Assessment and Simulation of Urban Ecosystem Resilience by Coupling the RAR and Markov–FLUS Models: A Case Study of the Jinan Metropolitan Area

Abstract

Confronting escalating urban ecological risks, this study investigates ecosystem resilience evolution in the Jinan metropolitan area’s mountain–plain structure. We establish a Resistance–Adaptability–Resilience (RAR) framework integrating ecosystem service value and landscape patterns. Using Optimal Multi-layered Geo-Detector and Markov–FLUS modeling, we quantify natural–socio-economic interactions and simulate resilience under three scenarios: inertial development, cultivated land protection, and ecological priority. The results show fluctuating resilience (0.1863→0.1876→0.1863) with functional intensification in high-value areas and escalating vulnerability in low-value regions, alongside the spatial dichotomy between the resilient southern mountains and northern plains, dominated by natural factors. Cultivated land protection degrades mountain resilience via slope farming, while ecological priority stabilizes it through transitional controls. The proposed “resilience red line–development permit” mechanism demonstrates terrain and policy integration optimizing resilience allocation. This framework offers strategies to reconcile ecological conservation and farmland security in urbanizing regions.

1. Introduction

Urban ecological risk issues have been increasingly prevalent in recent years, leading to significant detrimental effects on both human well-being and ecological quality [1]. Consequently, global entities have implemented various policy initiatives to enhance urban resilience by developing pertinent strategies and frameworks. Scholars are increasingly focusing on investigating ways to enhance urban resilience through the lens of ecosystem resilience [2,3].

Urban resilience encompasses a city’s capacity to sustain its functionality and structural integrity amidst external and internal changes while also fostering sustainable development [4]. This resilience is demonstrated through the city’s ability to withstand environmental stressors and maintain ecosystem stability [5]. Research on urban ecosystem resilience in China has primarily focused on the Yellow River Basin [6], the Yangtze River Delta [7], the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) [8], and the Beijing–Tianjin–Hebei region [9]. Recent comparative studies highlight distinct resilience patterns across the major river basins: The Yangtze River Basin exhibits higher adaptability and recovery capacities driven by economic factors. The Yellow River Basin emphasizes resistance mechanisms due to its fragile ecological foundation and historical sediment dynamics. The Greater Bay Area prioritizes green technology innovation and circular economy integration, advancing sustainability through low-carbon applications (e.g., near-zero-energy buildings and photovoltaic systems) and eco-industrial synergies. The Beijing–Tianjin–Hebei region focuses on cross-regional collaborative governance and ecological barrier construction, strengthening ecological security through air pollution joint control and large-scale sand fixation projects (e.g., the Saihanba Mechanized Forest Farm). The development of resilience patterns within and among urban agglomerations is still in its nascent phase [10]. Prior research has established theoretical assessment frameworks for ecosystem resilience, considering factors such as the economic–social–community management system [11], population clustering [12], and vernacular ecosystems [13]. However, limited studies have applied the Resistance–Adaptability–Resilience framework to examine the fundamental components of urban ecosystem resilience. Land cover change is a significant driver of human-induced alterations in urban areas, leading to the reduction of natural spaces, depletion of resources [14], and modification of landscape patterns. Therefore, it is imperative to investigate the resilience of urban ecosystems in response to land use changes. Recently, there has been a growing body of research examining the cascading effects of food price fluctuations on ecosystem service values (ESVs) [15], offering a novel lens through which to comprehend the interannual fluctuations in ecosystem resilience. Various factors influence urban ecosystem resilience, including NDVI, GDP, and soil type, among others [16]. Previous studies have commonly utilized the extended STIRPAT model, the spatiotemporal geo-weighted regression model (MGWR), and the geographic detector model (GDM) to examine the spatial and temporal heterogeneity of influencing factors [17]. Nevertheless, challenges remain in determining the most suitable spatial scale and in concurrently identifying multiple factors, due to limited simulations of ecosystem resilience. The Markov–FLUS model has been introduced to enhance the precision of simulating land use changes [18], thereby offering valuable insights for resilience forecasting.

How do land use transitions driven by urban expansion affect the spatial–temporal evolution of ecosystem resilience in the mountainous zones of the Jinan metropolitan area? What are the dominant socio-ecological factors determining the trade-offs between cultivated land protection and ecosystem resilience enhancement in ecologically fragile mountainous areas? Can scenario-based simulations using the Markov–FLUS model identify optimal pathways to reconcile agricultural land preservation with ecosystem resilience improvement? This study pioneers an integrated framework that synergizes three methodological advancements to advance ecosystem resilience research. (1) Multi-dimensional Resilience Quantification: The Resistance–Adaptability–Resilience (RAR) model integrates land use dynamics, ecosystem service value (ESV), and landscape pattern analysis, overcoming the simplification bias inherent in single-index evaluations prevalent in prior studies. (2) Multi-scale Driver Detection: By coupling the Optimal Multi-layered Geo-Detector (OMGD) model with the RAR framework, we enable simultaneous identification of dominant factors across 19 spatial scales (100 m–19 km) and their nonlinear interactions, addressing critical gaps in conventional single-scale analyses. (3) Policy-Responsive Scenario Simulation: The Markov–FLUS integration introduces adaptive land use simulations calibrated through ecological elasticity coefficients. This tri-model coupling establishes a novel “Assessment-Detection-Simulation” paradigm that resolves spatial mismatches between resilience evaluation (macro-scale) and driver analysis (micro-scale) through the OMGD model’s scale-adaptive detection, overcomes path dependency in land use modeling by embedding policy constraints (e.g., resilience red lines) into FLUS conversion rules, and provides actionable insights for reconciling ecological conservation with agricultural security in mountain–plain transition zones.

2. Materials and Methods

2.1. Study Area

The Jinan metropolitan area, is shown in Figure 1, situated in the central and western regions of Shandong Province and the lower reaches of the Yellow River in China, constitutes a significant component of the Shandong Peninsula urban agglomeration. Encompassing Jinan as its focal point, the area comprises 6 cities and 25 counties within approximately 22,300 square kilometers. With a substantial resident population and a vibrant economy, the Jinan metropolitan area experiences a typical temperate monsoon climate. Geographically, the area features a juxtaposition of low mountains, hills, and plains, with higher elevations in the south and east, gradually descending towards the north and west. Bounded by the plain of Shandong Province to the north and west, the area is flanked by mountains and rivers to the east and south, in proximity to the residual ridges of Mount Tai and the hilly terrains in the central and southern parts of Shandong Province. Characterized by undulating mountains, the area serves as a natural green barrier, offering abundant natural resources to the city.

2.2. Data Sources

Table 1. Data sources.

Data Type	Data Name	Year	Data Sources
Land use data	Land use	2003, 2013, 2023	Zenodo
Natural environment data	Annual precipitation	2003, 2013, 2023	National Qinghai–Tibet Plateau Science Data Center
	Annual average temperature	2003, 2013, 2023	National Qinghai–Tibet Plateau Science Data Center
	Agrotype	1995	Resources and Environmental Science and Data Center
	Digital elevation	2003	Resources and Environmental Science and Data Center
	Slope	2003	Based on DEM conversion
	Water system data	2023	OSM
Socio-economic data	Urbanization rate	2003, 2013, 2023	Shandong Statistical Yearbook
	Density of population	2003, 2013, 2023	https://landscan.ornl.gov/ (accessed on 12 November 2024)
	Gross domestic product	2003, 2013, 2023	Bulletin of Statistics on National Economic and Social Development of China
	Distance from township government	2023	State Statistical Bureau
	Distance from secondary road	2003	OSM
	Distance from secondary road	2023	OSM
	Food prices, grain yield	1988–2023	Compilation of cost and benefit data of agricultural products in China, Shandong Statistical Yearbook

presents the data sources for land use and driving factors analyzed in this study. Data utilized in the FLUS model, such as elevation, population density, temperature, precipitation, and proximity to national or provincial highways, were primarily sourced from the Resources and Environmental Science and Data Center of the Chinese Academy of Sciences, the National Qinghai–Tibet Plateau Science Data Center, CERN, and Open Street Map. Distances from national and provincial highways were calculated using Euclidean distance. Information on grain output and grain transaction prices, crucial for calculating ecosystem service value, was predominantly derived from the “Compilation of National Agricultural Product Cost and Income Data” and the “Shandong Statistical Yearbook”. Land cover data for the study area in 2003, 2013, and 2023 were obtained from China’s annual land cover dataset, with a 30 m resolution and spanning from 1985 to 2023, developed by Professors Yang Jie and Huang Xin from Wuhan University [19]. The dataset, encompassing 9 land cover types, demonstrated an overall accuracy of 80%. Through a secondary classification method, considering field conditions, the land use in the study area was categorized into cultivated land, forest land, grassland, water area, construction land, and unused land.

2.3. Methodology

This study established an integrated modeling framework to assess and simulate ecosystem resilience in the Jinan metropolitan area.

(1) The Resistance–Adaptability–Resilience (RAR) model was developed to systematically evaluate ecosystem resistance, adaptability, recovery, and resilience by integrating land use dynamics with landscape pattern indices. (2) The Optimal Multi-layered Geo-Detector (OMGD) model was introduced to identify dominant drivers through optimal spatial scale selection and interaction analysis (single-factor, dual-factor, and three-factor detection). (3) The Markov–FLUS model simulated ecosystem resilience evolution under three scenarios—inertial development, cultivated land protection, and ecological priority. High-resilience zones from 2023 were designated as restricted conversion areas to optimize resilience under ecological priority constraints.

2.3.1. Urban Ecosystem Resilience Model

Urban ecosystems exhibit resilience by effectively buffering, adapting to, and recovering from external disturbances. This study employed the Resistance–Adaptability–Resilience (RAR) model to establish an assessment framework [20]. The model emphasized the system’s dynamic attributes and feedback loops to depict urban ecology’s capacity to endure stress, recuperate, and maintain progress in the face of external threats and disruptions [21], ultimately aiming to “return to the original state”. Comprising resistance, adaptability, and resilience dimensions, the model mitigated the issue of oversimplification stemming from single-index evaluations [22]. By elucidating the interactions and harmonization among subsystems in urban ecosystems, it facilitated the realization of environmentally friendly and sustainable development in the region. The natural breakpoint method was used to classify each year of the same indicator, the classification threshold was obtained, and the average value of the multi-year threshold was taken to achieve the comparability and standardization of cross-year data. The data were divided into five categories: low, medium-low, medium, medium-high, and high. The specific formula for ecosystem resilience is outlined as follows [8]:

$$\mathrm{Re}silience=\sqrt[3]{P\times R\times E}$$

where P is resistance, R is adaptability, and E is resilience, and all are normalized.

(1) 

Urban Ecosystem Resistance Model

Resistance in ecology denotes a system’s capacity to withstand external disruptions while upholding its structural and operational equilibrium, serving as a pivotal gauge of ecosystem robustness and longevity. This resilience can be gauged through the prism of ecosystem service value, with the net agricultural output per unit area serving as a benchmark for assessing the worth of ecosystem services [23]. Ecosystem resistance is intricately linked to its inherent capacity to deliver ecosystem services. The evaluation of ecosystem resistance entailed the computation of ecosystem service value (ESV) using the equivalent factor method. The ESV for distinct land categories in the Jinan metropolitan area across various years is delineated in

Table 2. Ecosystem service value (CNY/hm²) per unit area in Jinan metropolitan area from 2003 to 2023.

Class I Service	Category II Services	Year	Cultivated Land	Forest Land	Greensward	Waters	Construction Land	Unused Land
Supply services	Food	2003	774.29	264.17	346.15	728.74	0	0
		2013	1817.75	620.17	812.64	1710.83	0	0
		2023	2359.32	804.95	1054.76	2220.54	0	0
	Raw material	2003	364.37	601.21	510.12	209.51	0	0
		2013	855.41	1411.43	1197.58	491.86	0	0
		2023	1110.27	1831.95	1554.38	638.41	0	0
	Water supply	2003	18.22	309.72	282.39	7551.60	0	0
		2013	42.77	727.10	662.94	17,728.43	0	0
		2023	55.51	943.73	860.46	23,010.35	0	0
Adjustment services	Climate regulation	2003	610.32	1976.72	1794.53	701.42	0	18.22
		2013	1432.82	4640.61	4212.91	1646.67	0	42.77
		2023	1859.70	6023.21	5468.08	2137.27	0	55.51
	Gas regulation	2003	327.93	5921.04	4745.94	2086.03	0	0
		2013	769.87	13,900.46	11,141.75	4897.24	0	0
		2023	999.24	18,041.89	14,461.27	6356.30	0	0
	Waste treatment	2003	91.09	1758.09	1566.80	5055.66	0	91.09
		2013	213.85	4127.37	3678.27	11,868.85	0	213.85
		2023	277.57	5357.05	4774.16	15,405.00	0	277.57
	Hydrological regulation	2003	245.95	4317.81	3479.75	93,133.42	0	27.33
		2013	577.40	10,136.64	8169.19	218,643.47	0	64.16
		2023	749.43	13,156.70	10,603.08	283,785.01	0	83.27
Support services	Soil formation and retention	2003	938.26	2413.96	2186.23	847.16	0	18.22
		2013	2202.69	5667.11	5132.48	1988.83	0	42.77
		2023	2858.95	7355.54	6661.62	2581.38	0	55.51
	Maintain nutrient cycling	2003	109.31	182.19	163.97	63.77	0	0
		2013	256.62	427.71	384.94	149.70	0	0
		2023	333.08	555.13	499.62	194.30	0	0
	Biodiversity protection	2003	118.42	2195.34	1985.83	2322.87	0	18.22
		2013	278.01	5153.86	4662.00	5453.26	0	42.77
		2023	360.84	6689.38	6050.97	7077.97	0	55.51
Cultural services	Aesthetic landscape	2003	54.66	965.59	874.49	1721.66	0	9.11
		2013	128.31	2266.84	2052.99	4041.82	0	21.39
		2023	166.54	2942.22	2664.65	5246.03	0	27.76
Total		2003	3652.83	20,905.83	17,936.20	114,421.84	0	182.19
		2013	8575.51	49,079.30	42,107.69	268,620.96	0	427.71
		2023	11,130.46	63,701.74	54,653.04	348,652.53	0	555.13

.

(2) 

Urban Ecosystem Adaptation Model

The adaptability of an ecosystem is closely linked to its stability [24], as indicated by the ecosystem landscape structure stability index [25]. This index relies on the stability of landscape pattern organization, particularly landscape connectivity and spatial heterogeneity. These factors play a crucial role in ecological forecasting, regionalization, monitoring [26,27], and the targeted regulation of landscape ecological space [28,29]. Evaluating landscape stability involves assessing landscape heterogeneity and connectivity indices, both of which positively influence ecosystem adaptability. While landscape heterogeneity and connectivity represent distinct aspects of ecosystem landscape structure, they are not interchangeable and carry equal importance [30]. The indicators used in this study were consistent with those outlined in

Table 3. Landscape index and weight.

Type	Landscape Index, Landscape Metrics	Weight
Landscape heterogeneity	Shannon diversity index	0.25
	Area weighted average fractal dimension of patches	0.25
Landscape connectivity	Fragmentation of landscape	0.5

, building upon prior research findings [8].

(3) 

Urban Ecosystem Resilience Model

Ecosystem resilience, or ecosystem resolution, denotes the capacity of land use types resembling natural ecosystems to withstand external pressures. Built land types exhibit relatively lower resilience to such pressures, rendering them more susceptible to significant damage [25]. For further details, the ecological elasticity model’s setting coefficient, as introduced by Peng et al. [13], can be consulted. The indicators used in this study were consistent with those outlined in

Table 4. Ecological elasticity coefficients.

Cultivated Land	Forest Land	Greensward	Waters	Construction Land	Unused Land
0.3	0.8	0.6	0.8	0.2	1

.

2.3.2. Optimal Multi-Layered Geo-Detector Model

Geo-Detector is a tool utilized for detecting and analyzing spatial heterogeneity [19], which unveils the underlying factors influencing the spatial distribution patterns of dependent variables by assessing the explanatory power of independent variables on dependent variables. The Optimal Multi-layered Geo-Detector (OMGD) model can effectively investigate three or more variables in a discrete and interactive manner. It automatically identifies the most suitable discretization approach for single- or multi-factor combinations using Geo-Detector q statistics. The scale detector within the model can iterate through various spatial scales to identify the optimal spatial scale for analyzing spatially stratified heterogeneity (SSH), thereby enhancing the explanatory capacity of the identified drivers [31]. In this research, a total of 19 grid scales ranging from 100 m to 19,000 m were selected for driver detection at 100 m intervals. Previous research has demonstrated that both natural and socio-economic factors play a role in shaping the spatial and temporal heterogeneity of ecosystem resilience [32,33]. The human factors considered in this study encompass urbanization rate (UR), population density (POP), gross domestic product (GDP), night light index (NTL), distance from highways (DHS), distance from main roads (DMR), distance from secondary roads (DSR), and distance from township government sites (DTG). Natural factors include soil type (ST), elevation (DEM), slope (PRE), distance from water bodies (DWB), average annual temperature (TMP), average annual precipitation (PRE), and normalized vegetation index (NDVI).

2.3.3. Markov–FLUS Model

The Markov model is a predictive method that relies on the current state of events and transition probabilities [34]. Liu Xiaoping et al. [35] incorporated an artificial neural network algorithm into the traditional cellular automata (CA) framework to model land use distribution by considering the existing land use and various influencing factors. By integrating the Markov model with the FLUS model, accurate simulations of land use changes can be achieved [36]. In this study, land use data for 2023 was generated through training and simulation using data from 2003 and 2013. The overall land use layout was essentially the same, with a Kappa coefficient of 0.83 and Overall Accuracy of 0.91. The constructed FLUS model had a high degree of accuracy and could be used for the multi-scenario simulation of land use changes for the year 2033. Subsequently, simulations for 2033 were conducted based on the actual land use data for 2023, with parameter settings informed by previous research [37].The neighborhood factor parameters are shown in

Table 5. Parameters of neighborhood factors.

Land Use Type	Cultivated Land	Forest Land	Greensward	Waters	Construction Land	Unused Land
Neighborhood weights	0.220053	0.218255	0.106687	0.037673	0.417226	0.000106

.

The neighborhood weights were calibrated using the following formula:

$${\mathrm{W}}_{\mathrm{i}}={\displaystyle \frac{{TA}_i-{TA}_{min}}{{TA}_{max}-{TA}_{min}}}$$

where W_i = neighborhood weight for the i-th land type, TA_i = expansion area of the i-th land use type (km²), TA_min = minimum land use expansion area across all types (km²), and TA_max = maximum land use expansion area across all types (km²).

(1) 

Natural Development

Based on the land use change trend observed between 2003 and 2023, as well as data from digital elevation models (DEMs), precipitation, temperature, gross domestic product (GDP), population, and other pertinent driving factors within the research area, our simulation forecasted the extent and geographical allocation of different land use types in 2033 in the absence of intervention.

(2) 

Protection of Cultivated Land

Ensuring the security of cultivated land is fundamental, and safeguarding cultivated land is a key national policy. To enhance protection, the rate of converting cultivated land to construction land was reduced to 40% of the original rate under the scenario of inertia development.

(3) 

Ecological Priority

In the inertial development scenario, the likelihood of cultivated land being converted to construction land was reduced to 30% of the initial value. Additionally, the probability of the other four land types transitioning to construction land was adjusted to 50% of the original probability. The high-value area of ecosystem resilience in 2023 was designated as a restricted conversion zone to safeguard regions with robust ecosystem resilience.

3. Results and Analysis

3.1. Spatial and Temporal Analysis of Urban Ecosystem Resilience

3.1.1. Spatial and Temporal Distribution of Urban Ecosystem Resilience

As shown in

Table 6. Typical values of ecosystem resilience in Jinan metropolitan area from 2003 to 2023.

Urban Ecosystem Resilience	2003	2013	2023
Minimum	0.0332	0.0276	0.0207
Maximum	0.4940	0.4994	0.5081
Average value	0.1863	0.1876	0.1863

, the average ecosystem resilience value in the Jinan metropolitan area increased from 0.1863 in 2003 to 0.1876 in 2013, indicating a temporary improvement in ecological conditions. However, this value subsequently declined to 0.1863 by 2023, signaling a renewed deterioration. Our comparative analysis of the 2003–2023 period revealed an overall pattern of initial enhancement followed by degradation. Notably, the maximum values (0.4940, 0.4994, and 0.5083) demonstrated a continuous upward trajectory, reflecting a progressive rise in high-quality ecosystem resilience conditions. Concurrently, low-quality ecosystem resilience parameters exhibited persistent deterioration throughout the study period, as evidenced by the declining minimum values (0.0332, 0.0276, and 0.0207). These trends collectively reveal an asymmetric development pattern characterized by a functional intensification in the high-value zones coexisting with a heightened vulnerability in the low-value regions.

As shown in Figure 2, the areas rated as low and medium-low consistently accounted for the highest proportion (>67%) throughout the study period. The proportions of medium-level areas were 14.69%, 15.23%, and 14.60% in 2003, 2013, and 2023, respectively, while medium–high-level areas constituted 8.88%, 8.23%, and 8.59% and high-level areas comprised 8.48%, 9.10%, and 9.18% during the same periods. These dynamics indicate unstable ecosystem resilience in the Jinan metropolitan area from 2003 to 2023, with ecological conditions demonstrating an initial improvement followed by deterioration. The fluctuating proportions of medium- and high-resilience zones, coupled with the persistent dominance of low-resilience areas, reflect a systemic instability in regional ecosystem resilience trajectories.

As shown in

Table 7. Changes in the area and proportion of ecosystem resilience in Jinan metropolitan area from 2003 to 2023 (unit: km², %).

Urban Ecosystem Resilience	2003		2013		2023
	Area	Percentage	Area	Percentage	Area	Percentage
Low	9669.83	43.40%	9059.01	40.66%	9778.77	43.89%
Medium-low	5471.45	24.56%	5966.01	26.78%	5287.74	23.73%
Medium	3272.04	14.69%	3394.21	15.23%	3253.32	14.60%
Medium-high	1977.65	8.88%	1833.75	8.23%	1913.37	8.59%
High	1888.56	8.48%	2026.55	9.10%	2046.33	9.18%

, low-rated areas dominated the regional ecosystem coverage in individual years, accounting for over 40% of the total area, while high-rated areas consistently occupied a smaller proportion, never exceeding 10% of the total area. In 2023, high-rated areas reached a historic high of 9.18% of the total area. Conversely, low-rated areas hit their lowest recorded proportion in 2013, covering 9059.01 km². Notably, 2013 marked the smallest combined share of low- and medium-low-rated areas at 67.44% (15,025.02 km²), whereas 2023 saw the largest combined proportion of high- and medium-high-rated areas at 17.77% (3959.7 km²). These fluctuations highlight the dynamic yet fragmented nature of ecosystem resilience distribution across the study period.

As shown in Figure 3, the composite index of ecosystem resilience was categorized into five levels using the natural breakpoint method. The rise in grain prices notably enhanced the overall ecosystem resistance, particularly in mountainous and riverine regions such as the central Shandong Province mountains and the Yellow River basin, which have emerged as new high-value areas for ecosystem resilience accumulation and growth. This increase has further underscored the value of ecosystem services. Over the past two decades, regional ecosystem resilience has transitioned from a low to a moderately low level, indicating a strong correlation between resistance changes and land use patterns in plain regions. The trajectory and spatial distribution of ecosystem adaptability and resilience have remained largely stable, with minor localized fluctuations. The spatial distribution of ecosystem suitability exhibited a mixed pattern, with the primary concentration falling within the middle and lower levels of ecological adaptability and above. Notably, there was a significant distribution of values ranging from low to high in the “urban–rural” and “plain–mountain” gradients. The ecological resilience across vast areas was generally low, with high values predominantly observed in the mountainous regions of central Shandong and its vicinity. Over the years, a consistent pattern of “high values in the central region and low values in the periphery” has emerged and persisted. The Jinan metropolitan area’s overall ecosystem resilience demonstrated a distinct “two zones and three zones” configuration, delineated by the “Ching–Linzi” ecosystem resilience boundary. South of this boundary, encompassing the Luzhong mountain area and its surroundings, lies a globally significant high ecosystem resilience zone, which includes the majority of areas with the highest ecosystem resilience values. Conversely, north of this boundary, lower values were more prevalent. A prominent belt characterized by moderate to high ecosystem resilience values has formed along the main Yellow River thoroughfare. Additionally, there were distinct belts of low ecosystem resilience both to the north and south of the “Chiping–Linzi” region. Areas with low ecosystem resilience were also scattered sporadically within regions characterized by moderate to low ecosystem resilience values. The distribution pattern observed during the study period exhibited relative stability.

3.1.2. Spatial Distribution of Hot and Cold Spots of Ecosystem Resilience

As shown in Figure 4 from 2003 to 2023, a clear regional clustering pattern of ecosystem resilience was evident, with 99% of hot spots concentrated in high-value areas and 99% of cold spots located in low-value areas. The distribution of cold and hot spots exhibited significant regional variations. Over time, the previously significant cold spot areas gradually transitioned into non-significant or significant hot spot areas. Specifically, the northwest region shifted from being predominantly a 99% cold spot area to a non-significant area, while the emergence of new 99% hot spot areas was largely attributed to the conversion of non-significant areas. However, in northwest China, 90% of the hot spots transitioned to 99% cold spots, primarily due to rapid economic development and accelerated land use changes during this period.

3.1.3. Dynamic Monitoring of Urban Ecosystem Resilience

As shown in

Table 8. Changes in ecosystem resilience levels in Jinan metropolitan area from 2003 to 2023.

Change Grade	Level Change
Significantly worse	−4 (high to low)
Obviously worse	−3 (high to medium-low/medium-high to low)
	−2 (high to medium/medium-high to medium-low/medium to low)
Slightly worse	−1 (high to medium-high/medium-high to medium/medium to medium-low/medium-low to low)
No change	0 (no level change, e.g., excellent to excellent)
Slightly better	1 (above, and vice versa)
Obviously better	2 (above, and vice versa)
	3 (above, and vice versa)
Significantly better	4 (above, and vice versa)

, according to the ecosystem resilience classification of the Jinan metropolitan area, the detected changes were further categorized into nine levels and seven grades. The change levels range from −4 to 4, where positive values indicate improvements in ecosystem resilience, 0 denotes no change, and negative values signify a deterioration. Level 0 was classified as “No change”, level −1 as “Slightly worse”, level 1 as “Slightly better”, levels 2 and 3 as “Better”, and level 4 as “Significantly better”.

Table 9. Changes in ecosystem resilience levels in Jinan metropolitan area from 2003 to 2023.

Change Grade	2003–2013		2013–2023		2003–2023
	Area	Percentage	Area	Percentage	Area	Percentage
Slightly worse	989.92	4.59	1392.39	6.46	1812.25	8.40
No change	18,793.34	87.15	19,391.07	89.93	17,784.98	82.48
Slightly better	1751.37	8.12	772.54	3.58	1905.78	8.84
Obviously better	28.96	0.13	7.59	0.04	60.58	0.28

shows the changes in ecosystem resilience of the Jinan metropolitan area from 2003 to 2023. During the study period, the ecosystem resilience changes were categorized into four grades: Slightly worse, No change, Slightly better, and Obviously better. The No change grade occupied the largest area, accounting for over 80% of the total, with the maximum area (19,391.07 km², 89.83%) observed during 2013–2023. Among the areas with deteriorating ecosystem resilience, only one degradation grade existed, while two improvement grades were identified. The Slightly better grade covered less than 10% of the area, and the Obviously better grade accounted for less than 0.3%. Significant changes occurred in regions linked to direct economic activities, such as the conversion of cropland and forestland to construction and industrial land. The spatial distribution of ecosystem resilience in the Jinan metropolitan area Figure 5 revealed that the degraded areas were primarily clustered around urban zones and most water bodies. The ecological deterioration around water bodies was associated with recent urban sewage leakage, industrial wastewater discharge, and shifts in regional landscape patterns driven by the expansion of aquaculture. Additionally, areas with declining ecosystem resilience expanded near the “Chiping–Linzi” line. Overall, the ecosystem resilience of the Jinan metropolitan area remained largely stable, with only a few regions showing signs of improvement.

3.2. Analysis of Influencing Factors of Ecosystem Resilience

3.2.1. Single-Factor Detection Analysis

As shown in Figure 6, the Optimal Multi-layered Geo-Detector identified the optimal spatial scale and 11 influential factors that passed a 95% p-value test. In 2003, the top four driving factors by q value were SLOPE (0.920), DEM (0.833), TMP (0.726), and ST (0.556), indicating their significant explanatory power for the spatial distribution of ecosystem resilience in the Jinan metropolitan area. By 2013, the impact of PRE had notably increased, while TMP’s influence had decreased. The leading factors by q value were SLOPE (0.918), DEM (0.777), PRE (0.682), and TMP (0.606), with a general weakening in their explanatory power. Moving to 2023, the top four driving factors by explanatory power were SLOPE (0.898), DEM (0.824), TMP (0.727), PRE (0.640), and a notable increase in ST (0.626). The factors DSR, DWB, and DTG exhibited a consistently low explanatory power over the years. The findings indicate that SLOPE and DEM significantly contributed to ecosystem resilience throughout the study period; areas characterized by steep topographic slopes predominantly consist of forests and grasslands with minimal human interference, thereby exhibiting heightened resilience. Over time, the influence of TMP and PRE on resilience increased. The cooling effect associated with the rising altitudes in the central mountainous region of Shandong impeded the northward movement of summer humid air, creating a rain shadow effect on the southeastern slope. This led to ample precipitation, further bolstering ecosystem resilience.

3.2.2. Multi-Actor Interaction

As shown in Figure 7, in the two-factor detection analysis, the robust explanatory abilities of SLOPE, DEM, and other variables resulted in a dual-factor linear attenuation for most factors. However, there were instances, such as UR∩POP, where a dual-factor linear enhancement was observed, illustrating a nuanced scenario of “two strong factors becoming weak, and two weak factors becoming strong”. This suggests that the resilience of the ecosystem in the Jinan metropolitan area was primarily shaped by natural factors, albeit with influences from social and economic factors.

The interaction of SLOPE with other influencing factors exhibited a substantial increase in explanatory power, placing it in the top tier. Similarly, the interaction of DEM with other influencing factors also demonstrated a high level of explanatory power, albeit in the second tier. In 2003, the highest q value of 0.91 was attributed to SLOPE∩TMP, followed by SLOPE∩DEM, SLOPE∩DTG, SLOPE∩POP, SLOPE∩ST, and SLOPE∩URq, all with a q value of 0.9. In 2013, the top q value remained at 0.91 for SLOPE∩TMP, followed by SLOPE∩DEM (0.9), SLOPE∩PRE (0.9), SLOPE∩DTG (0.89), SLOPE∩POP (0.89), SLOPE∩ST (0.89), and SLOPE∩UR (0.89). The inclusion of SLOPE∩PRE (0.9) enhanced the explanatory power, particularly in the single-factor detection. By 2023, the highest q value remained at 0.90 for SLOPE∩TMP, followed by SLOPE∩DEM (0.99), SLOPE∩PRE (0.89), SLOPE∩DTG (0.89), SLOPE∩GDP (0.89), SLOPE∩ST (0.89), and SLOPE∩UR (0.89), resulting in an overall decrease in explanatory power. However, there was a notable increase in the explanatory power of POP∩PRE (0.75), TMP∩PRE (0.75), TMP∩ST (0.72), and POP∩ST (0.71), indicating a growing influence of PRE, TMP, and other factors.

As shown in Figure 8, three-factor combinations typically exhibit higher values compared to single- and two-factor combinations, indicating that when the layer count is restricted, the amalgamation of three factors is more effective in elucidating the spatially stratified heterogeneity of ecosystem resilience. The findings revealed pronounced layer demarcations, with additional factors interacting with SLOPE and DEM demonstrating substantial explanatory capacity. In 2003, the highest q value attained was 0.90, with the most significant intersections being SLOPE∩PRE∩DEM and SLOPE∩PRE∩TMP. By 2013, the top q value remained at 0.90, with the most prominent intersections being PRE∩ST∩TMP and PRE∩ST∩SLOPE. In 2023, the highest q value observed was 0.90, corresponding to the intersection of SLOPE, TMP, and PRE. Over the three-year period, the most significant q value remained at 0.9; however, notable variations were observed in the interaction effects, particularly a substantial increase in the three-factor interaction q value involving ST in 2013. Overall, besides SLOPE, DEM, PRE, and TMP, the three-factor interaction analysis revealed that ST emerged as a primary determinant in elucidating the spatial variability of ecosystem resilience. Natural conditions form the foundation for socio-economic operations. People’s production and living activities, through actions such as reshaping land use types, altering carbon emissions, and afforestation, exert feedback on natural ecosystems, thereby affecting ecosystem resilience. For example, accelerated urbanization leads to increased food demand, inevitably resulting in the conversion of lower slope mountainous areas into farmland, which alters ecosystem resilience. This interaction between urbanization and slope enhanced the explanatory power of the relationship between urbanization and slope changes in ecosystem resilience.

3.3. Ecosystem Resilience Simulation and Prediction

The high-resilience zones from 2023 were designated as restricted conversion areas to optimize resilience under ecological priority constraints. As shown in Figure 9, in each simulation scenario, the regional ecosystem resilience pattern transitioned from “two zones and three zones” to “two zones and one belt”. The distribution of high ecosystem resilience areas along the Yellow River and the mountains remained fundamentally unchanged, with the continued validity of the “Chiping-Linzi” line. The ecosystem resilience in the northwest has strengthened due to increased food prices, leading to the vanishing of the middle and low ecosystem resilience zone. Similarly, the resilience zone of the low ecosystem in the southeast has disintegrated due to local ecosystem dynamics.

In the context of natural development, fragmented patches of high- and low-value ecosystems were irregularly scattered, exhibiting a certain degree of local ecosystem resilience. In terms of cultivated land protection, ecosystem resilience displayed regional patterns primarily influenced by forest land conversion to cultivated land, leading to changes in or the disappearance of some forest patches. This was evident in the enhanced regional connectivity at a similar level of ecosystem resilience. The pressure of cultivated land protection resulted in the conversion of a significant area of grassland in the central mountainous region of Shandong Province into cultivated land, creating a landscape known as “cultivated mountain”, significantly reducing regional ecosystem resilience. This was characterized by a larger low-value area compared to high-value areas. Compared to the natural development and cultivated land protection scenarios, the originally designated high-resilience areas were effectively preserved, and the conversion of forest land to cultivated land was strictly restricted. Large-scale encroachment of cultivated land into mountainous areas (termed “cultivated land moving up mountains”) was prevented, resulting in a more extensive spatial distribution of fragmented ecosystem resilience. This outcome demonstrated a robust landscape diversity and an enhanced stability of ecosystem resilience. These findings validate the effectiveness of delineating conversion zones based on ecosystem resilience thresholds, providing an optimal solution for balancing cultivated land protection and ecological conservation.

4. Discussion

Current methodologies for assessing socio-economic–ecological systems are constrained by pervasive limitations including simplification bias, scale disjunction, and static modeling approaches. To address these challenges, this study developed a multi-dimensional dynamic assessment framework through the systematic integration of the RAR model, the OMGD model, and the Markov–FLUS model. The ecosystem resilience of the Jinan metropolitan area exhibited a spatial pattern characterized by higher resilience in the southern regions and lower resilience in the northern areas. This pattern was primarily influenced by topographic barriers, the Yellow River ecological corridor, and land use in the plains. In the Yellow River Basin, the ecosystem service value significantly contributed to ecosystem resilience to a greater extent compared to other river systems. This disparity may be attributed to the combined effects of soil fertility and hydrological regulation resulting from sediment deposition [38]. The cultivated land protection policy in the Zhongshan Area of Shandong Province restricts construction land expansion but diminishes ecological restoration capacity, illustrating a trade-off between cultivated land utilization and ecological resilience in mountainous regions. The simulation outcomes revealed that solely constraining construction land expansion (under the cultivated land protection scenario) may result in the spatial displacement of ecosystem resilience [39], characterized by the coexistence of fragmented high-resilience areas and contiguous medium- to low-resilience areas. In contrast to the Yangtze River Basin, the Yellow River Basin prioritizes ecological preservation as a fundamental criterion, subsequently emphasizing resource efficiency and sustainable development within the basin [40]. In addition, unlike existing studies integrating geo-detectors with CA models, the tri-model framework of RAR, OMGD, and Markov–FLUS proposed in this research achieved three key breakthroughs: (1) Through the OMGD model’s multi-scale interaction detection, it revealed the nonlinear enhancement effects (q value reaching 0.9) of the three-factor combination—slope, precipitation, and soil type (SLOPE∩PRE∩ST)—on the spatial differentiation of mountain resilience. (2) By combining the ecosystem service value (ESV) and landscape stability analysis in the RAR model, it identified the ecological barrier function of Jinan’s “Chiping–Linzi Resilience Boundary, ” addressing the biases inherent in single-landscape index evaluations (e.g., Shannon diversity). (3) The dynamic “Resilience Red Line-Development Permit” coupling mechanism provided a scalable “land-for-ecology” policy tool for mountain–plain transition zones, demonstrating greater adaptability compared to static protection zoning in the Yellow River Basin and the Yangtze River Economic Belt.

This study introduces a novel dynamic coupling mechanism termed “toughness red line-development permission”, wherein regions with high toughness in 2023 are designated as red lines to restrict development activities. The proposed “Resilience Red Line-Development Permit” mechanism can be operationalized through three actionable steps, informed by the 2023 land use and ecosystem resilience data from Jinan City. Step 1: Delineating resilience red lines. Criteria: define boundaries based on high ecosystem resilience zones identified through RAR model outputs. Threshold integration: incorporate terrain constraints (slope > 25°), historical floodplains, and biodiversity corridors. Step 2: Development permit allocation. Quota system: allocate construction land quotas inversely proportional to economic development levels and disaster resilience capacities (e.g., red zones: 0.05 km²/km²; yellow zones: 0.15 km²/km²). Tradable permits: establish a regional permit trading platform, enabling counties with surplus resilience capacity (e.g., Southern Mountain Area) to transfer quotas to high-demand urban cores (e.g., Lixia District, Jinan). Step 3: Adaptive monitoring. Institutional coordination: create a cross-departmental oversight committee integrating resources, environmental, ecological, and planning agencies. Tech-driven enforcement: deploy IoT sensors and monthly satellite monitoring (30 m resolution) to track violations (e.g., unauthorized land conversions). Blockchain pilot: implement blockchain-based permit tracking in Jinan’s New and Old Kinetic Energy Conversion Zone to enhance transparency. Different regions have different natural and social conditions, and there is an interplay between ecological protection and land development. Therefore, the regions should be delimited according to the local conditions to promote the implementation of policies. In the Yellow River floodplain area, development permits are regulated by substituting construction land indicators to ensure the preservation of cultivated land. This approach aligns with the objective of enhancing the ecological barrier in the lower reaches of the Yellow River as outlined in the Ecological Protection and High-quality Development Plan of the Yellow River Basin in Shandong Province. Furthermore, it directly contributes to SDG11 (Sustainable Cities and Communities) and SDG15 (Land Ecology) of the United Nations Sustainable Development Goals. Notably, elements of this mechanism have been incorporated into the four strategies outlined in the “Fine Building High-quality People’s City” initiative of Jinan City, as detailed in the 2024 Government Work Report. This framework can serve as a valuable policy model for other mountain–plain metropolitan areas, such as the Zhengzhou metropolitan area. By integrating ecological adaptability into the rigid confines of spatial land planning [41,42], the ecosystem resilience of the Jinan metropolitan area exhibits distinctive features, particularly in response to rapid urbanization and agricultural modernization.

The resolution limitations of the land use data (30 m spatial resolution remains insufficiently sensitive to detect subtle land use changes) and methodological constraints in calculating the ecosystem service value persist. Land use change, as a complex dynamic process driven by synergistic multi-variate factors, exhibits temporal continuity and variability in conversion rules across different intervals [43], Better resolution data is better for capturing more subtle changes. While the OMGD model identifies dominant factors, its analysis of interactions is confined to statistical correlations, with the exclusion of factors at p ≥ 0.05 further limiting causal mechanism exploration. The Markov–FLUS model’s reliance on historical transition probabilities introduces “path dependence” biases due to the incomplete handling of external uncertainties. Additionally, fixed parameters in ecological priority scenarios neglect social adjustment mechanisms, and the findings from this region-specific structural analysis require broader validation.

Integrating multi-source high-resolution remote sensing data with field monitoring will enhance the sensitivity to land use variations, enabling dynamic “ecosystem resilience–human activity” feedback modeling. Coupling the OMGD model with complementary models could improve the analytical sensitivity and robustness. A social network analysis should be adopted to examine how policy communication networks and public environmental behaviors jointly drive resilience evolution, while adaptive policy integration into ecological scenarios would better simulate socio-institutional responses. A climate policy-integrated ecosystem resilience simulation platform, incorporating climate projections and land policy dynamics, is crucial for advancing risk assessment and adaptive planning. Systematic cross-regional validation frameworks should also be prioritized to generalize the findings.

5. Conclusions

Using the Jinan metropolitan area as a case study, this research assessed ecosystem resilience from 2003 to 2023 employing the Resistance–Adaptability–Resilience (RAR) model. This study analyzed and elucidated the factors influencing ecosystem resilience and the underlying mechanisms using the OMGD model. Furthermore, this study forecasted ecosystem resilience in 2033 through a Markov–FLUS simulation. Our conclusions are as follows:

(1)

The Jinan metropolitan area exhibits predominant ecosystem resilience at middle and low levels, with a stable “two regions and three zones” spatial pattern delineated by the Chiping–Linzi boundary, along which the resilience values demonstrate a diminishing gradient. Regional clustering is evident, marked by pronounced cold and hot spot distributions. Rapid land-use conversion has driven the transition of former cold spots to a non-significant status while elevating specific areas to significant hotspots. Our quantitative analysis revealed fluctuating average resilience (0.1863 in 2003 to 0.1876 in 2013 to 0.1863 in 2023), accompanied by divergent trends in extreme values: the maximum resilience increased steadily (0.4940 to 0.5081), while the minimum values declined sharply (0.0332 to 0.0207). Concurrently, the proportional coverage of both high- and low-rated areas expanded, with high-rated zones reaching 9.18% (2023) and low-rated areas persistently exceeding 40% of the total. These dynamics collectively illustrate an asymmetric development paradigm, characterized by functional intensification in the high-value regions coexisting with escalating vulnerability in the low-value zones, as evidenced by the widening resilience gradient (46.8% increase in value extremes) and persistent spatial polarization (>67% dominance of low–medium low areas).

(2)

The OMGD model revealed that ecosystem resilience’s global distribution was primarily influenced by natural factors, while also being impacted by social and economic factors. Factors such as slope, elevation, and temperature were identified as key drivers shaping regional ecosystem resilience distribution. Additionally, soil type emerged as a significant factor influencing the spatial distribution of ecosystem resilience, as indicated by the three-factor interaction analysis.

(3)

The simulation analysis of three scenarios—inertia development, cultivated land protection, and ecological priority—in the study area in 2033 indicated that the ecological priority scenario was the optimal model for enhancing land use structure and bolstering ecosystem resilience in the Jinan metropolitan area. Additionally, a dynamic relationship was observed between the plains and mountains, as well as between cultivated land protection and ecological priority.