the population shrinkage trend, S_P, and economic shrinkage trend, S_E, within each city were measured according to the mean values of two types of grid data over different periods, i. Recognizing the equal importance of these two factors, we employed an equal-weight superposition method to determine the comprehensive shrinkage index, ${CSI}_i$. The calculation formulas are as follows:

$$S_P={\displaystyle \frac{{\sum }_{i=a}^N{D}_P}{N}}$$

(1)

$$S_E={\displaystyle \frac{{\sum }_{i=a}^N{D}_E}{N}}$$

(2)

$${CSI}_i=S_P+S_E$$

(3)

where N is the number of grid cells contained in each city, and i represents the a^th time period.

3.1.3. Coding Urban Shrinkage Trajectories Based on the ${CSI}_i$

The Natural Breaks method was used to divide the ${CSI}_i$ for each time period into four levels. These levels were sequentially coded from 1 to 4, representing severe shrinkage, moderate shrinkage, mild shrinkage, and slight development, respectively. Subsequently, the coded ${CSI}_i$ values were arranged in order to determine the shrinkage trajectories of each city. Additionally, to mitigate the bias in stage division, the overall CSI_f values from all study periods were used to assist in determining the final coding for each city. Specifically, if a total of i time periods was obtained, the shrinkage trajectory coding K for each city was the sequential arrangement of all indices. The final digit of the code was determined according to the code derived from CSI_f. For example, a code of 132 indicated that a city was categorized into two development phases: severe shrinkage during the first period and mild shrinkage during the second period, with long-term monitoring indicating moderate shrinkage.

3.1.4. Classification of Shrinkage Types Based on Shrinkage-Trajectory Encoding

After the composition of the shrinkage trajectory encoding K for each city was encoded and previous research findings were referenced [48,49], different types of urban shrinkage were identified and classified into three categories: continuous shrinkage, intermittent shrinkage, and continuous development. Cases where K consisted of uniform numerical values suggested a stable developmental trajectory for a city. The specific nature of the trajectory—whether indicative of shrinkage or development—was determined according to the numerical values of the segmented encoding; for instance, encodings such as 111 or 222 denoted continuous shrinkage, whereas 333 or 444 signified continuous development. Variations within the encodings necessitated further analysis: the classification relied on the last digit and the sequence of the encoding numbers. If the last digit was smaller than the preceding values with a minor difference, the classification was assigned as intermittent shrinkage, exemplified in encodings like 332, 343, 232, etc.

3.2. Quantification of ESs and Selection of Multidimensional Driving Factors

3.2.1. Quantification and Integration of Ecosystem Services

Currently, guided by the 2005 Millennium Ecosystem Assessment (MA) [50], ESs are categorized into four types: supporting, provisioning, regulating, and cultural. This paper quantifies the provision of five services according to InVEST model—habitat quality (HQ), food production (FP), soil conservation (SC), water retention (WR), and carbon storage (CS).

HQ serves as an essential metric for biodiversity and ecological conditions; elevated HQ levels enable the sustenance of a more diverse species population [51]. FP is a critical energy source for humans, and its variability represents the strength of provisioning capacity. SC and CS services are utilized to mitigate climate issues and reduce soil erosion, thereby epitomizing regulatory services [52]. Water resources indirectly improve humans’ quality of life by affecting the growth and development of plants and crops, while WY services determine the adequacy of water resources. Relevant calculation formulas are provided in Supplementary Material S2.

The complexity and diversity of ecosystem functionality mean that a single unit can provide multiple services. Quantitatively integrating multiple ES indicators aids in conducting a more comprehensive assessment [53]. The comprehensive ecosystem services index (CESI) was calculated using the summation method (Formula (4)), and range standardization (Formula (5)) was employed to minimize potential dimensional errors between services. We assumed that all five types of ESs were equally important; hence, all weights were set to 1. The index ranged from 0 to 5, with higher values indicating a stronger ES provision.

$${CESI}_{sum}={\sum }_{i=1}^n{w}_i\times {ES}_i$$

(4)

$${ES}_i={\displaystyle \frac{{ES}_{ij}-{ES}_{ijmin}}{{ES}_{ijmax}-{ES}_{ijmin}}}$$

(5)

where ${CESI}_{sum}$ represents the comprehensive ecosystem services index for the i^th grid cell, $w_i$ is the weight value for the i^th service, ${ES}_i$ represents the normalized value of the i^th service, and ${ES}_{ij}$ denotes the original value of the service. ${ES}_{ijmax}$ and ${ES}_{ijmin}$ correspond to the maximum and minimum values of all services, respectively.

3.2.2. Selection and Classification of Driving Factors

Urban shrinkage encompasses four dimensions: population, economy, space, and society. A decline in population is the most visible indicator, with economic recession serving as a primary driver behind this decrease. Urban spaces and public facilities become underutilized and idle due to population loss, subsequently impacting residents’ welfare. The magnitude of changes in these elements is utilized to assess the specific conditions of urban development. Additionally, anthropogenic factors increasingly play a pivotal role in the alterations of ESs. Therefore, the selection of driving factors primarily focuses on human-related elements and the extent of changes in these factors. Drawing on previous research [26,54,55], this paper constructs a framework (

Table 3. Classification and connotation of driving factors.

Characterization Dimension	Indicator Content	Indicator Number and Abbreviation	Factor Index
Population	Population Size	X1 (NP)	Number of Permanent Population (Ten Thousand Persons)
	Population Growth	X2 (NGR)	Natural Growth Rate (%)
	Population Distribution	X3 (DP)	Population Density (Per Square Kilometer)
Economic	Industrial Structure	X4 (PSI)	Proportion of Secondary Industry (%)
		X5 (PTI)	Proportion of Tertiary Industry (%)
	Economic Development	X6 (GDP)	Per Capita GDP (Ten Thousand Yuan)
		X7 (CFR)	Per Capita Fiscal Revenue (Ten Thousand Yuan)
		X8 (TGDP)	Total GDP (Ten-Thousand Yuan)
Spatial	Expansion Status	X9 (UBA)	Urban Built-up Area Proportion Percentage (%)
	Infrastructure Development	X10 (RA)	Per Capita Urban Road Area (km2)
	Environmental Quality	X11 (GG)	Built-up Area Green Coverage (%)
Social	Social Security Capacity	X12 (CFE)	Per Capita Fiscal Expenditure (Ten Thousand Yuan)
	Consumption Levels	X13 (RSCG)	Total Retail Sales of Consumer Goods (Ten Thousand Yuan)
	Social Stability	X14 (UPR)	Unemployment Rate (%)

) for analyzing influencing factors, comprised of 14 indicators.

3.3. Exploring the Importance and Thresholds of Driving Factors

3.3.1. Model Fitting Choices

Existing research indicates that the relationship between urban development and ESs is complex, encompassing positive, negative, “inverted U-shaped”, and nonlinear correlations. Therefore, this paper utilized traditional linear models (ordinary least squares, OLS) along with two machine learning models (random forest and XGBoost) to comprehensively explore the dynamics between influencing factors and changes in ESs. OLS, a staple in statistical linear regression, aims to minimize the sum of squared differences between observed and predicted values, but it operates under the assumption that there exists a linear relationship between the variables. RF and XGBoost, both derivatives of the Classification and Regression Tree (CART) framework, address complex modeling challenges. RF mitigates overfitting by averaging predictions from multiple decision trees across varied data subsets, enhancing accuracy [56]. XGBoost is an enhancement of the GBDT algorithm, which incrementally corrects prediction errors from previous rounds by adding decision trees, and it includes regularization terms to manage model complexity and prevent overfitting [57]. These models are preferred in analytical research for their robust handling of high-dimensional and nonlinear data, offering the advantages of speed and precision in prediction. This study’s modeling is conducted using Python 3.11, with the XGBoost and RandomForest models implemented from the Scikit-learn machine learning library.

3.3.2. Designing the Learning Process

First, random points equivalent to 10% of the total number of grid cells in each city were generated, amounting to 36,340 points. These points were used to extract the variations of CESI and socioeconomic factors across different study periods, serving as response and driving factors. The data were then divided into three datasets based on different city shrinkage types. In the OLS model, the VIF for all factors was less than 10, indicating no multicollinearity. In the RF and XGBoost, each category of data was split into training and test sets in a 7:3 ratio. The model generalization error was minimized by adjusting parameters and employing 10-fold cross-validation. Model performance was ultimately assessed using the coefficient of determination (R²), the root mean square error (RMSE), and the mean absolute error (MAE) to determine the specific parameters and select the final model. A comparative analysis of model performance is provided in

Table 4. Model performance comparison.

	Continuous Shrinkage			Intermittent Shrinkage			Continuous Development
Evaluation Index	R2	RMSE	MAE	R2	RMSE	MAE	R2	RMSE	MAE
OLS	0.6997	0.1507	0.1137	0.6545	0.1532	0.1142	0.6344	0.1433	0.1041
RF	0.7336	0.1370	0.1032	0.7411	0.1327	0.0984	0.7423	0.1214	0.0865
XGBoost	0.725	0.1419	0.1047	0.7253	0.1380	0.1023	0.7311	0.1238	0.0884

, with the RF ultimately being chosen to construct the research framework. Within the continuous development context, intermittent shrinkage context, and continuous shrinkage context, the final parameters for the RF model were set to mtry = 14, 14, 12 and ntree = 650, 300, 100.

3.3.3. Interpreting Model Learning Outcomes

Based on the Shapley values from game theory [58], the SHAP (Shapley additive explanations) model is a visualization method that transforms machine learning outputs into a “glass box”. This model allows for the interpretation of model outputs both globally and locally by calculating the marginal contributions of each feature within the model while considering interactions among features [59]. Sorting each feature’s marginal contributions according to specific rules reveals how various indicators influence the model’s outcomes. Positive SHAP values indicate a beneficial impact, with larger values having a greater effect. Compared to other explanation methods, tree-based SHAP models offer faster computation speeds and higher accuracy, making them particularly suitable for understanding the impact of each feature on service changes in depth. In this study, the visualization results included SHAP summary plots, which depict variable importance, and SHAP dependency plots that elucidate how elements depend on features. These visualizations were created using the SHAP package in Python with the aim of explaining the contributions of different influencing factors to service changes and how elements impact these changes.

4. Results

4.1. Encoding Urban Shrinkage Types

4.1.1. Spatiotemporal Patterns of Urban Shrinkage Trends

During the T₁ period (Figure 3a), the S_P values for all cities were negative, with extremes of −0.021 and −1.294, indicating significant population losses predominantly in northern HLJ, southeastern JL, and southeastern LN. Provincial capitals experienced relatively minor population declines, with S_P values of −0.02, −0.03, and −0.15, respectively. The S_E values (Figure 3d) were positive, showing a similar spatial distribution to S_P, with extremes of 0.05 and 1.57; lower values were associated with locations in northern HLJ and the northwest and southeast corners of JL, suggesting slower economic growth in these areas, while southern LN showed better overall economic development. In the T₂ period (Figure 3b), 10 cities had positive S_P values, ranging from 0.53 to −1.43, indicating a changing and increasingly divergent population development trend compared to T₁. Population decline was most pronounced in northern HLJ and JL, although slight population increases were noted in cities like Huludao and Dandong. A high percentage of cities in HLJ and JL showed declining S_E values, with 35.29% of cities experiencing a decrease.

Over the 20 years (Figure 3c), S_P remained negative with a low of −2.67. HLJ accounts for 83.33% of cities experiencing severe population decline. Additionally, half of the cities in JL and 21.42% in LN also reported substantial population losses. High S_E values were concentrated along the core axis of the “Harbin-Changchun urban cluster” and progressively moved southward, mirroring the economic focus shift in the northeast (Figure 3f). The consistent population decline and economic slowdown in the TPNC were evident.

4.1.2. Spatiotemporal Trajectories and Encoding of Urban Shrinkage Intensity

During T₁ (Figure 4), the proportions of cities experiencing severe shrinkage, moderate shrinkage, mild shrinkage, and slight development were 29.41%, 26.47%, 23.53%, and 17.65% respectively, with 70% of severely shrinking cities in HLJ and 83.33% of slightly developing cities in LN. In T₂ (Figure 4), the overall shrinkage trend slowed down, largely due to economic development. Shrinkage trends were relatively balanced across the TPNC, although classifications varied. HLJ had the highest proportion of severely shrinking cities at 83.33%.

Longitudinal observations (Figure 4) revealed persistent, significant shrinkage in HLJ, with only one city classified as mildly shrinking; the extent of shrinkage in southern LN was less severe compared to central JL. It is essential to emphasize that differences and similarities between T₁ and T₂, and longitudinal observations demonstrate the necessity of dividing the observation into different periods to enhance research accuracy.

According to the coding criteria, cities in TPNC exhibited 14 different shrinkage trajectories, categorized into 13 cities with continuous shrinkage, 13 with intermittent shrinkage, and 8 with continuous development. Geographically (Figure 3g), shrinking cities were predominantly located in HLJ, especially within the “siphon range” of provincial capitals (Harbin, Changchun, Shenyang, and Dalian); a few were border and port cities, such as those along the northern border of HLJ with Russia and the southern border of JL with North Korea. Except for Panjin, which is expanding its port economy, all resource-based cities underwent varying degrees of shrinkage, accounting for 62.5% presented in

Table 5. Classification and encoding of urban shrinkage phases.

Shrinkage Types	Encoding Composition
Continuous development (CDC)	444 (Changchun; Dalian; Harbin; Shenyang)
	434 (Panjin; Liaoyang)
	433 (Anshan)
	333 (Jilin)
Intermittent shrinkage (ISC)	343 (Daqing)
	332 (Chaoyang; Dandong; Jinzhou; Yingkou; Songyuan)
	232 (Fuxin; Huludao; Liaoyuan)
	132 (Baicheng)
	131 (Qiqihaer)
	121 (Heihe; Baishan)
Continuous shrinkage (CSC)	222 (Benxi; Siping)
	221 (Tonghua; Tieling)
	211 (Fushun; Jiamusi)
	111 (Hegang; Jixi; Mudanjiang; Qitaihel; Shuangyashan; Suihua; Yichun)

.

4.2. Temporal and Spatial Patterns of ESs

Figure 5 illustrates the temporal and spatial patterns of various services during the study period. Overall, except for habitat quality and carbon sequestration capacity, all other services exhibited improvement, with notable spatial heterogeneity.

From 2000 to 2020, the average HQ in TPNC slightly declined from 0.4939 to 0.4926, demonstrating a “deterioration-improvement” fluctuation pattern. The change rates for the T₁ and T₂ were −1.92% and 1.7%, respectively. In HLJ, both the deterioration during T₁ and the subsequent improvement in T₂ were minimal, while LN showed the opposite trend; JL exhibited the most severe degradation over the two decades. Spatially, regions with high HQ were concentrated in Yichun City in HLJ, areas predominantly covered by forests and grasslands. Conversely, lower values were observed in highly developed urban cores and their vicinities, such as the heavy industrial zones in Shenyang. Central JL and southern LN experienced notable declines over the study period.

FP in NTPC increased dramatically, from 69.36 tons/km² to 139.98 tons/km², representing a growth rate of 62.10%. During T₁, significant enhancements in FP were observed with standard value change rates of 111.41% (HLJ), 120.12% (JL), and 123.85% (LN), respectively. T₂, however, saw a substantial slowdown compared to T₁, with growth rates only reaching 18.76%, 20.07%, and 17.56% of the T₁ rates. Spatially, the distribution of FP exhibited high consistency with the spatial pattern of arable land. Over time, high-value areas gradually shifted northward, transitioning from a balanced distribution to concentration in HLJ’s fertile black soil regions and flat plains. Conversely, FP in the other two provinces gradually declined, with significant decreases in southern LN and central JL.

The average SC followed a dynamic “increase-then-decrease” trend, rising overall by 37.41%. The three-year averages were 7109.99, 10,976.31, and 11,358.99 tons/km², respectively. HLJ consistently grew at rates of 0.45% and 38.19% throughout the periods, unlike the other two provinces, which mirrored general trends. Spatial analysis revealed higher SC levels in the east and north due to land use and topography. Increases during T₁ were notable in southeastern coastal LN and northern HLJ. In T₂, enhancements were pronounced across the region, particularly in the east and west of LN. The western area showed gradual improvements, with the north and east also experiencing enhancements.

Over 20 years, the WR capacity exhibited a fluctuating “increase-decrease” pattern, ultimately rising by 6.23 × 10⁷ tons, an increase rate of 57.21%. Annual averages were 59.59, 158.60, and 139.25 tons/km², respectively. The significant fluctuations in WR are primarily influenced by the interannual variations in precipitation and evapotranspiration: 2010 was a typical wet year for the study area, but the gains from forest expansion have been outweighed by increased evapotranspiration due to warming climates [60], leading to a slight decline in WR during the T2 period. Additionally, ecological restoration projects that have been progressively implemented since the 1990s, including two phases of the Natural Forest Conservation Program (NFCP) and multiple phases of the “Three-North” Shelter Forest Program, along with the construction of nature reserves, have also made significant contributions to the improvement in WR [61]. Only HLJ consistently increased its WR by 26.56%, while other provinces showed initial declines followed by increases, with LN’s capacity surging by 124.67%. The spatial distribution of WR predominantly showed an “eastern high, western low” pattern, closely mirroring vegetation distribution. During T1, high-value areas shifted from the east and north to the south, becoming more dispersed. In T2, they concentrated more in southeastern LN and northern HLJ. Rapid urban expansion increased impervious surfaces, suppressing evapotranspiration and consequently raising WR in southern LN.

CS displayed a dynamic pattern of decreases followed by increases, overall showing a slight net decline of 1.45% from 2000 to 2020. The average CS was 8316.26 tons/km² in 2000, dropping to 7194.82 tons/km² in 2010 and partially recovering to 8197.27 tons/km² by 2020. JL saw a minor increase, contrasting with HLJ’s overall 2.53% decline, which was still less severe than LN’s 3.56% reduction. Spatial analysis revealed that high-value CS zones were primarily in areas with extensive vegetation cover, such as the Mudanjiang in eastern JL, and Dandong in southeastern LN. During T₂, these high-value areas in T₁ shifted eastward, while regions in the south experiencing significant urban expansion, consistently showed decreased CS over the two decades.

Temporal changes of ${\mathrm{C}\mathrm{E}\mathrm{S}\mathrm{I}}_{\mathrm{s}\mathrm{u}\mathrm{m}}$ indicate a 41.34% increasein in overall service levels over 20 years. During T₁, the $\mathrm{C}\mathrm{E}\mathrm{S}\mathrm{I}$ rose at a rate of 45.33% but significantly slowed in T₂, dropping to 2.48%. In LN, $\mathrm{C}\mathrm{E}\mathrm{S}\mathrm{I}$ deteriorated at a rate of −1.57%, while HLJ improved its services by 3.39 times the average of TPNC. The spatial distribution exhibited a pattern of “high in the east and north, low in the west and south”, intensifying from west to east over time. In T₂, regions of degradation and improvement were located in the southeast and northeast corners, respectively.

4.3. Spatiotemporal Dynamics of Ecosystem Services in Shrinking Cities

Based on the classification of cities, the cross-sectional values and dynamic changes of Ess’ mean values for different types of cities were compared to depict the force of shrinkage (Figure 6). Generally, continuously shrinking cities exhibited relatively optimal ESs, exceeding the average of the study area, with the greatest improvements and mildest deteriorations across all periods. Their average HQ was 0.56, surpassing the other two city categories by 19.46% and 23.08%, and their SC values exceeded those of continuously developing cities by 30.58%. Due to the rapid expansion of impervious surfaces in urban built-up areas, cities that are continuously developing demonstrated optimal WR, with average values across three different years surpassing those of other categories by 0.01 and 0.04, respectively. Unlike SC, average sub-service levels in continuously developing cities were marginally higher than in intermittent shrinking cities. This phenomenon may be related to the policies implemented in continuously developing cities to restore ecosystems degraded due to rapid development.

Service-capacity enhancements varied across city types, typically increasing with the degree of shrinkage. Continuously shrinking cities showed the most substantial improvement, with the $\mathrm{C}\mathrm{E}\mathrm{S}\mathrm{I}$ average increasing by 0.62 over 20 years; the $\mathrm{C}\mathrm{E}\mathrm{S}\mathrm{I}$ for intermittent shrinking and continuously developing cities rose from 1.14 and 1.23 in 2000 to 1.72 and 1.79 by 2020, increasing at rates of 50.33% and 45.23%, respectively. Improvements during T₁ were similar, at 0.51, 0.55, and 0.58. However, from 2010, improvements in continuously shrinking cities were 3.49 and 5.09 times greater than those in intermittent shrinking and continuously developing cities, respectively, with only continuously developing cities experiencing deterioration in T₂.

Further comparison of service changes across TPNC showed that, except for CS, HLJ—the most severely shrinking—consistently performed well in all periods. The change rates in $\mathrm{C}\mathrm{E}\mathrm{S}\mathrm{I}$ in T₂, in descending order of shrinkage severity, were 5.73%, 1.77%, and −1.11%. Natural gradients have been proven to affect ESs, and comparisons of temporal and typological cross-sections confirm that different stages of urban shrinkage can modify these disparities.

4.4. Relationship between Driving Factors and Changes in ESs

4.4.1. Factor Importance Analysis

Figure 7, Figure 8 and Figure 9 present global bar and SHAP summary plots illustrating each factor’s influence on changes in ESs. The x-axis denotes SHAP values, and the y-axis lists various factors; each dot symbolizes a sample, with the color intensity (red for high and blue for low) reflecting the feature values. A greater dispersion of data points indicates a more substantial impact on service improvement.

In cities with continuous development (Figure 6), changes in ESs were significantly affected by the proportions of tertiary (PTI, contribution = 0.18) and secondary industries (PSI, 0.04), the unemployment rate (UPR, 0.03), green coverage (GG, 0.01), and the natural growth rate (NGR, 0.01). The decreasing contributions from economic, social, spatial, and demographic dimensions to service changes were 74.93%, 13.46%, 7.88%, and 3.73%, respectively. High variability points (red) for PTI, NGR, and GG were predominantly on the left, suggesting negative impacts from larger changes in these factors. Conversely, significant changes in other factors might be beneficial, as their red points are predominantly clustered on the right side of the figure.

For cities experiencing intermittent shrinkage (Figure 8), the importance of various indicators shifted significantly. The five most critical indicators were a permanent population number (NP, 0.18), green coverage (GG, 0.05), population density (DP, 0.03), the unemployment rate (UPR, 0.02), and per capita fiscal expenditure (CFE, 0.02). Changes in population and economic factors had the most substantial impact on service improvement, accounting for 68.11% and 14.01%, respectively, as much as twice the impact of social dimensions. High variability in NP, PD, per capita fiscal revenue (CFR), and per capita road area (RA) were beneficial for service improvement; only minor changes in GG and fluctuations in UPR could preserve ES capabilities without harm.

The impact of various factors on cities with continuous shrinkage was complex (Figure 9). The top five indicators by importance were population density (PD, 0.06), the permanent population number (NP, 0.06), the proportion of secondary industry (PSI, 0.05), the per capita GDP (GDP, 0.03), and total retail sales of consumer goods (RSCG, 0.03).The differences in importance between them were minor. Population and economic dimensions contributed 39.32% and 30.36%, respectively, to service improvement, while the influence of social and spatial dimensions was roughly equivalent. However, in such cities, data points for population-related factors often clustered to the right of the median, differing slightly from other types of cities. Increases in the proportion of UBA negatively impacted service improvements, while significant declines in the share of PSI and CFE typically inhibited enhancements in service capacity.

The impact of various factors on urban services shifted with the degree of urban shrinkage (Figure 10). GDP and PSI consistently exerted positive influences across different stages of cities. With greater shrinkage, the significance of factors such as NP, PD, GDP, and GG gradually increased, with SHAP-value proportions climbing from initial values of 0.19%, 0.11%, 1.84%, and 4.08% to 21.37%, 22.09%, 12.18%, and 11.37%, respectively. In contrast, contributions from the PTI, RA, and UPR diminished as shrinkage increased, declining at rates of 0.18 (99.33%), 0.01 (45.76%), and 0.03 (92.39%). The influence of NGR and GDP initially rose and then fell. Spatial-dimension factors exerted a relatively minor average impact of 0.03 compared to other dimensions.

4.4.2. Non-Linear Relationship of Driving Factors on ESs

Figure 11, Figure 12 and Figure 13 show the feature-dependence plot for the impact of various factors on SHAP values, revealing significant non-linear relationships between variables and service improvements. The scatter points in the diagram represent the changes in a specific city over a certain period; thus, the total number of points is twice the number of cities in that category. In continuously developing cities (Figure 11), changes in population-related factors uniformly negatively correlate with service improvements, with the NGR showing the strongest negative correlation—greater declines led to higher SHAP values. Adjustments in the PSI are beneficial for service enhancement when within the −5%-to-5% range. Interestingly, increases in GG have a negative impact, though the extent of this impact gradually decreases.

For cities undergoing intermittent shrinkage (Figure 12), several intriguing findings emerged: slight population growth and densification had positive effects; maintaining the current PTI was the minimal baseline for preserving current service levels; and a gradually increasing PSI was beneficial. Additionally, increases in RA beyond 5 square meters and maintaining a UPR between 0 and 0.8% also produced positive effects.

For cities experiencing continuous shrinkage (Figure 13), improving the level of ESs could be achieved by increasing the PTI by 10% to 15%. However, increases in UBA and RA could lead to indiscriminate urban expansion, significantly reducing the service capacity of these cities. Higher UPR and increased GG had positive effects. While the threshold relationships varied across the three city types, all demonstrate that increases in NGR had negative impacts. Increases in GDP and CFE were beneficial for all cities, and the beneficial range of GG increased with the intensity of shrinkage.

5. Discussion

5.1. Contribution of Urban Shrinkage to ES Enhancement

Human activities serve as the central driving force behind urban growth and decline, continuously impacting ecosystems. However, the relationship between urban shrinkage and ESs remains insufficiently understood. By quantifying the evolution of ESs from 2000 to 2020, it was found that, despite a decline in HQ and CS, both the individual subsystems and the overall service provision levels improved. Moreover, the extent of shrinkage was directly proportional to the degree of improvement and inversely proportional to the degree of degradation. These findings suggest that regional urban shrinkage can lead to enhancements in certain ecosystem services, with the effects varying across different stages of urban development.

Specifically, the overall HQ had consistently declined across the TPNC, with cities experiencing continuous shrinkage showing the least deterioration. SC also provided supporting evidence: the mean standard values for cities categorized by decreasing degrees of shrinkage were 0.01 0.01, and 0.01, respectively. Cities experiencing continuous shrinkage showed improvement levels 3.49 times and 5.09 times greater than those undergoing intermittent shrinkage and continuous development, respectively. Only cities without shrinkage saw a decline in CESI. These findings align with previous research, which identified intensive human activities as a critical factor in the degradation of ESs [62].

Research indicates that in well-developed cities, frequent human activities have progressively encroached upon ecological source areas such as forests, grasslands, and water bodies, converting them into built-up land. This process increases the fragmentation of urban ecological patches, disrupts the corridors facilitating the flow of ecological elements, and consequently diminishes HQ. This phenomenon aligns with observations made in rapidly urbanizing areas such as the Guangdong–Hong Kong–Macau Greater Bay Area and the Yangtze River Delta [63]. Furthermore, the imbalance between the supply and demand for food production services in rapidly developing areas forces continuously developing cities to intensify the exploitation of arable land in order to meet food supply demands [64]. Intensive agriculture and increased use of chemical fertilizers potentially alter the internal structure of the soil, thereby reducing its regulatory capacity and leading to a decline in SC ability. WY, closely linked to precipitation and evapotranspiration, is also affected by the rapid expansion of impervious surfaces in urbanized areas. This expansion inhibits evapotranspiration, resulting in an improvement in WY in the southern part of Liaoning, where urban shrinkage is minimal.

Conversely, cities experiencing sustained shrinkage encounter issues such as diminished economic vitality and slowed infrastructure development due to population loss, leading to decreased land-use intensity and human disturbances. This finding aligns with conclusions suggesting that reducing human interventions such as migration could slow ecosystem degradation by adjusting the composition and quality of ecological assets in different bioregions [65,66]. Additionally, “high housing vacancy rates” serve as another typical characteristic of shrinking cities. The demolition of abandoned buildings and the lack of adequate land management measures result in extensive urban vacant plots [67], providing opportunities for urban rewilding and the restructuring of urban spaces: shrinking cities can harness the vast potential of naturalizing vacant land to improve urban ecosystem services. For instance, planting forests on unused lands enhances overall WR and CS, and converting “brownfields” into green infrastructure can boost biodiversity and enhance resident welfare. Research by Zhang et al. found that land abandonment due to migration created new growth spaces for vegetation, significantly enhancing ESs [68]. It is observable that, in cities undergoing shrinkage, the deceleration of urban expansion mitigates external anthropogenic disturbances; concurrently, vacant lands facilitate opportunities for the regeneration of vegetation, thus enhancing ESs. This insight substantiates the notion that regional urban shrinkage can, to a certain extent, ameliorate specific ecosystem supply services, presenting a distinct contrast to the impacts associated with urban expansion.

It is worth noting that, apart from HQ, continuously developing cities slightly outperformed those experiencing intermittent shrinkage in overall ES provisions. One plausible explanation is that sustained economic development has heightened the emphasis on the significance of livable urban environments, leading to an increased pursuit of high-quality life that, in turn, fosters the augmentation of urban green spaces [62]. Consequently, despite the transformation of the majority of land into impervious surfaces during urban expansion, the introduction of numerous green infrastructure, such as city parks and street trees, has been facilitated [69,70]. This trend of improved greening levels in economically better-off cities in the northeast has been confirmed in multiple studies [71]. Furthermore, effective government management and a solid economic foundation enable such cities to more successfully implement and enforce policies aimed at urban greening and improving living environments, thereby making significant contributions to the enhancement of ecosystem service capacities [72].

In contrast, cities experiencing intermittent shrinkage face severe challenges: Firstly, resource depletion and population loss have undermined their intrinsic economic growth drivers, increasing fiscal pressures. However, as these cities are not severely shrinking, governments have not fully shifted their development strategies. They often struggle with unresolved contradictions between population loss and spatial expansion [73], and they frequently resort to land finance and intensified resource extraction as primary means of alleviating fiscal stress. Secondly, the ecosystem degradation resulting from previous development receives less focus because the synergistic effects of urban greening and economic growth are typically constrained to more developed urban areas. The increase in anthropogenic disturbances and the lack of mitigation measures can reasonably explain the findings of the lowest comprehensive supply capacity in such cities.

5.2. Identifying Key Factors Driving ES Changes in Different Types of Shrinking Cities

In our analysis of driving forces, we identified marked disparities in the impact of various factors on ES changes across different urban types. Cities undergoing continuous development (CDC) showed the highest sensitivity to economic factors, significantly exceeding the other two city categories by 61.87% and 81.01%, respectively. Notably, significant shifts in PTI generally hindered ES enhancements, while minor variations in the PSI positively influenced ES improvements. An increase in the share of the tertiary sector signals favorable urban development trends, attracting more population migration and promoting the construction of comprehensive support services, thereby facilitating the expansion of urban infrastructure. Indeed, such cities are on the brink of exceeding their ecosystems’ carrying capacity due to dense populations and economic growth, and unrestrained expansion could lead to irreversible consequences, indirectly exerting a negative impact on service improvements. This also explains the adverse effects of a higher NGR. Moreover, an increase in the PSI intensifies urban dependence on resources and diminishes the advanced nature of the industrial structure. Insufficient endogenous development momentum and a lack of subsequent industries reduce urban efficiency and employment rates, gradually leading to population outmigration. Against the backdrop of China’s new stage of economic and social development, the significant decline in external demand and the increasingly pronounced issues of overcapacity in coal and steel exacerbate the impact on cities. An increased share in the secondary sector can significantly challenge urban areas, precipitating urban shrinkage. Surprisingly, we found that UPR and RA had significant negative and positive impacts, respectively, in these cities. Rising unemployment rates suggest a relative decline in a city’s “development potential” [1], which in turn squeezes out high-quality talent and technological innovation. The accelerated outflow of labor further weakens these cities’ ability to attract development factors. Mounting development pressures compel urban policy shifts from improving quality of life to economic development, forcing cities to mitigate economic decline through land finance and lowering environmental standards for incoming businesses. Consequently, ecosystems face increased disturbances and a reduction in their recovery capabilities, as the effects of ecological restoration measures are relatively insufficient compared to the extent of damage. Therefore, the decreasing negative impacts of unemployment rates with increasing urban shrinkage, and the seemingly limited positive impact of increased GG on such cities, appear to be adequately explained. Additionally, the increase in the proportion of roads underscores the advanced and improved infrastructure, reducing the cost of intercity-element flow and providing opportunities for resource redistribution, thus greatly “catalyzing” population outflows and reducing human disturbances. Numerous studies indicate that the development of transportation infrastructure such as roads and high-speed rail can boost regional economic growth; it also increases the likelihood of urban shrinkage, with road systems having a more substantial contribution [74]. Lu and colleagues’ study in HLJ province also shows that the mileage of highways impacts the ecological benefits of shrinking cities, thus explaining the positive effects of increased road ratios [75]. Overall, the findings essentially trigger urban shrinkage from multiple dimensions, further elucidating the role of shrinkage in service improvement.

In cities experiencing intermittent shrinkage (ISC), the influence of population, economic, spatial, and social dimensional factors sequentially decreased, yet the impact of each sub-factor on such cities was relatively complex. These cities are at a crossroads of developmental-phase transitions, where varying changes in factors can lead to different outcomes. On one hand, an increase in RA accelerates the free mobility of populations, whereas a reduction in the GG detrimentally affects the quality of life for urban residents. These elements have a positive effect on ESs by “passively” steering a city towards shrinkage. Conversely, escalations in NP, GDP, and CFR can decelerate, or even invert, the shrinkage trajectory, thereby realigning these cities towards a path of continuously developing cities. Economic expansion acts as a magnet, drawing population concentrations that subsequently provide the essential labor force and consumer markets necessary for further economic dispersal. Consequently, these cities can “proactively” enhance their ESs through strategic policy initiatives aimed at elevating foreign trade, fostering technological innovation, and accelerating the purification of industrial structures. This aligns with research conducted by Han [76].

For cities undergoing continuous shrinkage, the impact of various factors on ES improvement is relatively balanced, but the positive contributions of indicators representing social welfare—total retail sales of consumer goods (RSCG) and per capita fiscal expenditure (CFE)—are most significant in these cities. A plausible explanation is that an increase in CFE (e.g., investment in education, health, and public infrastructure) can enhance residents’ quality of life and bolster their purchasing power, thereby driving increases in RSCG and government tax revenues, alleviating urban fiscal pressures. Furthermore, ecological governance primarily relies on government fiscal expenditures. Once fiscal levels surpass the pressure threshold, investments in environmental protection within the fiscal expenditure structure are secured, thereby improving ESs. Residents whose sense of well-being is enhanced will perpetuate the aforementioned process, ultimately establishing a positive feedback loop. Evidence from Dresden and certain “Rust Belt” cities in the USA indicates that, when governmental development shifts from a growth pursuit to improving the welfare of existing residents, the living environment is significantly enhanced, and surprisingly, the population begins to grow again [77,78]. Conversely, the blind expansion of built-up areas indicates that governments have not yet accepted the reality of shrinkage. Such expansion not only contradicts the current demographic trends of population loss in these cities but also hinders the maintenance of fiscal balance and the advancement of ecological restoration activities, thus negatively impacting ES enhancement. This also explains why most of the red data points in X9 are concentrated on the left side of Figure 8b.

A further exploration of the relationship between the importance of various factors and the degree of urban shrinkage is warranted. Contributions from CFE and GG in built-up areas increased as shrinkage intensified. Stable fiscal foundation ensures the continuous improvement of residents’ welfare. On one hand, this can enhance government enforcement of environmental policies and financial commitments to ecological restoration projects, directly benefiting ESs by providing public environmental services [79]. Additionally, fiscal expenditure serves a policy-guiding role, which can steer the direction and scale of non-governmental industry investments and reduce the reliance of existing industries on natural resources, thereby indirectly benefiting ESs. The increasing importance of green coverage as cities experience more intense shrinkage is related to the availability of vacant urban spaces suitable for conversion into large parks or green infrastructures. Cities undergoing continuous shrinkage have more space than other kinds of cities. Such cities, with their early occurrences of abandoned industrial lands, have benefited from long-term management strategies that reduce ecological damage, thereby actively improving ESs both qualitatively and quantitatively. Experiences from cities like Cleveland and Detroit in the United States and Leipzig in Germany underscore the substantial contributions of vacant lands in enhancing ESs in shrinking cities. Consequently, an increased vegetation cover and the improved connectivity of landscape patterns alleviate ecosystem pressures to some extent, and the proactive green transformation of industries reduces pollution emissions. These cities experience an enhancement in ESs under a dual mechanism, aligning with He’s research findings [80]. The importance of the GDP diminishes as city shrinkage intensifies, with respective values of 18.48%, 13.34%, and 12.18%. Although these figures slightly deviate from findings in regions with diverse economic backgrounds, such as the Pearl River Delta and the Loess Plateau [38,81], our study addresses a research gap in areas facing economic downturns.

5.3. Threshold Analysis for Sustainable Urban Development

Threshold analysis highlights the need to limit human disturbances within the stable range of ecosystems, setting benchmarks for decision-making and management actions from an ecological perspective. Industrial restructuring and transformation play a pivotal role in the process of urban shrinkage. For the TPNC, finding an industrial structure that balances the ecological impact and urban development is crucial. In cities with ongoing development, adjusting the secondary industry share by −10% to 15% and maintaining tertiary-sector growth below 10% may optimize the industrial structure for maximum benefits. Research by Li and others [82] has shown that industrial diversity can foster positive, stable urban development. Therefore, in such cities, adjusting PSI can be achieved by strategically diversifying local resources to enhance industrial variety.

In cities experiencing shrinkage, an increasing of PSI appears to have a minimal impact on those undergoing temporary shrinkage; however, enhancing the tertiary sector has the opposite effect. Thus, in such cities, adjustments to and the optimization of the industrial structure should prioritize stabilizing the existing tertiary sector. Conversely, in cities with continuous shrinkage, adjustments not exceeding −10% in the secondary sector and development between 0–20% in the tertiary sector may benefit service improvements. But no upper threshold for secondary-industry growth has been observed. Theoretically, it is feasible to continuously develop heavy industry to foster urban prosperity, which in turn gradually boosts the tertiary sector and thereby enhances ESs. However, for cities heavily reliant on resources, this approach inevitably leads to further urban decline or even abandonment. These results, derived from machine learning algorithms, do not consider the practical reality that cities should be preserved, rather than allowed to disappear. This underscores the importance of interpreting mathematical analyses in light of real-world conditions. The distinct threshold relationships observed among the three types of cities demonstrate that changes in ecosystem service capacity are closely linked to the stage of shrinkage, rather than simply the occurrence of shrinkage. It is crucial to define and categorize urban development stages accurately; a one-size-fits-all analysis may yield biased results. Scholars like Gao have also raised similar points, emphasizing the need for nuanced analysis that accounts for varying urban conditions [83].

5.4. Revitalizing Shrinking Cities through Ecological Management

Discussions of urban shrinkage often focus on its negative impacts—economic downturns leading to diminished vitality, high unemployment and housing vacancy rates, rising crime rates, and escalating social conflicts [84]. However, urban shrinkage is not inherently negative. In our research, we confirm the positive role of urban shrinkage in enhancing ESs. Research by Hu and colleagues on cities like Leipzig in Germany, Detroit in the USA, and Yichun in China’s Northeast has revealed that, despite the physical decline in infrastructure and natural environments in shrinking cities, the levels of residents’ life satisfaction and well-being are not lower than in other cities and may even surpass those in growing cities, indicating a potentially higher quality of life [76,85,86]. Shrinking cities can alleviate certain negative externalities such as traffic congestion and shortages in educational and healthcare resources to some extent. Residents in these cities are not subjected to high-pressure work environments or congested public transportation, contributing to an improved urban environment and a stronger sense of attachment to the city. Although shrinking cities may not reverse their gradually declining objective characteristics currently, managers should capitalize on the potential positive effects generated via the ecological environment during urban shrinkage. Starting from a human-centered perspective, they should explore diverse angles for urban transformation and devise effective ES management strategies to rejuvenate and revitalize shrinking urban areas.

In continuously developing cities, the primary focus for enhancing ESs lies in “mitigation”. This entails alleviating ecosystem stress by reducing human-induced disturbances and slowing the pace of urban expansion. Measures such as increasing building density, optimizing land-use efficiency, and curbing the spread of built-up areas and farmland reclamation can prevent damage to ecological source areas and corridors. Additionally, these cities can leverage the “Matthew Effect” of urban development to alleviate mounting environmental pressures by promoting regional cooperation to ensure a more equitable distribution of resources and minimizing the concentration of residents in core urban areas [87].

For cities experiencing intermittent shrinkage, “guidance” emerges as the pivotal approach for improving ESs. This can be achieved through policy interventions that steer industrial restructuring, actively promote external openness, and provide incentives such as tax reductions to attract enterprises and talent. By proactively enhancing technological capabilities, these cities can improve resource efficiency and reduce pollution, thus achieving both economic growth and ecosystem optimization. Spatial planning also plays a crucial role in guiding compact and efficient urban land use, ensuring a synergistic alignment between spatial planning and landscape configuration.

For cities undergoing prolonged shrinkage, the core strategy centers around “transformation”. The extended period of decline may prevent these cities from quickly returning to normal levels of urban development. However, city planners and managers can shift their development paradigm by acknowledging the reality of urban shrinkage and proactively embracing the environmental opportunities it presents. The focus should be on enhancing residents’ quality of life and improving living environments as central goals of urban revitalization. This can be accomplished by repurposing obsolete spaces, expanding green infrastructure, and reorganizing urban layouts under a “slimming and strengthening” governance approach, thereby fostering equitable resource distribution and sustainable urban regeneration [88].

5.5. Limitations and Prospects

Despite the findings that regional urban shrinkage can enhance ecosystem service capacities, this study had several limitations. First, the identification of shrinking cities was based on a composite index of population decline and economic recession, which could have introduced biases into the results, and the ten-year research interval may have been too lengthy to capture more immediate changes effectively. Furthermore, existing research underscores the impact of scale effects and spatial heterogeneity on studies; this paper has focused only on a broader urban scale and employed a random forest model that did not adequately address spatial heterogeneity. This approach might have overlooked some detailed impacts, which would be disadvantageous for proposing adaptive recommendations for the ecosystem management of shrinking cities from a precision governance perspective. Therefore, future work should consider other aspects of urban shrinkage and could employ a more diversified and precise indicator system to enhance the robustness of our methodology in identifying urban shrinkage. Moreover, optimizing the study from a multi-scale spatial perspective will be our subsequent exploration direction. We aim to integrate theoretical research with field surveys to propose more scientifically sound and reasonable policies for the ecosystem management of shrinking cities.

6. Conclusions

This research confirms that urban shrinkage can improve ESs. By analyzing long-term remote sensing data, this study developed a comprehensive shrinkage index and categorized cities in the TPNC into continuously, intermittently, and continuously developing types. It then quantified the spatiotemporal patterns of ESs from 2000 to 2020 and employed RF and SHAP models to explore and visualize the relationships between service changes and driving factors. The findings indicate the following: (1) Over 20 years, the degree of shrinkage in the TPNC has continuously intensified, with resource-based cities experiencing the most severe shrinkage. (2) Improvements were observed in all services except HQ and CS. Among the three types of cities, those experiencing continuous shrinkage demonstrated the most notable overall enhancement. (3) The impact of the same factor varied across cities with different types of shrinkage. Overall, the economic and social dimensions had the most pronounced average impacts. The importance of GG increased with the degree of shrinkage. Cities experiencing shrinkage were most sensitive to changes in PSI. The ecological reuse of vacant spaces and the reduction of human disturbances contribute to the enhancement of ESs. (4) Nonlinear relationships indicate that increases in the natural growth rate negatively impact service improvements, whereas increases in CFE and GDP contribute positively. Enhancing living environments and resident welfare can boost ES capacities. Cities undergoing shrinkage should capitalize on their potential ecological benefits and formulate appropriate policy measures to facilitate future urban transformation and the enhancement of human well-being.