Land Use Change Dynamics in Metropolitan Areas: A Cross-Regional Comparison Across China, Japan, and the United States

Abstract

Metropolitan areas are key carriers of economic growth and regional sustainable development. Comparing land use and land cover changes (LUCC) across multiple metropolitan areas can provide pathway references for the sustainable development of emerging metropolitan areas. However, current approaches are limited by two major shortcomings: (1) the lack of methods capable of providing a comprehensive comparison of LUCC processes across multiple metropolitan areas and (2) the difficulty in effectively visualizing the results of comprehensive and complex LUCC analyses. Here, we introduce a novel comparative intensity analysis (CIA) model to comprehensively compare LUCC processes across metropolitan areas. The challenge of visualization is addressed by the newly proposed Intensity Deviation Maps. Three metropolitan areas were selected as research objects: the Chang-Zhu-Tan Metropolitan Area (CZT) in China, the Chukyo Metropolitan Area (CMA) in Japan, and the Dallas–Fort Worth (DFW) Metropolitan Area in the United States. Findings reveal a metropolitan evolution characterizing three stages and mechanisms underlying cross-regional differences: (1) The first stage is rapid, unorganized expansion, which requires enhanced regulation to achieve sustainable land resource utilization; (2) the second stage shifts from external expansion to internal renewal, with a focus on urban resilience and the well-being of residents; and (3) the third stage seeks external breakthroughs to expand its influence. We uncover context-shaped heterogeneous LUCC: policy-driven rapid CZT construction land expansion, population-driven high intensity in the DFW, and low CMA intensity amid stagnation. This study deepens understanding of global metropolitan LUCC and informs sustainable land use planning. The CIA model provides methodological support for cross-regional LUCC research.

1. Introduction

Land use and land cover change (LUCC) profoundly impacts global carbon emissions, ecological environment, air quality, landscape, and biodiversity [1,2,3] and drives global environmental change [4]. As the primary engines of global urbanization, metropolitan areas host 57% of the global population on less than 2% of the terrestrial surface [5]. These regions are critical for economic growth and international competitiveness but are also hotspots of anthropogenic land-use intensity [6].

While mature metropolitan areas in developed nations have established stable spatial structures, emerging metropolitan areas in developing countries are experiencing rapid, often unorganized expansion [7]. These newly established metropolitan areas are at high risk of replicating the urban diseases faced by their mature counterparts. Therefore, comparing LUCC trajectories across metropolitan areas at different developmental stages—such as those in China, Japan, and the United States—is essential for understanding global urbanization mechanisms and formulating sustainable planning frameworks. However, existing LUCC research has predominantly focused on megacities and well-established metropolitan areas [8,9,10], often due to the prevalence of urban challenges within these areas. These emerging metropolitan areas are at high risk of replicating the problems currently faced by their more mature counterparts. However, rapidly developing emerging metropolitan areas often receive insufficient attention due to the relatively limited visibility of their urban challenges.

China, the fastest-developing country over the past 50 years, has established 17 new metropolitan areas since 2021 [11], which are key to its new urbanization. Japan, the first to propose metropolitan areas, has the Tokyo Metropolitan Area as a model [12] and is highly urbanized, with most people in compact metropolitan zones [13]. The USA, the most developed country, was the first to standardize metropolitan statistical areas [14]. These three countries represent rapid economic development, high urbanization, and advanced development; analyzing their metropolitan LUCC trajectories offers transferable insights for global metropolitan development. Numerous scholars have conducted extensive research on specific regions within these three countries [15,16,17]. Comparative studies among these three countries have primarily focused on urban agglomerations. For example, Cao et al. [18] leveraged impervious surface datasets to conduct a comparative analysis of spatial structural disparities between China’s Beijing–Tianjin–Hebei urban agglomeration and the USA’s Boston urban agglomeration; Yang et al. [19] quantified and compared impervious surface expansion rates across urban agglomerations in coastal China, Japan, and the United States. These studies have thoroughly examined expansion patterns and rates in core cities and global metropolises. However, China’s metropolitanization is a relatively recent phenomenon, and the literature contains few comparative LUCC analyses at the metropolitan scale. What are the differences in LUCC among metropolitan areas in developed and developing countries? Do mature metropolitan areas and emerging metropolitan areas exhibit different LUCC patterns? Can a conceptual framework be derived to characterize their developmental trajectories? These questions currently lack clear answers.

Quantifying the process of LUCC is a crucial step towards understanding LUCC in metropolitan areas. Several methods are available for quantifying LUCC, such as the single and integrated land use dynamic degree [20] and the land-use transition matrix [21]. The land-use transition matrix, extensively used in LUCC studies, quantifies both the magnitude and composition of land-category conversions [22,23]. However, as Pontius [24] noted, while transition matrices effectively present the area of conversions between categories, they fail to account for the proportional relationship between the converted area and the total area of a specific category. In certain contexts, the intensity of a transition can be more meaningful than its absolute area. When analyzing LUCC, one should, as in the article [24], consider relative proportions by calculating the intensity of each component, in order to more comprehensively understand the characteristics of the change. To overcome the aforementioned limitation of the transition matrix, Aldwaik and Pontius [25] first proposed Intensity Analysis—an analytical framework tailored to identify land categories experiencing more pronounced changes in intensity throughout the LUCC process. The method offers a three-level analytical framework—interval, category, and transition—for measuring the intensity of land-use changes [26]. Owing to its distinct advantages in interpreting the dynamics of land use, Intensity Analysis has been widely applied in fields such as urban expansion monitoring and desertification assessment [27,28,29].

A key challenge in comparative LUCC studies is addressing differences in area size across regions. Current methods include the Single Land Use Dynamic Degree (SLUDD) and Intensity Analysis. The most commonly used SLUDD expresses the expansion or contraction of land types by calculating their net changes [30]. Kuang et al. [8] used SLUDD to compare urban land expansion rates in six megacities in China and the USA. He et al. [31] introduced the “urban growth rate”, a type of SLUDD, to quantitatively assess changes in the size of six urban agglomerations in China and the USA. SLUDD effectively addresses the issue of differing area sizes in cross-regional comparisons. However, SLUDD only reflects net change in a single land-use type, neglecting conversions with other land uses and failing to distinguish between gains and losses [30]. Intensity Analysis, which focuses on the intensity of change rather than the amount, is another effective method for cross-regional comparative studies [29]. Its hierarchical and graphical analysis can effectively reveal the details and mechanisms of land change [30]. However, the method was originally designed for longitudinal analysis within a single study area [25]. Consequently, the initial framework lacks a complete and interpretable mathematical formulation for comparative studies involving multiple regions. There is therefore a pressing need to adapt and enhance Intensity Analysis to leverage its full potential in cross-regional LUCC comparisons.

Intensity Analysis determines the nature of a transition by comparing its intensity to a uniform intensity value. The results are typically visualized using bar charts, where a “uniform line” represents the uniform intensity. If a bar, representing a specific transition, extends beyond this line, the transition is considered “fast/active/target”. Conversely, if the bar falls short of the line, the transition is deemed “slow/dormant/avoid” [30]. Generating a large number of charts makes it difficult to extract salient patterns. Furthermore, traditional Intensity Analysis often struggles to intuitively display the relative magnitude of deviations across regions with vastly different sizes, which can introduce uncertainty in comparative interpretations. Similarly, an increase in the number of land categories expands the number of bars in each chart, further complicating the visual interpretation. To reduce complexity and the number of graphs, researchers often choose short or long-interval time periods [32,33,34]. This is why long-term and multi-temporal scale studies are currently lacking in applied research using Intensity Analysis.

In response to the aforementioned gaps, we refine Intensity Analysis by establishing a comparative intensity analysis (CIA) model. We demonstrate the validity and advantages of the CIA model by explicitly comparing its visualization capabilities and interpretability with traditional Intensity Analysis frameworks. We apply it to three metropolitan areas in China, Japan, and the United States. Using this model, we compare LUCC dynamics over 35 years (1985–2020) at the interval, category, and transition levels. We then visualize the resulting patterns with Intensity Deviation Maps, enabling readers to distill key changes from complex LUCC information. The research provides applied insights into the development of emerging metropolitan areas, helping them to formulate planning frameworks and achieve sustainable development.

These three regions were strategically selected to construct a robust and comparable empirical foundation for the proposed conceptual framework. As detailed in

Table 1. Key Characteristics of the Study Areas (1985–2020).

Characteristic	CZT (China)	CMA (Japan)	DFW (USA)
Latitude	27°12′ N~28°39′ N	33°43′ N~36°28′ N	32°03′ N~33°26′ N
Area (104 km2)	1.89	2.15	2.33
Topography	Plains (80%) & Hills	Mountains (40%) & Plains	Low Plains
Number of important cities	9 Cities	7 Cities	8 Cities
Population 1985 (Million)	8.48	10.26	3.60
Population 2020 (Million)	14.68	11.29	7.67
Pop. Growth (1985–2020)	+73%	+10%	+113%
GDP (2020, Billion USD)	237.2	522.9	546.5
Metropolitan Stage	Emerging (Rapid Expansion)	Mature (Stagnation/Renewal)	Developed (High Growth)

, they share critical similarities that serve as control variables: all are situated near 30° N latitude with subtropical monsoon climates, possess comparable land areas, host populations of approximately 10 million, and exhibit similar polycentric urban structures. Furthermore, they exhibit a comparable economic scale (as of 2020) and serve as the primary economic engines within their respective regions. By controlling for these natural and scalar variables, we can more effectively isolate and identify the impact of their distinct developmental stages on LUCC mechanisms.

2. Materials and Methods

2.1. Study Area

The study focuses on three representative metropolitan areas (Figure 1). To ensure spatial stability over time and align with the jurisdictions where government policies are implemented, the boundaries for all three regions were selected based on administrative units. Table 1 summarizes their key geographic and socioeconomic characteristics.

The CZT (Figure 1a) is located in Hunan Province, south-central China. It covers the administrative jurisdictions of Changsha, Zhuzhou, and Xiangtan, representing a rapidly emerging metropolitan area in a developing economy. The CMA (Figure 1b) is situated in central Japan, centered on Nagoya City. The study area includes Aichi, Mie and Gifu Prefectures, serving as a mature industrial hub connecting Tokyo and Osaka. The DFW (Figure 1c) Metropolitan Area, also known as the Metroplex, is located in northern Texas, USA. It comprises the core counties anchored by Dallas and Fort Worth, representing a high-growth model in a developed economy.

2.2. Data Source and Processing

The core data of LUCC analysis are land-use data. Long-term time series land-use/land-cover datasets at the global scale are relatively scarce. In 2024, the GLC_FCS30D dataset—a long-term, high-resolution land-use and land-cover product—was released [35]. The dataset has a 30 m spatial resolution and attains an overall accuracy of 80.88% (±0.27%) across ten primary land-use/land cover classes. Therefore, the land-use data were selected from GLC_FCS30D. Eight time points of GLC_FCS30D were used in this study: 1985, 1990, 1995, 2000, 2005, 2010, 2015, and 2020. Drawing upon the National Standard of China (GB/T 21010-2017 [36]) and the “Production–Living–Ecological Space” theory [37], we reclassified the original land-use categories into three functional classes: construction land (CL), ecological land (EL), and agricultural land (AL). Specifically, categories such as forests, shrublands, grasslands, wetlands, and water bodies were grouped into EL because their primary utility is ecological regulation and service provision, distinguishing them from the production-oriented AL and the living/economic-oriented CL. This reclassification strategy serves two primary purposes: (1) to prioritize the cross-regional applicability of the CIA model by standardizing diverse land covers into comparable functional categories and (2) to minimize the interference of minor internal variations unrelated to macro-comparative analysis. The reclassification rules are shown in

Table 2. The reclassification rules and rationale based on land use function and Eco-environmental Quality Index (EQI).

Original Land Types	Reclassified Land Types	EQI	Primary Classification Rationale
Impervious surfaces	Construction Land (CL)	0.010	Dominated by anthropogenic development; serves as the primary carrier for human settlement and economic activities.
Cropland	Agricultural Land (AL)	0.293	Primary function is food production; represents semi-natural ecosystems managed for economic output.
Forest	Ecological Land (EL)	0.883	Core ecological component; functions include carbon sequestration, water conservation, and biodiversity support.
Shrubland	Ecological Land (EL)	0.883	Natural vegetation cover with no direct economic output; provides soil conservation and ecological buffering.
Grassland	Ecological Land (EL)	0.798	Essential for soil/water conservation and ecological buffering; functionally distinct from intensive agriculture.
Wetland	Ecological Land (EL)	0.521	Critical for biodiversity maintenance, water purification, and hydrological regulation.
Water body	Ecological Land (EL)	0.521	Provides essential ecosystem services including hydrological regulation and habitat provision.

Note: The Eco-environmental Quality Index (EQI) values are derived from Yang et al. [37]. Other land categories such as Tundra and Permanent Snow/Ice were excluded from the table as they have negligible or zero distribution in the study areas.

.

The vector boundary data for CZT, CMA and DFW are, respectively, sourced from the National Catalogue Service for Geographic Information (www.webmap.cn, accessed on 6 March 2025), the United States Geological Survey (USGS, https://apps.nationalmap.gov/downloader/, accessed on 9 March 2025) and the Geospatial Information Authority of Japan (www.gsi.go.jp/kiban, accessed on 14 March 2025). The DEM data with a 30 m resolution for all areas are derived from the ASTER GDEM data product on USGS. ArcGIS Pro (Version 3.4.3) and R (Version 4.5.1) were used for mapping and data analysis.

2.3. CIA Model Summary

The CIA model is an enhancement of Intensity Analysis [25], with its workflow shown in Figure 2. It includes interval, category, transition levels, and Intensity Deviation Maps, with two core improvements: (1) Adding regional symbol $r$ to upgrade mathematical formulas and (2) Introducing intensity deviation to standardize change intensity measurement, enabling granular differentiation of LUCC states and addressing traditional Intensity Analysis’ binary classification limitation.

The CIA model retains Intensity Analysis’ core logic. It focuses on LUCC intensity instead of direct size comparison, given that size comparison lacks meaning across regions of different scales. Due to formula complexity from diverse symbols, the model’s core logic is graphically presented in Figure 3 for clarity.

2.3.1. Interval Level

Traditional Intensity Analysis uses region-specific average intensity as a reference; this study’s improved “uniform intensity” is calculated across all regions. Comparing each region-interval’s change intensity with overall uniform intensity identifies regional change speed in specific intervals.

Equation (1) calculates the intensity ${(S}_{rt})$ of gross change in region $r$ during interval $t$. Equation (2) shows the uniform intensity $\left(U\right)$ of change for all intervals in all regions [25]. Unlike a simple arithmetic mean of change rates, $U$ represents the “Equal Intensity” hypothesis: it assumes that the total change occurring across all metropolitan areas is distributed uniformly relative to their combined area size. This approach prevents smaller regions with high volatility from skewing the baseline, ensuring that the comparison is weighted by the actual land scale.

Equation (3) defines the intensity deviation $\left(D_{rt}\right)$. Here, the value 0 serves as a heuristic threshold rather than a statistical significance level. A positive $D_{rt} (D_{rt}>0)$ indicates that the region’s change speed is faster than the cross-regional average, while a negative value implies a slower rate.

$$S_{rt}={\displaystyle \frac{{\sum }_{j=1}^J\left[\left({\sum }_{i=1}^J{C}_{rtij}\right)-C_{rtjj}\right]}{{\sum }_{j=1}^J\left({\sum }_{i=1}^J{C}_{rtij}\right)}}\times 100\mathrm{\%}$$

(1)

$$U={\displaystyle \frac{{\sum }_{r=1}^R{\sum }_{t=1}^T{\sum }_{j=1}^J\left[\left({\sum }_{i=1}^J{C}_{rtij}\right)-C_{rtjj}\right]}{{\sum }_{r=1}^R{\sum }_{j=1}^J{\sum }_{i=1}^J{C}_{rtij}}}\times {\displaystyle \frac{100\%}{T}}$$

(2)

$$D_{rt}=S_{rt}-U$$

(3)

where $S_{rt}$ stands for the change intensity in region $r$ during interval $t$. $U$ is the uniform intensity across all regions. $D_{rt}$ is the intensity deviation in region $r$ during time interval $t$. The subscripts $i$ and $j$ denote the land types. $J$ is the number of categories. $C_{rtij}$ identifies the size in region $r$ that changes during interval $t$ from category $i$ to category $j$. $R$ and $T$ are the number of regions and intervals, respectively. In the following, the meanings of the notions will remain as stated above.

2.3.2. Category Level

The category level further extracts the conversion intensity information of each land category based on the interval level. Equations (4) and (5) provide a measure of the intensities associated with gains and losses for each land type [25]. Step by step, these formulas work as follows: (1) The numerator calculates the total area of a specific land category that changed and (2) the denominator normalizes this change by the category’s size at the end (for gain) or start (for loss) of the interval. This normalization ensures that categories with small initial areas but high turnover are correctly identified as active. By comparing the gain intensity $G_{rtj}$ with the uniform intensity $S_{rt}$ of the corresponding period, it is determined whether the gain of the land category is active or dormant.

Equation (6) is the mathematical expression of the intensity deviation at the category level. It is the difference between the conversion intensity of a land category and the uniform intensity of the interval level in which it is located. The larger $D_{rtj}$ is, the more active the gain of land category $j$ becomes in region $r$ during interval $t$; conversely, the more dormant it is. The larger $D_{rti}$ is, the more active the loss of land category $i$ becomes in region $r$ during interval $t$; conversely, the more dormant it is.

$$G_{rtj}={\displaystyle \frac{\left({\sum }_{i=1}^J{C}_{rtij}\right)-C_{rtjj}}{\left({\sum }_{i=1}^J{C}_{rtij}\right)}}\times 100\mathrm{\%}$$

(4)

$$L_{rti}={\displaystyle \frac{\left({\sum }_{j=1}^J{C}_{rtij}\right)-C_{rtii}}{{\sum }_{j=1}^J{C}_{rtij}}}\times 100\mathrm{\%}$$

(5)

$$\{\begin{array}{c}{D}_{rtj}=G_{rtj}-S_{rt}\\ D_{rti}=L_{rti}-S_{rt}\end{array}$$

(6)

2.3.3. Transition Level

The transition explores the intensity of land conversion from various categories to a particular category, expanding on the framework set by the category level. Equation (7) quantifies the transition intensity, $R_{rtij}$, from land category $i$ to $j$, where $i$ ≠ $j$. Equation (8) quantifies the uniform transition intensity, $W_{rtj}$, from other land categories to category $j$ [25]. If $R_{rtij}$ > W_rtj, then the gain of category $j$ targets category $i$ in region $r$ during interval $t$. If $R_{rtij}$ < $W_{rtj}$, then the gain of category $j$ avoids category $i$.

Equation (9) determines the intensity deviation of the conversion from land category $i$ to $j$ in region $r$ during interval $t$ [38]. Step-by-step comparison works as follows: We compare the specific transition intensity ${(W}_{rtij})$ against the uniform intensity ${(W}_{rtj})$. The larger $D_{rtij}$ is, the more the gain of land category $j$ is targeted from land category $i$. On the contrary, the smaller the $D_{rtij}$ is, the more the gain of land category $i$ is avoided from land category $j$.

$$R_{rtij}={\displaystyle \frac{C_{rtij}}{{\sum }_{j=1}^J{C}_{rtij}}}\times 100\%$$

(7)

$$W_{rtj}={\displaystyle \frac{\left({\sum }_{i=1}^J{C}_{rtij}\right)-C_{rtjj}}{{\sum }_{i=1}^J\left[\left({\sum }_{j=1}^J{C}_{rtij}\right)-C_{rtij}\right]}}\times 100\mathrm{\%}$$

(8)

$$D_{rtij}=R_{rtij}-W_{rtj}$$

(9)

2.3.4. Intensity Deviation Maps

The Intensity Deviation Maps are designed to overcome the difficulty in visualizing the complex results generated by Intensity Analysis. The CIA model typically involves multiple study areas, intervals, and land categories. The information on LUCC will increase exponentially with the increase in study areas, interval, and land category. It is often challenging for decision-makers to extract salient and systematic patterns from intricate change categories; accordingly, visualizing CIA model outputs is necessary for clarity.

The Intensity Deviation Maps can be decomposed into three maps: the Interval Deviation Map, the Category Deviation Map, and the Transition Deviation Map. These maps correspond to the three levels of the CIA model, respectively. In the Intensity Deviation Maps, bubbles are employed as carriers for information representation (Figure 4). Bubbles incorporate two information dimensions (size and color) corresponding to intensity and intensity deviation, respectively. The gradation of size was achieved through Jenks Natural Breaks Classification [39]. The size of bubbles indicates the magnitude of change intensities. The larger the bubble, the greater the intensity. The color of bubbles represents the degree of intensity deviation. Higher saturation of red indicates faster/more active/more targeted changes. Conversely, higher saturation of blue indicates slower/more dormant/more avoidant changes.

Beyond intuitively expressing the intensity and intensity deviation, the Intensity Deviation Maps further characterize the stationarity of LUCC. When bubbles within the same row exhibit similar coloration, it indicates that the LUCC is stationary. By examining the horizontal color differences of the bubbles, the researchers can swiftly assess the stationarity of the conversion process throughout the entire period [38].

2.3.5. Conceptual Comparison with Traditional Intensity Analysis

To illustrate the advancement of the CIA model, consider a comparison between a large Region A (10,000 km², change 100 km²) and a small Region B (1000 km², change 50 km²).

In Traditional Intensity Analysis, each region is analyzed in isolation. Region A has a change intensity of 1%, and Region B has 5%. Without a unified baseline, one can only conclude that “B is faster than A,” but cannot quantify their deviations relative to the entire system.

In the CIA Model, we calculate a unified intensity $\left(U\right)$ for the combined area (11,000 km², total change 150 km²), resulting in $U$ ≈ 1.36%. We then derive the intensity deviation: Region A (1% − 1.36% = −0.36%) is defined as “Slow/Dormant,” while Region B (5% − 1.36% = +3.64%) is “Fast/Active.” This standardization allows for the direct visualization of “Deviation” (Bubble Color) across multiple regions, which is impossible with traditional independent analysis.

3. Results

3.1. Overall Analysis of LUCC

Figure 5 presents the percentage of land categories for the CZT, DFW, and CMA from 1985 to 2020. Across all three metropolitan areas, EL consistently constituted the largest share, though it exhibited a declining trend. Specifically, EL decreased from 82% to 72% in DFW, from 77% to 75% in CMA, and from 54% to 52% in CZT.

CL showed significant growth, particularly in DFW and CZT. In DFW, CL nearly doubled, rising from 9% in 1985 to 17% in 2020, representing the largest percentage increase (8%) among the three regions. CZT started with the lowest CL proportion at only 1% in 1985 but expanded to 6% by 2020. In contrast, CMA maintained a relatively stable but high proportion of CL, increasing slightly from 11% to 14%.

Regarding AL, distinct regional characteristics were observed. CZT maintained a high proportion of AL, ranking as the second-largest category, although it decreased from 45% to 42%. Conversely, AL accounted for the smallest proportion in both CMA and DFW. From 1985 to 2020, AL in CMA decreased marginally from 12% to 11%, while in DFW, it saw a slight increase from 9% to 10%.

3.2. Detection and Comparison of LUCC Intensity

3.2.1. Change Detection at Interval Level

Figure 6 illustrates the LUCC intensity and intensity deviation for three metropolitan areas across seven intervals. Regarding the magnitude of conversion intensity, the order is DFW > CZT > CMA. DFW’s change intensities increase and then decrease, with the maximum intensity occurring in the interval 2000–2005. CZT’s maximum intensity interval is 1990–1995. If we disregard the maximum intensity interval, we can observe that CZT’s change intensities also increase and then decrease. CMA’s change intensities decrease and then increase, with the maximum intensity interval being 1985–1990. In terms of intensity deviation, DFW exhibits a consistently fast characteristic, with the fastest interval being 2000–2005. Conversely, CMA exhibits a steadily slow characteristic. The LUCC intensity characteristics of CZT are unstable, with fast change periods in 1990–1995 and 2000–2015.

3.2.2. Change Detection at Category Level

Figure 7 reports category-level gain and loss intensities and their deviations across seven intervals for CZT, CMA, and DFW. CZT shows the highest gain intensity of CL, with gain deviations consistently active and loss deviations persistently dormant. This pattern indicates rapid and largely irreversible CL expansion in CZT relative to CMA and DFW. In CMA, both CL gains and losses are minimal, and their deviations hover near zero, implying small inflows and negligible outflows from 1985 to 2020 and thus only modest CL growth, as illustrated in Figure 5. DFW’s CL trajectory resembles CMA’s, but with higher gain intensity; as a result, the CL area in DFW has nearly doubled. Across all regions, CL losses remain small and dormant, while CL gains in CZT are steadily active, underscoring the stability of transitions into CL.

EL exhibits relatively low change intensities everywhere because it is the dominant land category. In CZT, EL gains were active and losses dormant from 1985 to 1995; thereafter, gains became dormant, and losses were active from 2000 to 2015, producing a slight net decline. CMA shows a similar temporal pattern but began with a larger EL base, so despite lower change intensities, it experienced a greater absolute reduction in EL. In DFW, both EL gains and losses are largely dormant, with losses exceeding gains, leading to a net decrease in EL area.

AL is the most volatile category. Gains and losses are both active across the study period, especially in DFW, where AL churn is sustained and its intensity surpasses that of any category in any region. In CZT, the 1990–1995 interval is notable for dormant gains and active losses, indicating a substantial AL reduction; in 2005–2010, the pattern reverses, with gains active and losses dormant, suggesting partial recovery of AL. CMA mirrors DFW’s concurrent gains and losses pattern but at lower intensity, reflecting more moderate turnover. Overall, the contrast in category-level intensities highlights rapid and stable CL expansion in CZT, subdued CL dynamics in CMA with modest growth, and strong AL volatility—most pronounced in DFW—against generally declining EL, especially where losses become intermittently active.

3.2.3. Change Detection at Transition Level

Figure 8 illustrates the intensity and intensity deviation of transitions from category $i$ to $j$. The analysis results of transition to CL are shown in Figure 8a,b. In all three regions, although the intensities of transition from EL and AL are similar, their intensity deviations are quite different. The transitions from EL to CL are represented in blue, signifying that the gain in CL has come at the expense of EL. On the contrary, transitions from AL to CL are colored red, signifying that CL gains preferentially drew from AL. The transition from EL to CL shows a stationary avoiding characteristic, while the transition from AL to CL shows a stationary targeting characteristic.

Figure 8c,d shows the transition to EL. The transition from CL and AL exhibits significant differences in intensity and intensity deviation. The transition intensity from CL to EL remains minimal across all regions, and such transitions have been steadily avoided. Conversely, the transition intensity from AL to EL demonstrates substantial magnitude. The intensity deviation indicates that the gain in EL steadily targeted AL across all regions. DFW has the largest intensity deviation, followed by CMA, and the lowest is CZT. There are two reasons for this result. First, the intensity deviation is calculated as the difference between $R_{rtij}$ and the overall transition intensity. When the transition intensity from a specific category to EL in Region A increases, the uniform intensity to EL increases correspondingly. Under such conditions, all $D_{rtij}$ values in Region A are amplified due to magnified internal disparities in transition intensity. Second, in regions where the proportion of AL is smaller, the same area of AL converted to EL will convert a larger intensity of itself. In the DFW region, the intensity of transition from AL to EL exhibits the biggest bubbles, which magnify the uniform intensity of transition to EL in DFW. Moreover, DFW’s initial proportion of AL is smaller than that of CMA and CZT. DFW experienced the greatest intensity of conversion despite having the smallest proportion of AL. Therefore, the intensity deviation of AL conversion to EL in DFW is significantly greater than in CMA and CZT.

Figure 8e,f show the transition to AL. CL still resists loss, as its bubbles are small and blue. EL serves as the main source for land transitioning into AL. It maintains the scale of AL, preventing a large-scale reduction. The intensity deviation of EL conversion to AL is slightly red, consistent with previous explanations. This is due to the large initial proportion of EL; the amount converted is small relative to this initial size. The scale of conversion varies among the three regions. CZT experienced a high-intensity conversion from EL to AL during the 2005–2010 period, while CMA saw a low point during the same period. In the DFW, the transition from EL to AL offset the large-scale loss of AL. Unlike CZT and CMA, which saw a decline in the proportion of AL, DFW’s proportion of AL even increased by 1% throughout the study period.

4. Discussion

4.1. Comparison and Summary of Results Between CZT, CMA and DFW

This study selected CZT, CMA, and DFW for research to compare the similarities and differences in their LUCC patterns. At the interval level, DFW exhibited the highest intensity of LUCC, followed by CZT, with CMA showing the lowest. Human activities are a primary driver of LUCC [40]. Over the past few decades, DFW has experienced significant population growth. Since 1985, the population of DFW has increased by 113% [41], compared to 73% for CZT and 7% for CMA. This substantial population increase has amplified the demand for housing, infrastructure, and food, thereby intensifying LUCC, particularly the expansion of CL [42].

The intensity of LUCC in CZT initially increased and then decreased, aligning with the overall trend of land change in China [43]. However, the period of greatest change intensity in CZT was 1990–1995, during which the population increased by only 2.5%. Therefore, population growth was not the primary driver of this dramatic change. Instead, institutional factors played a decisive role. After China implemented the Tax-Sharing System Reform in 1994, the central government centralized tax revenue, leaving local governments with a significant fiscal gap between shrinking revenue and rising expenditure responsibilities. To bridge this gap, local governments turned to ‘land finance’ [44]. They accelerated the conversion of rural land into construction land (CL) to generate substantial revenue from land use rights sales, utilizing their monopoly over land supply to capture value-added income [45]. Consequently, at the category level (Figure 7a), the gain intensity of CL in CZT became very active starting in 1990. This gain intensity gradually decreased only after 2010, when the central government introduced policies to restrict illegal land requisition and disorderly urban expansion [44].

CMA exhibited the lowest and declining intensity, attributed to its early saturation and Japan’s unique constraints. With a massive population concentrated on limited habitable land [46], CMA had already reached a high level of development by 1985. The only notable expansion occurred during the bubble economy (1985–1990). Subsequently, the bubble burst in the early 1990s caused land prices to plummet and transactions to freeze, sharply restricting further urban expansion [47]. The economy then entered a ‘Lost Decade’ of prolonged stagnation [48], which structurally suppressed demand for new development and led to a continued decrease in LUCC intensity [49].

DFW, CZT, and CMA are significant metropolitan areas situated within the world’s three leading GDP nations in 2020 [50]. Derived from three distinct yet comparable case studies, these LUCC trajectories offer a lens into the evolution of metropolitan areas. We distill these patterns into a conceptual framework comprising three characteristic stages of development: In the first stage, the extremely rapid expansion of CL is a common trend in the development of emerging metropolitan areas. This expansion is characterized by two key features: first, irregular land-use changes are prevalent during periods of weak regulation, and second, surrounding farmland is often the first to be encroached upon. The bubbles for CZT and DFW continuously increased before 2010 and 2000, respectively (Figure 6). Figure 8b demonstrates that the primary source of CL was AL. As mentioned earlier, CMA’s first phase was completed before 1985. These characteristics are also supported by evidence from numerous other studies [51,52,53,54]. In the second stage, when the expansion of CL reaches a certain level, the government strengthens regulation and controls expansion [21], and the LUCC intensity gradually decreases. The metropolitan area shifts from external expansion to internal renewal, with a focus on macro-level urban resilience [55,56] and micro-level people’s lives and community governance [57] CZT is currently at this stage, as can be seen in Figure 6, where its LUCC intensity is continuously decreasing. In the third stage, metropolitan area development seeks external breakthroughs to expand its influence [58]. CMA and DFW in this study have begun to transition and are in the early stages of the third stage. In Figure 6, their LUCC intensity is increasing again after 2015.

In addition to this conceptual framework, LUCC in metropolitan areas is also influenced by the economic development trajectories and policies of their respective countries. DFW, located in the Southern United States, is among the largest and fastest-growing metropolitan areas in the country. A substantial population influx—increasing by approximately 113% from 1985 to 2020 (Table 1)—has fueled economic growth and a real estate boom. Liberalized economic land transaction policies have further stimulated LUCC. CZT, located in central China, has developed rapidly with the proposal of the Rise of Central China strategy. Stimulated by land finance and the four trillion investment plan, the LUCC intensity of CZT increased rapidly [59]. Consequently, governments must strengthen regulatory oversight and enforcement when developing planning policies. Following 2010, stricter land policies from the central government limited the continued increase in CZT’s LUCC intensity. CMA is Japan’s third-largest metropolitan area. However, Japan’s high urbanization rate and stringent land control policies have constrained the intensity of its LUCC, preventing uncontrolled expansion [60]. In countries or regions facing land scarcity, authorities tend to implement stricter land policies and other measures, such as defining urban growth boundaries, to prevent disorderly urban sprawl [61]. Policymakers generally favor internal redevelopment over outward expansion. The policies driving LUCC in these metropolitan areas are shown in Figure 9.

4.2. Enhancement of Intensity Analysis

The CIA model builds upon the existing Intensity Analysis framework. Numerous case studies have demonstrated the effectiveness of Intensity Analysis [62,63]. We have enhanced the original Intensity Analysis in three key aspects. First, we incorporated the area symbol $r$ and intensity deviation metrics, which were absent in the original formulation. Second, we eliminated the calculation that converts overall change intensity into annual changes. When time intervals are consistent, removing this step simplifies the formula. Third, we introduced Intensity Deviation Maps to provide an intuitive visual representation of the results at both the category and transition levels. Furthermore, we removed two equations used to analyze the intensity of transitions from a losing category at the end of the time interval, as these have been shown to be inappropriate by scholars, including the original author [64,65].

Although the CIA model is a mathematical accounting framework that does not require statistical validation in the predictive sense (like machine learning models), its validity is demonstrated by its alignment with known historical events. For instance, the model correctly identified the active CL expansion in CZT during 2005–2010 (Figure 6), which coincides with the Rise of Central China strategy. Similarly, the dormant status of CMA accurately reflects Japan’s economic stagnation. This consistency between quantitative deviation metrics and qualitative historical contexts confirms the model’s reliability in characterizing macro-level LUCC patterns.

4.2.1. Effect of Intensity Deviation

Intensity deviation is valuable because it quantifies the difference between varying intensities and a uniform intensity, providing a clear measure of the magnitude of this difference. For example, Figure 8 illustrates the degree to which land class conversions are targeted or avoided. When comparing numerous data to a uniform intensity, deviation data is often more effective than direct comparisons [66,67]. Indeed, researchers have explored alternative methods to avoid extensive direct comparisons. Luo et al. [68] used avoidance and tendency judgments instead of direct comparison with uniform intensity. Similarly, Li et al. [69] used “tendency” and “suppression” to represent the relationship between change intensity and uniform intensity: change intensity greater than uniform intensity was classified as “tendency,” and vice versa as “suppression”. However, this binary classification, common in previous studies, overlooks measurable differences within each category. Our research refines this by further subdividing the results after the initial binary classification. Intensity deviation, unlike simple tendency/avoidance judgments, provides continuous numerical values. A larger intensity deviation indicates a stronger tendency, while a smaller deviation indicates a greater degree of avoidance. This use of intensity deviation allows for more discriminative expression and analysis of the results [38].

4.2.2. Comparison Between Different Expressions

The previous outcomes of the Intensity Analysis can be expressed in three ways, as illustrated in Figure 4 and Figure 10. Bidirectional bar graphs (Figure 10a) are the most widely used representation in practical applications of Intensity Analysis [70,71]. These graphs are divided into two sections: the left side indicates the magnitude of change, while the right side reflects the intensity of change. The left-side bars offer an intuitive visualization of change size, where the length of the bar directly corresponds to the area converted for each land category. However, bidirectional bar graphs have two key limitations. First, land categories with larger initial areas are likely to show larger converted areas simply because they have more area available to be converted [38]. For example, consider grassland and water bodies with initial areas of 500 km² and 50 km², respectively. If both experience a conversion of 50 km², the bidirectional bar graph would show bars of equal length, potentially masking the fact that the water body experienced a complete (100%) conversion—a devastating impact on the aquatic ecosystem. This critical difference can be easily overlooked [72,73]. Second, direct size comparisons across multiple regions are often misleading due to variations in total area [31]. Under a unified coordinate system, bars representing changes in larger regions will naturally appear longer than those in smaller regions, regardless of the relative intensity of change. Therefore, bidirectional bar graphs are not a scientifically objective method for expressing results of the Intensity Analysis in multi-region comparative studies.

To address the shortcomings of bidirectional bar graphs, Li et al. [69] proposed a simplified approach, removing the bar on the left and retaining only the bar on the right. This right-side bar visualizes the key result of the Intensity Analysis: the change intensity for each land category. A dotted line superimposed on the bar chart represents a baseline of uniform intensity. Researchers then determine whether a land category’s change intensity is significant by comparing the length of its bar to this uniform intensity line [74]. This method is effective when analyzing data with few time intervals and land categories. However, when managing numerous time intervals and land classifications, the bars become densely packed and small, making it difficult to extract meaningful information [75].

To address the challenge of effectively representing Intensity Analysis results when dealing with numerous time intervals and land categories, Yang et al. [76] employed a confusion table. This table presents a binary classification (Figure 10b), using two colors to distinguish between change intensities that are either greater than or less than the uniform intensity. In certain contexts [68,77], this binary classification graph may offer advantages over the bidirectional bar graph. However, as previously discussed, this binary classification inherently sacrifices detail, obscuring the nuanced differences within the results. The relationship between change intensity and uniform intensity is not simply a binary distinction.

Figure 10. Expression methods of Intensity Analysis results: (a) bidirectional bar graph and (b) binary classification graph [25,76].

Figure 10. Expression methods of Intensity Analysis results: (a) bidirectional bar graph and (b) binary classification graph [25,76].

Our Intensity Deviation Maps (Figure 4) use bubbles instead of bars to compare against uniform intensity. The size and color of these bubbles represent the core results of the Intensity Analysis: the change intensity and its relationship to the uniform intensity. Red bubbles indicate intensity greater than uniform, while blue bubbles indicate intensity less than uniform. Furthermore, saturation levels represent the magnitude of deviation from uniform intensity; higher saturation indicates a greater difference. By comparing the color saturation, the tendency and avoidance degree of land category conversion can be intuitively distinguished [78].

4.3. Influence of Time Interval

Temporal resolution critically affects LUCC detection. Our analysis identified an anomalous intensity spike in CZT (1990–1995) that would be smoothed out in decadal studies [38]. Coarse time intervals risk masking such short-term but high-impact policy shocks [79]. While finer intervals capture detailed dynamics, they generate an overwhelming volume of data, posing a visualization challenge.

Traditional graphical methods, such as stacked bar charts, become cluttered when dealing with multiple regions and intervals [76]. Previous studies often compromised by reducing the number of intervals or land categories [63,80]. In contrast, the Intensity Deviation Maps proposed in this study effectively compress multidimensional information into a single view, including intensity, deviation, and stationarity. This approach resolves the trade-off between temporal resolution and visual interpretability, making it particularly suitable for long-term, multi-regional comparative research supported by annual datasets [35].

4.4. Implications for Extrapolation

Since the introduction of Intensity Analysis by Aldwaik and Pontius [25], it has been widely applied in numerous case studies [30,72,81]. Unlike traditional methods that focus on magnitudes, Intensity Analysis emphasizes the intensity of LUCC [25]. A key advantage of Intensity Analysis is its ability to enable cross-regional comparisons of LUCC, even when land areas differ [29,78], a capability not initially anticipated by its creators. Several studies have employed Intensity Analysis in regional comparative research [29,65,74,82]. However, in their applications, they merely apply Intensity Analysis to two distinct regions, without providing a comprehensive explanation at the level of mathematical formulas. Clear mathematical formulations are crucial for standardizing scientific methodologies and providing a common language for interdisciplinary research [20]. The CIA model summarizes the formulaic expressions from previous literature [38,64,78] and presents a systematic and clear mathematical formulation. For regional comparisons using intensity analysis, later researchers can now utilize the enhanced CIA model, which offers an interpretable and complete mathematical formula.

Unlike previous studies that compared parameters in isolation [31,83,84], the CIA model provides a comprehensive comparison of LUCC using a top-down, three-level system. This model expands the methodological framework for comparative analysis and strengthens the theoretical foundation for applying Intensity Analysis across different regions. Furthermore, the integrated approach of intensity deviation and Intensity Deviation Maps is equally applicable to case studies within single regions [38]. Intensity deviation is central to the CIA model because it unifies the intensity of change across different regions and land categories with a uniform intensity. The use of Intensity Deviation Maps significantly compresses information density, highlighting the LUCC information that decision-makers wish to observe. Researchers can leverage the strengths of both intensity deviation and Intensity Deviation Maps to advance their respective fields.

4.5. Limitations and Possible Improvement

Our study has several limitations that we aim to address in future research. First, this study simplified land use into three broad categories (CL, EL, and AL) to facilitate cross-regional comparison. While this aggregation improves the overall accuracy of the land-use data by reducing spectral confusion between similar sub-classes, it inevitably masks internal heterogeneity and specific ecological functions. For instance, the differing ecological values of forests, wetlands, and grasslands are conflated within the category of EL. Consequently, this coarse resolution may lead to specific biases, such as an underestimation of ecological fragmentation or an oversimplification of agricultural dynamics in regions with diverse cropping systems. However, given that the primary objective of this study is to identify macro-level developmental stages of metropolitan areas, this framework remains robust for capturing the dominant trends of land-use transformation. Future research should consider more granular classifications to capture these nuanced ecological processes.

Second, the CIA model benefits from a larger number of regional comparisons to identify areas with more pronounced LUCC. Expanding the number of regions would provide a richer dataset. With sufficient data, it would become possible to classify different metropolitan areas into distinct development stages. For instance, Peng and Quan [81] conducted the Intensity Analysis of 114 cities in the Mid-spine Belt of Beautiful China and categorized urban development into four stages. Mapping metropolitan areas to their corresponding development stage would have significant implications for policy guidance.

Third, regarding the robustness of the CIA model, the uniform intensity is dependent on the composition of the study regions. Adding a significantly larger or faster-changing region to the dataset would shift the uniform intensity, potentially altering the Fast/Slow classification of other regions. This is an inherent feature of comparative analysis—intensity is relative to the group being compared. Future studies should conduct sensitivity analyses by varying the set of regions to distinguish between local idiosyncrasies and global trends.

5. Conclusions

Using a CIA framework coupled with Intensity Deviation Maps, we systematically compared LUCC in three metropolitan areas—CZT in China, CMA in Japan, and DFW in the USA—from 1985 to 2020 at interval, category, and transition levels. Synthesizing these patterns, we proposed a conceptual framework of metropolitan evolution characterizing three stages: rapid expansion, internal renewal, and external breakthrough.

Methodologically, the CIA model, coupled with Intensity Deviation Maps, advances cross-regional research by standardizing change intensities against a uniform baseline. This approach effectively resolves the challenge of comparing regions with disparate sizes and visualizing complex transition matrices intuitively.

In terms of policy, the identified stages offer actionable guidance for sustainable planning. For emerging metropolises in the expansion stage, strict growth boundaries are critical to prevent disorderly sprawl. Mature regions in the renewal stage should prioritize stock land revitalization and ecological restoration. Finally, regions seeking external breakthroughs must balance renewed population growth with conservation. Recognizing these stages enables policymakers to transition from quantity-oriented expansion to quality-oriented sustainability strategies.