Probability-Based Framework for Applying the Ecological Area Ratio: Insights from South Korea’s New Towns

Abstract

As urbanization intensifies, the ecological area ratio (EAR) has become an essential tool for assessing ecological performance in urban development projects. However, conventional EAR systems remain largely prescriptive and surface-oriented, lacking adaptability to diverse planning contexts. This study proposes a probability-based EAR reference table developed from empirical data collected across six representative large-scale urban development districts. EAR values were statistically analyzed for 16 land-use types to construct a reference table using mean and quartile indicators. The table was then applied to seven newly planned towns to evaluate its predictive utility. The results showed that predicted EAR values closely aligned with institutional targets and revealed meaningful internal variation depending on land-use composition. Green space and parks showed the highest ecological contributions, while multi-family housing, despite moderate unit-area performance, played a key stabilizing role due to its large spatial footprint. Correlation analyses further demonstrated that EAR composition varied across housing supply types, shaped by differing regulatory frameworks and design priorities. By transitioning EAR from a uniform ratio to a data-driven, probabilistic guideline, this study offers both a practical estimation tool and a strategic planning aid. The findings provide actionable insights for more adaptive, equitable, and ecologically robust urban development practices.

1. Introduction

The intensification of climate change and the acceleration of urbanization have precipitated various environmental challenges in urban areas, including the urban heat island effect, the disruption of the hydrological cycle, and the loss of biodiversity [1,2,3]. The expansion of cities increases the area of impervious surfaces at the expense of vegetated land, which results in decreased infiltration, increased surface runoff, and weakened ecological resilience [4]. To address these problems, various tools have been suggested to help urban designers restore ecological functions and support natural cycles. In South Korea, the ecological area ratio (EAR) has become an important measure for assessing the portion of land within a development site that enhances ecological performance [5]. EAR comprehensively considers ecological value and natural circulation functions, which are difficult for conventional indicators such as the building coverage ratio or floor area ratio to account for [6].

The ecological area ratio (EAR) was officially introduced in 2006 by the Korean Ministry of Environment as part of the environmental impact assessment system for large-scale development projects. Since its inception, the regulation has undergone multiple revisions (e.g., 2012, 2018, 2022) to expand its scope and refine the weighting criteria for different surface types. Over nearly two decades, EAR has been applied to a wide range of urban contexts including residential complexes, public facilities, and urban renewal districts, serving as both a regulatory requirement and a planning guideline. Empirical studies have documented its role in guiding ecological compensation, promoting minimum green coverage standards, and evaluating post-construction ecological performance [7]. However, despite this history of application, standardized methods for planning-stage estimation and cross-project comparability remain limited. This challenge is not unique to EAR: in recent decades, multiple cities worldwide have also developed ratio-based greening metrics to overcome similar limitations.

In recent decades, various greening metrics have been adopted to quantitatively and qualitatively assess ecologically functional surfaces for environmental sustainability [8,9]. Multiple cities in various countries have institutionalized ratio-based systems such as the biotope area factor (BFF) in Berlin, Germany; green space factor (GSF) in Malmö, Sweden; landscaping for urban spaces and high-rises (LUSH) in Singapore; green area ratio (GAR) in Washington DC, USA; and urban greening factor (UGF) in Greater London and 10 other borough-level jurisdictions in the UK. As presented in

Table 1. Urban greening metrics around the world.

Country	Policy Name	Application Scope	Calculation Method	Key Features	Related Studies
Germany	Biotope Area Factor (1990)	Selected redevelopment areas covering approximately 5% of Berlin	Sum of (area by land-use type × ecological weight) divided by total site area	First ecological land-use planning tool in Germany; target values vary by building type and development intensity	[12,13,14]
Sweden	Green Space Factor (2001)	City of Malmö, including Bo01 and Flagghusen districts (approx. 160 km2)	Sum of (surface area × ecological weight) divided by total lot area	Incorporates Green Point system requiring at least 10 out of 35 ecological items; implemented at the municipal level	[15,16]
Singapore	Landscaping for Urban Spaces and High-Rises (2009)	Nationwide application, differentiated by strategic and non-strategic zones	Green Plot Ratio: total green area divided by total development site area	Regulations vary by gross plot ratio and softscape proportion; stricter criteria in high-activity zones	[17,18]
United States	Green Area Ratio (2013)	City of Washington D.C., across all zoning classifications	Sum of (surface area × ecological weight plus bonus for native vegetation) divided by lot area	Zoning-specific minimum targets; native species receive additional weighting; zoning map available via GIS platform	[19,20]
United Kingdom	Urban Greening Factor (2015)	Greater London Authority and 10 additional borough-level jurisdictions in England	Sum of (surface type area × ecological weight) divided by total lot area	Minimum target values by land-use type (e.g., residential, commercial, industrial); tailored to local environmental conditions	[21,22,23]
Republic of Korea	Ecological Area Ratio (2006)	All national development projects subject to Strategic or Environmental impact assessment	Ratio of ecologically functional soil area to total development site area; weights assigned by surface type	Three-phase system (current, target, planned ratios); target ratio varies by project type (e.g., urban ≥30–40%, tourism ≥60%); allows up to 50% reduction based on site conditions	[24,25]

, these metrics differ in scope and calculation but reflect a common shift away from a rigid area-based approach toward a more flexible, ecologically meaningful, and context-sensitive approach. For instance, BFF is applied selectively in dense redevelopment zones where targets are differentiated by building type and development intensity. GSF combines surface-based weighting with a green point system that encourages biodiversity in specific districts. LUSH uses the green plot ratio (GnPR) to measure the total vegetated area for which criteria vary according to the urban activity and zoning status. GAR integrates bonuses for native species and provides zoning-specific thresholds that are supported by a geographic information system (GIS)-based platform. UGF assigns ecological weights to different surface types and allows for local customization. These metrics reflect a clear trend toward flexible and multidimensional planning tools that incorporate ecological function, spatial form, and social benefit. In contrast, South Korea’s EAR is largely quantitative and surface-oriented and focuses primarily on horizontal green coverage. Despite being established in 2006 and revised multiple times, EAR continues to rely on fixed weights and uniform target ratios for each project type with limited adaptability to site-specific constraints. Empirical studies have pointed out that only a narrow set of land-cover types account for most of the value of EAR, particularly natural soil and artificial soil, while other types make negligible contributions [10,11]. Moreover, there is a lack of standardized data or predictive tools to support planning-stage estimation or post-construction validation. These limitations highlight the need for a more integrated and performance-oriented approach aligned with global green planning practices.

To address this, we selected six urban development districts in the Seoul Metropolitan Area that capture diverse geographical settings (eastern, western, and southern sectors), development stages (initiated between the 2000s and 2010s), and ecological contexts (residential, commercial, and mixed-use functions). This representativeness was intended to ensure that the probability-based EAR reference values are grounded in spatial and ecological heterogeneity. Based on these data, this study aimed to enhance the practical applicability of EAR by developing statistically grounded guidelines specific to land-use and land-cover types. The probability-based EAR reference table was then applied to seven new towns to evaluate its predictive utility. A focused correlation analysis was conducted using multifamily housing blocks to examine how individual land-cover types contribute to the overall EAR. This study makes three core contributions. First, it transitions EAR from a uniform recommendation to a probabilistic and data-driven standard that reflects urban land-use and land-cover types and development realities. Second, it proposes a process for estimating EAR at the planning stage even in the absence of finalized design details. Third, by clarifying which land-cover types substantially contribute to EAR, it provides guidance for both policy refinement and sustainable urban design practices. It is expected to support more adaptable and effective urban greening strategies and contribute to South Korea’s broader goals of ecological resilience and climate adaptation.

2. Materials and Methods

Figure 1 shows the overall workflow of the study, which included the quantitative analysis of empirical data from urban development districts, the development of a probability-based EAR reference table, and the application of the table to planning the land use of new towns.

2.1. Terminology

In South Korea, an “urban development district” is an area established under the Urban Development Act that is designated for the systematic development of infrastructure and various land uses including residential, commercial, public, and green spaces [26]. Urban development districts are typically developed under public-sector leadership and are characterized by comprehensive planning and phased land development. The urban development districts selected for empirical analysis were all large-scale state- or city-led projects. A “new town” is a type of large-scale and government-planned urban development district that was created to address housing demand and urban congestion in major metropolitan areas. New towns are typically categorized into three phases based on their planning and implementation period: phase 1 was established in 1989, phase 2 was established in 2003, and phase 3 was established in 2018. This study applied the probability-based EAR reference table to phase-2 and phase-3 new towns. In this study, land use refers to the functional designation of space within an urban development district (e.g., residential, commercial, public facilities, and green space), whereas land cover denotes the physical surface characteristics present within each land-use parcel (e.g., natural soil greening, rooftop greening, permeable pavements). The probability-based EAR reference values were derived by linking 16 land-use categories with their observed distribution of land-cover types.

2.2. Data Collection

Table 2. Details of the six reference sites selected for empirical EAR data analysis.

Category	Hanam Misa	Hwaseong Dongtan	Incheon Geomdan	Suwon Gwanggyo	Wirye NewTown	Yangju Okjeong
Administrative Region	Gyeonggi-do		Incheon Metropolitan City	Gyeonggi-do
Area (m2)	874,883	1,123,023	1,031,659	932,966	742,925	709,904
Population	92,501	285,878	187,076	77,500	110,719	106,351
Project Period	3 June 2009 – 30 June 2022	11 July 2008 – 31 December 2014	June 2007 – December 2026	December 2005 – December 2012	5 August 2008 – 31 December 2024	5 August 2008 – 31 December 2024

lists the six completed urban development districts located within the Seoul Metropolitan Area that were selected as reference sites: Hanam Misa, Hwaseong Dongtan, Incheon Gemdan, Suwon Gwanggyo, Wirye New Town, and Yangju Okjeong. Seven new towns were selected as application sites of the probability-based EAR reference table: Gimpo Hangang, Paju Unjeong, and Yangju Hoecheon are phase-2 new towns while Goyang Changneung, Hanam Gyosan, Namyangju Wangsuk, and Namyangju Wangsuk2 are phase-3 new towns. Figure 2 shows the spatial distribution and classification of the reference and application sites. Relevant planning documents were obtained from the Korea Land and Housing Corporation (LH), which included master plans, landscaping and pavement layouts, and EAR-specific calculation drawings.

The land cover was classified according to Ministry of Environment Notice No. 2022-61, which defines 19 types [27] including natural ground, artificial ground (categorized by soil depth), rooftop greening (classified by soil thickness), permeable pavements, wall greening, and areas on detention and infiltration facilities. Each category was assigned a weight in the range of 0.1–1.0 in accordance with its ecological contribution. Based on these weights, block-level EAR values were calculated using plan-based measurements. To improve the accuracy of classification derived from computer-aided design drawings, additional verification was conducted using Google Earth imagery, KakaoMap Skyview aerial imagery (50 cm, 2023), and photos taken during a field survey. When features were obscured by shadows or unavailable, earlier imagery was consulted. Parcel-level land-cover areas were delineated in AutoCAD 2025 based on visually identifiable features, with an effective minimal mapping unit of approximately 1.0 × 1.0 m. All imagery sources were complemented with official planning drawings and EAR calculation documents from the LH, which corresponded to the designated development periods of the six reference sites (2005–2022), thereby ensuring temporal consistency. In cases where a land-cover type could serve multiple functions (e.g., rooftop greening providing both stormwater retention and habitat value), the classification followed MOE Notice No. 2022-61, which assigns a single ecological weight to each standardized category. This ensured consistency and avoided double-counting, while implicitly capturing multifunctionality through the assigned weights.

This specification highlights the spatial limits of the analysis and ensures comparability across reference and application sites. The selected reference sites were all large-scale urban development districts, typically exceeding several hundred thousand square meters, which was important to ensure sufficient diversity of land-use and land-cover types for generating statistically meaningful EAR distributions. While the approach could in principle be applied to smaller sites (e.g., ~1000 m²), the reliability of the probability-based reference values would be limited due to insufficient heterogeneity. Therefore, similar scale across reference sites was not an explicit requirement but a practical condition to ensure representativeness. All data were organized into a geospatial database, as visualized in Figure 3. Land-use classification standards, data collection timing, and source documents were aligned across reference and application sites, ensuring comparability. Although some ecological variability naturally arises from site-specific contexts, the use of probability-based EAR values derived from multiple heterogeneous districts provided robustness against localized bias and improved the generalizability of the results.

In South Korea, the EAR is administered under the national guideline issued by the Ministry of Environment [28]. Since the 2016 second revision, a district-level target of 40% has been specified for development projects outside existing urban cores; accordingly, we treat 40% as a mean-level compliance threshold when interpreting results in this study. Where a different threshold is mandated by a jurisdiction, the framework applies by substituting that value.

2.3. Development of the Probability-Based EAR Reference Table

The probability-based EAR reference table was constructed according to the statistical distribution characteristics of EAR values by land-use type based on data collected from six reference sites. For each block $b$ and land-cover class $k$, the ecological contribution was defined as the product of the surface area and the ecological weight:

$$C_{k,b}=A_{k,b}\times w_k$$

where $A_{k,b}$ is the mapped area and $w_k$ is the ecological weight of class $k$. The block-level EAR was then calculated as

$${EAR}_b={\displaystyle \frac{\sum _k{C}_{k,b}}{T_b}}$$

with $T_b$ being the total block area. To further examine the structure of contributions, we also derived the composition ratio of each class withing a block:

$$r_{k,b}={\displaystyle \frac{C_{k,b}}{\sum _j{C}_{j,b}}}$$

where $j$ is a dummy index running over all land-cover classes withing block $b$. These normalized values were then converted into EAR composition ratios (%) for subsequent analysis.

To ensure analytical reliability, the dataset was refined by removing outliers using the interquartile range (IQR) method (values below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR) and handling missing values accordingly [29,30]. The cleaned data were then structured for statistical summarization using empirical means, medians, and quartiles rather than assuming a specific distributional form.

Table 3. Ecological weights of land-cover types [31].

Coverage Type		Weight
Area on Detention and Infiltration Facility		0.30
Artificial Ground Greening	10 cm ≤ Soil depth < 40 cm	0.50
	40 cm ≤ Soil depth < 90 cm	0.60
	Soil depth ≥ 90 cm	0.70
Natural Soil Greening		1.00
Partial Pavement	Artificial ground	0.50
	Natural ground	0.50
Permeable Joint Pavement	Artificial ground	0.14
	Natural ground	0.23
Rooftop Greening	10 cm ≤ Soil depth < 20 cm	0.50
	20 cm ≤ Soil depth < 30 cm	0.60
	Soil depth ≥ 30 cm	0.70
Total Permeable Pavement	Grade 1 Permeability, Artificial Ground	0.28
	Grade 1 Permeability, Natural Ground	0.40
	Grade 2 Permeability, Artificial Ground	0.21
	Grade 2 Permeability, Natural Ground	0.30
Wall Greening		0.40
Water Space	Impermeable	0.70
	Permeable	1.00

summarizes the ecological weights for each land-cover type, which reflect the ecological performance of the surface as defined by MOE Notice No. 2022-61 [31]. For example, natural soil greening and permeable water space were both assigned a full ecological weight of 1.0 whereas artificial ground greening with a shallow soil depth or wall greening was assigned a lower ecological weight to reflect its reduced ecological functionality. The weights span from 0.1 for low-functionality surfaces (e.g., low-depth artificial greening, permeable joint pavement) to 1.0 for fully functional natural elements (e.g., natural soil, permeable water space). Most values are clustered in the mid-range (0.3–0.7), reflecting partially functional surfaces such as artificial ground greening, rooftop greening, and permeable pavements. This distribution highlights the gradation of ecological performance across surface types rather than a binary classification of green versus non-green. Although the majority of weights are 0.50 or larger, their impact on modeling is moderated by the varying shares of land-cover types in actual projects. In other words, higher weight values do not uniformly inflate EAR outcomes, since block-level EAR depends on the proportional composition of land-cover types observed in diverse districts.

EAR is derived from standardized planning drawings and tends to be calculated under consistent regulatory and institutional conditions. While the exact distribution of EAR values may vary across sites, their empirical patterns can be effectively summarized using descriptive statistics. Accordingly, mean, median, and quartile values were calculated for each land-use type. This approach provides a robust and interpretable summary of the observed data distributions without requiring strict normality assumptions. The mean indicates the typical EAR level, while the IQR reflects the expected variability across different sites. This approach was adopted because mean and quartile indicators provide a robust and easily interpretable summary of empirical distributions without relying on strict normality assumptions. While the mean captures the central tendency, quartiles represent variability in a way that is statistically sound and practically useful for planning applications. The resulting probability-based EAR reference table contained statistical values for each land-use type centered on the mean and quartiles. It was designed to allow preliminary estimation of the overall EAR during the early planning stages using only the projected areas for each land-use type. Unlike traditional approaches that apply a uniform EAR target, this table enables the establishment of more reliable EAR targets based on empirical data from actual sites.

2.4. Validation

The probability-based EAR reference table was applied to seven application sites to evaluate its practical utility: three phase-2 new towns (Gimpo Hangang, Paju Unjeong, and Yangju Hoecheon) and four phase-3 new towns (Goyang Changneung, Hanam Gyosan, Namyangju Wangsuk, and Namyangju Wangsuk2). For each application site, parcel-level land-use types and their respective areas were extracted from district-level development plans. The corresponding EAR values from the table were applied to estimate the overall EAR for each application site. The estimated EAR values were then compared against the target EAR values specified in the planning documentation for each application site. Application sites were selected a priori using two objective criteria: (i) program stage (phase-2 or phase-3 new towns) and (ii) public availability of parcel-level land-use schedules and official EAR targets from LH at the time of data collection. These criteria yielded seven towns (three phase-2, four phase-3); no sites were excluded based on expected performance, and none of the application sites were used to derive the reference table, ensuring out-of-sample validation. The degree of agreement between the estimated and target values served as an indirect validation of the predictive reliability of the probability-based EAR reference table, demonstrating its potential applicability in preliminary planning stages. To minimize potential bias toward towns that may appear more positively impacted by the probability-based approach, the reference table was derived from multiple heterogeneous districts rather than a single case. This ensured that the applied EAR values were grounded in a broad spectrum of urban and ecological contexts, thereby minimizing the likelihood of systematically favoring any particular application site.

3. Results and Discussion

3.1. Probability-Based EAR Reference Table

Table 4. Probability-based EAR reference table.

Land-Use Type	Mean	Q1 (25%)	Q2 (50%)	Q3 (75%)
Commercial Facility	0.174	0.117	0.146	0.178
Green Space	0.971	0.999	1.000	1.000
Industrial Facility	0.185	0.139	0.170	0.214
Multifamily Housing	0.335	0.302	0.338	0.399
Office Facility	0.169	0.115	0.141	0.219
Park (landscaped)	0.748	0.614	0.735	0.843
Park (preserved)	0.848	0.723	0.799	0.882
Plaza	0.452	0.144	0.262	0.485
Public Land	0.689	0.328	0.615	0.808
Public, Cultural, and Sports Facility	0.260	0.156	0.226	0.294
Religious Facility	0.206	0.065	0.109	0.153
Road	0.099	0.057	0.192	0.339
School	0.247	0.204	0.230	0.265
Single-family Housing	0.092	0.034	0.049	0.098
Transportation Facility	0.061	0.048	0.094	0.130
Utility and Supply Facility	0.241	0.064	0.106	0.183

presents the probability-based EAR reference table, which contains the mean, Q1 (25th percentile), Q2 (median), and Q3 (75th percentile) EAR values for each of the 16 land-use types. Clear stratification can be observed among the land-use types in terms of ecological performance. Land-use types associated with open space and vegetation had the highest EAR values. Green space exhibited a mean EAR value of 0.97 with all quartiles converging near 1.0, which indicates a consistently high ecological function. Preserved parks and landscaped parks, which primarily had land covers of natural ground or artificial ground with soil depths exceeding 90 cm, followed with mean EAR values of 0.85 and 0.75, respectively. A second group of land-use types exhibited moderate EAR values in the range of 0.20–0.45, which included multifamily housing (0.34), schools (0.25), public, cultural, and sports facilities (0.26), religious facilities (0.21), and utility and supply areas (0.24). These land-use types typically relied on a combination of greening strategies such as rooftop greening and permeable pavement that contributed to their EAR values. In contrast, lower EAR values were observed for land-use types such as industrial (0.19), commercial (0.17), office (0.17), and transportation facilities (0.06) as well as single-family housing (0.09) and roads (0.10). These land-use types primarily comprised impervious surfaces, which had a low ecological weight.

Beyond the expected high values for green space and parks, two observations are noteworthy. First, several non-green categories (for example, office and utility and supply) show moderate EAR values (approximately 0.17 to 0.24), which likely reflect engineered greening such as rooftop greening and permeable pavements. Second, while green space has quartiles clustered near 1.0, moderate categories exhibit wider IQR, suggesting that green space contributes primarily to stability whereas non-green categories account for much of the variability across sites.

The probability-based EAR reference table provides statistically grounded benchmarks that reflect the typical and expected ranges observed in actual sites. It offers a practical tool for estimating EAR in the absence of empirical data and for supporting land-use planning and assessment. In planning applications, the median (Q2) can be used as a baseline per-type target, with Q3 for ambitious programs and Q1 where constraints are severe; a site-level target can then be approximated by the proportion-weighted sum ${\sum }_k{s}_k{T}_k$, where $s_k$ is the land-use proportion and $T_k$ is the selected percentile for type $k$. The IQR (Q3−Q1) is reported as an indicator of expected stability and can be used to flag mixes that may require design safeguards.

3.2. Application of the Probability-Based EAR Reference Table

3.2.1. Phase-2 New Towns

Figure 4 illustrates the actual land-use compositions of the three phase-2 new towns based on district-level planning data. Across all three application sites, multifamily housing emerged as the most dominant land-use type with an area ratio of 28% in Gimpo Hangang and Yangju Hoecheon and 31% in Paju Unjeong. Green space was also prominent with an area ratio of 17–21%, which reflects efforts to incorporate open space and ecological buffers into the plans for the new towns. Single-family housing showed site-specific variation with area ratios of up to 19% in Gimpo Hangang but only 13% in Paju Unjeong. Institutional and community service land-use types, which included public, cultural, and sports facilities as well as schools and utilities, had a modest combined area ratio of less than 15% at each application site. Landscaped and preserved parks were present at all three application sites with preserved parks slightly more prominent in Gimpo Hangang. In contrast, land-use types with high-density urban functions such as commercial, office, and industrial facilities each had an area ratio of less than 8%, which indicates that the development of the phase-2 new towns was predominantly oriented toward residential and open spaces. The land-use compositions reflect a balanced mix of residential density and ecological planning with some site-level differences in the proportion of green infrastructure and public amenities.

Figure 5 presents the predicted EAR distributions for the three phase-2 new towns, which were obtained by applying the probability-based EAR reference table to the actual land-use composition of each application site. All three application sites exhibited EAR distributions centered near the planned target EAR of 40%. Yangju Hoecheon showed the narrowest IQR, which indicates relatively uniform EAR contributions across its parcels. Gimpo Hangang showed the widest IQR, which suggests a more heterogeneous mix of parcel-level EAR values resulting from its higher share of single-family housing and preserved parks. This reading is consistent with Section 3.3, Section 3.4 and Section 3.5, which describe how site-level mean and IQR vary primarily with the spread of land-use proportions and the presence of cover types with inherently higher variance, rather than with the phase designation per se. In terms of the mean EAR (green dot), all three application sites achieved values close to the planned target EAR of 40% with differences of less than ±2%. These results suggest that the actual land-use compositions dominated by multifamily housing and green space with relatively low shares of impermeable urban land-use types are generally compatible with the target EAR. Applying the probability-based EAR reference table to actual land-use compositions yields predicted EAR distributions that not only met planning targets but also revealed meaningful internal variations based on parcel-level land-use planning strategies. Accordingly, we interpret IQR as an indicator of stability and use it to flag mixes that may require design safeguards, while detailed drivers are discussed in Section 3.3, Section 3.4 and Section 3.5.

3.2.2. Phase-3 New Towns

Figure 6 presents the actual land-use compositions of the four phase-3 new towns based on district-level planning data. In contrast to the phase-2 new towns, the phase-3 new towns showed a stronger emphasis on parks and open spaces. Landscaped parks accounted for the largest single land-use type at all four application sites with area ratios of 22–29%, which was followed by multifamily housing at 15–22% and green space at 15–17%. Institutional and service land-use types such as schools, public facilities, and utilities had area ratios of 7–11% at the four application sites while single-family housing remained relatively limited at 2–4%. Land-use types with urban functions such as commercial, office, and industrial facilities were present but consistently minor with area ratios of less than 10% each. These land-use compositions indicate a policy shift toward maximizing areas with ecological and public benefits. The consistent allocation of land to green space and public parks alongside compact residential planning suggests a strategic approach to improving urban environmental quality and ecological resilience at the district level.

Figure 7 shows the predicted EAR distributions of the four phase-3 new towns according to the probability-based EAR reference table. Substantial differences can be observed across the four application sites. Goyang Changneung exhibited the highest overall EAR with a mean above 60% and a narrow IQR of approximately 55–66%, which can be attributed to its high area ratios of green space and preserved parks. Similarly, Namyangju Wangsuk2 maintained EAR values well above the planned target EAR of 40% with a mean of approximately 50% and a relatively balanced distribution. In contrast, Namyangju Wangsuk and Hanam Gyosan showed EAR distributions closer to the target EAR. In particular, Namyangju Wangsuk had a wide IQR of roughly 35–50%, which indicated a more mixed land-use composition and lower ecological weights for certain parcels. Hanam Gyosan had a more compact IQR but also hovered near the target EAR of 40%, which suggests a moderate ecological performance. These results highlight the sensitivity of EAR to site-specific land-use compositions. While the probability-based EAR reference table provided a consistent methodological framework, the differences in predicted EAR distributions clearly reflect the influence of land-use type allocation, particularly regarding the area ratios of open space and high-density residential areas.

3.3. Contributions of Land-Use Types to the EAR Distribution

The scatter plot in Figure 8 illustrates the contribution of each land-use type to the overall EAR of each application site. As expected, land-use types with high ecological weights such as green spaces and parks made the largest contributions to the overall EAR. Despite their moderate area ratios, these land-use types exhibited a substantial ecological impact due to their high EAR coefficients (mean: 0.75–1.0). In contrast, while public land had a relatively high EAR coefficient (mean: 0.69), its contribution was minimal because of the very limited area ratio. While multifamily housing had a moderate EAR coefficient (mean: 0.34), it had the largest area ratio in most application sites, so it consistently made a relatively high contribution of 7–10% to the overall EAR. Furthermore, its tightly clustered points suggest that this land-use type not only contributes significantly to the mean EAR but also functions as a stabilizing element that reduces inter-site variations in land-use composition. Conversely, land-use types with low ecological weights such as roads, transportation facilities, and commercial facilities consistently made negligible contributions to the overall EAR of all sites, which confirmed their limited ecological function. These findings underscore the importance of evaluating both the ecological weight and spatial distribution of land-use types when assessing their actual contributions to EAR.

3.4. Correlation Between Land-Use Type and EAR Distributions

To investigate the factors driving the differences in ecological performance among application sites, a correlation analysis was conducted between the area ratio of each land-use type and two EAR indicators: the mean EAR and IQR. While the mean EAR reflects the overall level of ecological contribution, the IQR measures the internal variability among parcel-level EAR values. Unlike the variance, which is sensitive to outliers and skewed distributions, IQR provides a more robust and interpretable metric for evaluating consistency in ecological design, which is particularly useful in the context of uneven land distribution when planning new towns [32,33]. A positive correlation with the mean EAR suggests that a given land-use type improves the overall ecological performance with its presence while a negative correlation with the IQR indicates that the land-use type supports greater internal consistency across parcels. These two directions of correlation offer complementary insights into performance and stability rather than a dichotomy of good versus bad [34,35]. To visualize these relationships, Figure 9 presents the Pearson correlation coefficients between the area ratios of the land-use types and the two EAR indicators across the seven application sites [36,37].

The landscaped and preserved parks both showed strong positive correlations with the mean EAR (r = 0.75 and 0.61, respectively), and with the IQR (r > 0.63). These land-use types have land-cover types with high ecological weights such as natural ground or deep artificial soil, so they elevate the overall EAR when present in significant amounts. At the same time, their area ratios varied substantially across application sites depending on the design strategy, which increases the IQR. Notably, the contribution of landscaped parks to the mean EAR rose from 5% in phase-2 new towns to 25% in phase-3 new towns, which reflects a shift toward park-centered ecological planning between phases. Likewise, utility and supply facilities showed a marked increase in their contribution to the mean EAR from 0.08% in phase-2 new towns to 0.20% in phase-3 new towns, which is likely due to the inclusion of eco-functional infrastructure such as infiltration basins or bioswales. Office facilities followed a similar pattern with high correlations to both the mean EAR (r = 0.75) and IQR (r = 0.63). Although this land-use type typically had a small area ratio, it exhibits large variations between application sites depending on the presence or absence of rooftop greening and permeable pavements.

In contrast, multifamily housing exhibited the strongest negative correlation with the mean EAR (r = −0.80) and a similarly strong negative correlation with the IQR (r = −0.73). Although it had a moderate ecological performance (mean EAR: 0.34), it consistently had the largest area ratio in most application sites, which explains why new towns with a higher area ratio of this land-use type tended to show both a lower overall EAR and reduced IQR. Other land-use types such as roads, schools, and religious facilities also demonstrated negative correlations with both indicators. These land-use types typically consist of impervious surfaces with minimal ecological function, and their standardized configurations lead to both a low mean EAR and IQR. Lastly, green space showed a near-zero correlation with the mean EAR but a moderately negative correlation with the IQR (r = −0.66). Despite its high ecological weight, it is generally implemented as uniform natural ground, which resulted in consistently high but unvarying EAR values across districts. These findings highlight how certain land-use types influence not only the level but also the structural coherence of the ecological performance of a site.

Taken together, these associations indicate complementary roles for performance and stability. Preserved and landscaped park categories are positively associated with the mean EAR, and consistent with the cross-site synthesis, utility and supply facilities show a similar positive association. Stabilizing institutional types (for example, schools and religious facilities) have standardized layouts and maintenance obligations that constrain coverage choices and thus lower IQR. Both green space and multifamily housing exhibit negative associations with IQR because coverage mix discretion is limited. In green space, contiguous natural ground keeps EAR uniformly high yet relatively invariant, while in multifamily housing standardized block typologies, parking requirements, and structural constraints on engineered greening restrict feasible mixes, and larger shares further narrow dispersion. Conversely, broader within-type coverage discretion in parks and in office and utility categories, such as the proportions of natural-soil versus artificial-soil greening, the density and surfacing of trail and plaza networks, and the inclusion of detention or water features, underlies their positive association with IQR. Built-up categories that rely mainly on engineered greening (for example, rooftop greening and permeable pavements) are sensitive to budget and structural constraints and therefore exhibit wider within-type variability. For planning use, we read the correlations as descriptive guidance: increasing shares of high-weight park types raise the site-level mean, maintaining institutional types helps narrow IQR, and designs that depend on engineered greening should anticipate variability and incorporate safeguards. We frame this section as exploratory spatial data analysis that characterizes distributional patterns and associations for guidance rather than as causal claims. Potential influences and extension paths are summarized in Section 4.2.

3.5. Regional Variations in EAR Distributions

Despite the use of a unified probability-based EAR reference table, significant variations were observed in both the mean EAR and IQR across the seven application sites. These discrepancies illustrate that the ecological performance is not solely determined by the application of standardized coefficients but is instead highly sensitive to the composition, proportion, and interaction of land-use types within each application site. Goyang Changneung, Hanam Gyosan, and Namyangju Wangsuk2 had the highest mean EARs (above 0.46) but also the largest IQRs (above 0.16). They shared a common pattern of high area ratios for green spaces, preserved parks, and utility and supply facilities, which all had a strong positive correlation with the mean EAR. However, they also lacked stabilizing land-use types such as public land, school, and religious facilities, which were previously shown to be associated with lower variability in the EAR. As a result, while the overall ecological contribution was high due to the abundance of land-use types with high ecological weights, internal consistency was reduced owing to site-specific flexibility in the design of land-cover types with ecological functions. Gimpo Hangang and Namyangju Wangsuk both demonstrated intermediate values for the mean EAR and IQR. Their land-use compositions did not strongly favor any single category and instead reflected a balanced mix of multifamily housing, green spaces, parks, and road. These land-use types tended to contribute reliably but not exceptionally to the EAR.

Finally, Yangju Hoecheon and Paju Unjeong had the lowest mean EARs (around or below 0.40) but also the narrowest IQRs (approximately 0.13). These application sites were characterized by large area ratios for multifamily housing and green space while the combined area ratio of preserved and landscaped parks was roughly half that of other application sites. Multifamily housing was consistently associated with both a low EAR and IQR, which contributed to overall ecological uniformity. In this context, the lack of design diversity, especially the limited presence of land-cover areas with high ecological weights, explains the low but stable EAR of these application sites. Part of the observed dispersion also reflects within-type heterogeneity arising from parcel-level coverage-type mixtures (for example, differences in rooftop soil-depth tiers or pavement permeability grades). Consequently, districts with similar land-use shares can still display different IQRs. In summary, these patterns confirm that standardized EAR estimation alone does not ensure uniform ecological outcomes. Rather, it is the strategic combination and structural role of land-use types that determines both the level and stability of EAR at the district scale. A high ecological performance can be achieved through the inclusion of land-use types with high ecological weights, but stability requires a deliberate balance of composition and spatial consistency.

4. Discussion

4.1. Correlation Analysis of Multifamily Housing Subtypes

Multifamily housing was the largest land-use type by area at all application sites. A correlation analysis was conducted between land-cover types and multifamily housing subtypes within housing developments, which included public sale, public rental, and private sale units.

Table 5. Correlation coefficients between land-cover types and EAR values of multifamily housing subtypes.

Coverage Type		Multifamily Housing (Aggregate)	Public Sale	Public Rental	Private Sale
Area on Detention and Infiltration Facility		0.00	0.16	−0.01	0.16
Artificial Ground Greening	10 cm ≤ Soil depth < 40 cm	0.00	−0.52	−0.03	0.18
	40 cm ≤ Soil depth < 90 cm	0.01	−0.25	−0.04	0.06
	Soil depth ≥ 90 cm	−0.25	−0.26	−0.30	−0.32
Natural Soil Greening		0.30	0.34	0.40	0.17
Partial Pavement	Artificial ground	-	0.09	−0.02	0.17
	Natural ground	-	0.09	−0.02	0.06
Permeable Joint Pavement	Artificial ground	0.02	0.05	−0.01	−0.18
	Natural ground	0.02	0.05	−0.01	−0.18
Rooftop Greening	10 cm ≤ Soil depth < 20 cm	0.04	0.13	−0.11	0.13
	20 cm ≤ Soil depth < 30 cm	0.05	0.11	−0.04	0.05
	Soil depth ≥ 30 cm	0.00	0.11	−0.00	0.11
Total Permeable Pavement	Grade 1 Permeability, Artificial Ground	0.01	0.06	−0.03	-
	Grade 1 Permeability, Natural Ground	0.01	0.06	−0.03	−0.02
	Grade 2 Permeability, Artificial Ground	0.04	−0.09	−0.16	0.14
	Grade 2 Permeability, Natural Ground	0.05	−0.21	−0.17	0.18
Wall Greening		0.01	0.18	-	0.16
Water Space	Impermeable	0.08	0.21	−0.07	− 0.26
	Permeable	0.00	-	−0.02	-

Note: Pearson correlation coefficients (r) at the parcel level (n ≥ 30 per subtype). Values are reported as descriptive associations; no multiple-comparison adjustment was applied and p-values are not reported to avoid overinterpretation under multiplicity. “-“ indicates that at least one variable was constant and the correlation was not computed.

presents the correlation coefficients. Natural soil greening had positive correlations with all housing subtypes particularly public rental units (r = 0.40), which may reflect that publicly managed housing tends to integrate land-cover types with higher ecological weights. In contrast, deep artificial ground greening (≥90 cm) showed a negative correlation, especially with private sale units (r = −0.32), which indicates that intensive soil layers are less commonly applied in private-sector developments. Other features such as permeable pavements, detention and infiltration areas, and rooftop greening exhibit mixed or weak correlations, which reflects differing design strategies across housing subtypes. Public sale housing is supplied with the support of the national or local government and public agencies to assist various socioeconomic groups such as national merit recipients, newlyweds, young adults, and low- to middle-income households. Similarly, public rental housing is provided by the government or public institutions but is specifically targeted toward low-income or non-homeowning populations with reduced deposits and rents. In contrast, private sale housing is supplied by private developers or corporations based on market principles and their own criteria. These institutional differences in multifamily housing subtypes may partly explain the varying associations with land-cover types of different ecological weights [38,39,40]. Building on these differences, guidance can be refined in subsequent work by incorporating project-level subtype flags and reporting stratified quartiles, as outlined in Section 4.2. Following an exploratory framing, we use correlations to characterize patterns rather than to make confirmatory claims and interpret them descriptively; where p-values are shown, they are unadjusted.

Based on the above correlation analysis,

Table 6. Interpretation of EAR-correlated factors by multifamily housing subtype.

Category	EAR Positively Correlated Factors (+)	EAR Negatively Correlated Factors (−)	Interpretation
Capital Region	Area on detention and infiltration facility, Natural soil greening	Artificial ground greening (Soil depth ≥ 90 cm), Permeable pavement	EAR mostly relies on natural ground
Public sale housing	Natural soil greening, Wall greening (weak)	Artificial ground greening (10cm ≤ Soil depth < 40 cm), Artificial ground greening (Soil depth ≥ 90 cm)	Uniform layout, high proportion of natural ground
Public rental housing	Natural soil greening	Artificial ground greening (Soil depth ≥ 90 cm), Most other types	Simple structure, low EAR, dependent on natural ground
Private Sale housing	None (generally weak)	Artificial ground greening (Soil depth ≥ 90 cm), Some permeable pavements	Despite diversity in composition, low EAR contribution

Note: Interpretations summarize descriptive associations from Table 5 and Section 4.1. They are intended for exploratory guidance rather than confirmatory claims.

outlines factors associated with each multifamily housing subtype and their positive or negative correlation with EAR. Across all subtypes, natural soil greening consistently had a strongly positive correlation with EAR with the strongest correlations observed in public rental units (r = 0.40) followed by public sale units (r = 0.34) and private sale units (r = 0.17). These correspond to modest to moderate positive associations, suggesting that public housing subtypes, especially rental units, are structurally aligned with land-cover surfaces that have high ecological weights. This outcome is consistent with their institutional design conditions, which emphasize low building coverage, fixed open space ratios, and standardized greening strategies. Conversely, deep artificial ground greening (≥90 cm) showed negative correlations across all housing subtypes, most notably with private sale units (r = −0.32). This pattern may reflect budgetary and structural constraints associated with intensive deep-soil systems, which are reported to entail higher installation and maintenance costs and load requirements; privately financed projects operating under market constraints may therefore favor lighter or surface-oriented greening instead [41,42,43]. Public rental units also exhibited a negative correlation (r = −0.30) though to a lesser extent while public sale units differed markedly by displaying a more pronounced negative correlation (r = −0.52), which is indicative of their distinctive design approach. This suggests that, while public sale units still follow a government-led framework, they selectively substitute deep soil installations with more manageable or visible surface treatments. Other land-cover types showed varied trends. Rooftop greening (10–30 cm) exhibited weakly positive correlations with public sale and private sale units (r ≈ 0.13–0.14) but not with public rental units. Wall greening had more positive correlations with private housing (r = 0.16), which is consistent with its use as a design-driven feature in market-oriented developments. Permeable joint pavements and detention/infiltration areas showed weak or inconsistent correlations across all housing subtypes, which is consistent with their secondary role in EAR composition.

These patterns reveal clear differences in how land-cover types are deployed by each housing subtype. Public rental housing is characterized by the strongest regulatory controls with uniform spatial arrangements and prescriptive greening requirements. As a result, its EAR is dominated by the consistent application of natural soil and basic permeable surfaces while roof and wall greening are virtually absent. Although public sale housing is still governed by public planning, it allows more design flexibility and compensatory strategies, which explains its moderate correlation with natural soil greening and strongly negative correlation with deep artificial greening. Rather than including costly subsurface features, developers of public sale housing may favor rooftop or pavement-based solutions to meet green ratio requirements within a more relaxed regulatory scope. Private sale housing operates under the loosest constraints and incentive-based design, which may explain its moderate correlations with rooftop and wall greening, weak correlation with natural soil, and strongly negative correlation with deep artificial greening. This suggests a tendency toward highly selective and often visible features used strategically for marketing or bonus incentives rather than systemic ecological planning.

These findings indicate that the ecological structure of multifamily housing is influenced not only by the area ratio but also by the underlying planning obligations, regulatory regime, and economic priorities, which differ across housing subtypes. To ensure more consistent ecological performance, differentiated EAR guidelines may be needed that acknowledge these structural and institutional differences in development practice [44,45]. Operational considerations differentiated by housing subtype are presented in Section 4.2. More broadly, this study highlights the importance of decoupling land-use composition and regulatory frameworks when evaluating the effectiveness of EAR for large-scale urban developments. The combination of quantitative correlation analysis with typological differentiation offers empirical evidence that ecological outcomes are shaped not only by the land-cover type and its ecological weight but also by spatial governance, design standards, and policy incentives. Unlike prior approaches that relied on single policy thresholds or fixed coefficients, these insights can inform more targeted and equitable green infrastructure strategies, particularly in the context of rapidly expanding new towns where the uniform application of EAR standards may mask significant spatial and institutional variations.

4.2. Practical Applicability and Limitations

This study follows a planning-consistent set of 16 land-use types for all analyses. Within-type heterogeneity is indirectly represented through parcel-level mixtures of 19 coverage types and summarized by the empirical quartiles and the IQR. For applications where a jurisdiction uses a different land-use taxonomy or records sub-class flags (for example, tenure or procurement mode for housing subtypes, road hierarchy, floor-area-ratio bands, roof soil-depth tiers, or pavement permeability grades), the framework can be extended in a comparability-preserving manner by reporting stratified quartiles for the relevant sub-classes and by remapping local categories to the 16-type scheme when aggregating site-level targets. Where such metadata are not available, the type-level quartiles and IQR provide a compact and consistent basis for planning decisions. These procedures make the results applicable to studies that use different or coarser land-use and cover taxonomies. Where only aggregate land-use proportions are available, type-level percentile targets offer defaults and the interquartile range provides a simple uncertainty band.

Although the empirical reference values and the 16-type taxonomy were calibrated on large-scale urban development districts, the framework can be applied beyond urban settings. In predominantly rural contexts where non-built covers dominate, the EAR may approach the upper bound and therefore has limited discriminative power at the district scale. In such cases, applications are better scoped to settlement footprints or infrastructure corridors, and jurisdictions may consider localized weights or additional classes (for example, cropland, pasture, managed forests, riparian buffers) together with complementary metrics of habitat quality, connectivity, or water quality.

In practice, the probability-based EAR can be implemented with standard planning datasets: parcel-level land-use proportions and coverage-type mixes mapped to the 16-type scheme. At the concept stage, agencies select per-type percentile anchors from Table 4, compute a proportion-weighted site target, and consult type-level IQRs to screen stability; at design development, projects document the coverage mix used to meet the anchors; at approval, agencies check compliance against the site target with the IQR used as a diagnostic. To keep references current under evolving land uses and climate, agencies can maintain a versioned reference table, re-estimating quartiles on a periodic schedule or when sufficient new sites are added, while retaining the 16-type mapping for comparability. Candidate demonstrations such as case studies or simulation-based policy testing are noted as future work.

The analyses are descriptive and exploratory, and we refrain from causal claims. We did not explicitly model potential confounding from development stage and density (for example, floor-area-ratio bands), regulatory and institutional obligations, socioeconomic context, or site-specific geography (for example, slope and hydrology). Where spatial-pattern metadata are available, the framework can be extended to account for within-complex layout and diversity by reporting stratified quartiles or localized weights with respect to configuration and diversity metrics (for example, connectivity, patch-size distributions, edge density, patch richness and evenness, or Shannon diversity), and by incorporating basic spatial autocorrelation diagnostics. These extensions can be implemented without altering the 16-type scheme by localizing weights or stratifications. More generally, future work can implement formal identification and uncertainty propagation. In addition, this study adopted the ecological weights defined by MOE Notice No. 2022-61, which ensures consistency with national standards. While this enhances comparability across sites, empirical validation or sensitivity analysis of the weights was not within the scope of this study. Future research could complement the framework with field-based assessments or uncertainty testing to further strengthen the robustness of EAR applications.

Building on the associations reported in Section 4.1, we outline non-prescriptive options by housing subtype, using Table 4 percentile references as contextual anchors. For public rental housing, planners may consider anchoring EAR targets near the median (Q2) while prioritizing high-weight covers, and consult the type-level IQR to review stability. For public sale housing, maintaining targets around Q2 with a priority on natural soil is a feasible option; where deep-soil systems face structural or budget constraints, performance can be met through compensatory high-weight measures (for example, detention and infiltration areas or higher-grade permeable surfaces), and the IQR can flag schemes that may require safeguards. For private sale housing, jurisdictions may encourage the presence of at least one high-weight cover class per block where feasible (for example, natural soil or deep-soil greening), and, where roof greening is feasible, select thickness classes associated with higher ecological weights. Specific thresholds and mixes should remain jurisdiction- and project-specific; Table 4 provides default anchors, and the IQR serves as a stability diagnostic, consistent with the exploratory framing. The EAR is a coverage-weighted, surface-based index and does not directly quantify biodiversity or vertical vegetation structure. Where such data are available, companion indicators can be co-reported, and crosswalks to established greening metrics can be implemented by recoding coverage classes; quantitative benchmarking is beyond the scope of this study and is noted as future work.

5. Conclusions

This paper develops a data-driven, probability-based guideline for the EAR that replaces single-number thresholds with land-use-specific percentile references and a stability readout based on the interquartile range, enabling early-stage target setting while preserving comparability to the 16-type policy scheme. Using empirical data from large-scale development districts, we estimated distributional reference values (means and quartiles) for 16 land-use types and demonstrated early-stage applicability by predicting site-level EAR in new towns. Land-use types with high ecological weights such as green spaces and parks consistently delivered a strong ecological performance while multifamily housing contributed significantly to the overall EAR despite its moderate ecological weight because of its high area ratio. Further correlation analysis of multifamily housing subtypes revealed that the EAR distribution is influenced not only by the land-cover type but also by the regulatory and institutional structures tied to each housing subtype. Public rental housing showed a strong correlation with land-cover types with high ecological weights such as natural soil greening whereas private sale housing favored more cost-efficient or visually prominent elements such as rooftop and wall greening. These patterns underscore the importance of development of differentiated EAR strategies that consider planning constraints and development incentives. Despite the application of the probability-based EAR reference table as a standardized framework of reference, significant variations were observed across the application sites, which indicates that ecological consistency requires more than uniform application of a single standard. The strategic balance and distribution of land-use types are equally critical. Relative to prior work centered on single thresholds or uniform coefficients, this study provides land-use-specific percentile targets together with an explicit IQR-based stability indicator, enabling early-stage estimation within the 16-type scheme and supporting more targeted, equitable green infrastructure strategies.