Integrated Modeling of Wildfire Ignition Risk in the Military–Civilian Interface of the Korean DMZ

Integrated Modeling of Wildfire Ignition Risk in the Military–Civilian Interface of the Korean DMZ | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Integrated Modeling of Wildfire Ignition Risk in the Military–Civilian Interface of the Korean DMZ Sujung Heo, Sujung Ahn, Song Hee Han, Sung Eun Cha, Mi Na Jang, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7872076/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Military–civilian interface zones (MCIZs) adjacent to the Korean Demilitarized Zone (DMZ) represent complex wildfire environments shaped by restricted access, intensive military activities, and adjacent civilian land use. This study develops a spatially explicit wildfire ignition risk assessment framework for the DMZ and Civilian Control Zone (CCZ) in Paju, South Korea, by integrating Random Forest (RF), Generalized Additive Models (GAM), and Geographically Weighted Regression (GWR). A dataset of 318 wildfire ignition events (2001–2024), including 78 associated with military activities, was analyzed. The RF model achieved high predictive accuracy (AUC = 0.81), identifying proximity to military training zones, relative humidity, wind speed, and proximity to built infrastructure as dominant ignition drivers. GAM revealed narrow nonlinear thresholds—relative humidity at 13.8–14.0% and wind speed at 13.5–14.0 m/s—corresponding to peak ignition probabilities. GWR demonstrated pronounced spatial heterogeneity, with military proximity exerting stronger influence in eastern and northern sectors, while meteorological effects varied geographically. Based on these outputs, an integrated modeling framework was established to synthesize probabilistic model results into spatially explicit ignition susceptibility maps. The proposed approach provides a scientifically rigorous and operationally applicable method for quantifying ignition risk in politically sensitive, access-restricted landscapes, offering valuable insights for adaptive wildfire prevention and spatially informed governance of transboundary fire risk. Wildfire ignition risk Integrated modeling framework Random Forest (RF) Generalized Additive Model (GAM) Geographically Weighted Regression (GWR) Spatial heterogeneity Military–civilian interface zones (MCIZs) Demilitarized Zone (DMZ) Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Wildfires have increasingly emerged as a critical global threat, exacerbated by the intersection of climate change, land-use transformations, and the expansion of human activity into fire-prone landscapes (Bowman et al., 2017 ; Moreira et al., 2020 ). While substantial advances have been made in wildfire modeling and mitigation, much of the research has focused on either densely populated wildland–urban interface (WUI) zones or remote forested regions (Murray et al., 2023 ; Chen et al., 2024 ). Transitional areas such as military–civilian interface zones (MCIZs), however, remain substantially underexamined despite their distinct geographies of risk, which are shaped by restricted access, concentrated military activity, and adjacent civilian development (Kim et al., 2021 ). This research gap is particularly pronounced in Republic of Korea (“Korea”), especially within the western Demilitarized Zone (DMZ) and Civilian Control Zone (CCZ), where live-fire military exercises, agricultural residue burning, and climate-sensitive vegetation dynamics converge to elevate wildfire ignition risk (Kim & Lee, 2024 ). From 2001 to 2024, national fire statistics and data from the Fire Information for Resource Management System (FIRMS) indicate that over 95.65% of wildfires in these regions are attributed to anthropogenic ignition sources, with military activities (56.52%) and civilian open burning (20.29%) representing the primary causes (Korea Forest Service, 2025 ; FIRMS, 2025 ). Nevertheless, systematic spatial modeling of wildfire risk and rigorous evaluation of mitigations remain limited for these sensitive zones (Lee et al., 2022 ). Recent advancements in data science and spatial analysis, including machine learning, spatial regression, and geospatial modeling, provide promising avenues to capture the multi-dimensional and spatially heterogeneous nature of wildfire ignition risk (Marcos et al., 2021 ; Bisenic, 2022 ; Andrianarivony & Akhloufi, 2024 ). Random Forest (RF) models have demonstrated strong performance in ignition risk prediction and variable selection (Heo et al., 2023 ), Generalized Additive Models (GAM) facilitate robust exploration of nonlinear thresholds and ignition probabilities (Detmer et al., 2025 ), while Geographically Weighted Regression (GWR) enables the assessment of spatial heterogeneity across complex landscapes (Punzo et al., 2022 ). However, few studies have integrated these methodologies to jointly address predictive accuracy, interpretability, and spatial non-stationarity—particularly within the context of access-restricted military landscapes (Kim et al., 2021 ; Kim & Lee, 2024 ). To address these gaps, this study proposes a spatially explicit wildfire ignition risk assessment framework specifically tailored to MCIZs. By integrating RF, GAM, and GWR, the framework systematically identifies the principal climatic, topographic, anthropogenic, and military drivers of ignition, quantifies their nonlinear thresholds, and evaluates their spatial heterogeneity across the DMZ–CCZ landscape. Building on these complementary outputs, an integrated modeling framework was developed to synthesize probabilistic model results into spatially interpretable ignition susceptibility maps, providing a unified representation of wildfire risk patterns. The research is guided by three key questions: (1) Which factors exert the strongest influence on wildfire ignition in MCIZs? (2) Under what threshold conditions (e.g., low relative humidity, high wind speeds, or proximity to military training areas) does ignition risk sharply increase? (3) How do these effects vary geographically, revealing spatial patterns of ignition vulnerability? 2. Study area The present study focuses on the northern region of Paju City, Korea, which is characterized by its direct proximity to the DMZ and the CCZ (Fig. 1 ). This locale constitutes a complex socio-ecological frontier, where zones of military activity—including installations, live-fire training grounds, and restricted buffer strips—converge with areas dedicated to civilian settlements, agriculture, and transportation infrastructure. Following the Korean Armistice Agreement of 1953, the DMZ has remained under stringent military supervision with civilian access strictly prohibited, while the adjacent CCZ has permitted limited agricultural activity and residential development, resulting in divergent land use practices across relatively short distances (Kim et al., 2021 ). The western sector of Paju’s CCZ, contiguous with the DMZ, exemplifies the dynamic interface between military infrastructure and rural livelihoods. The land use mosaic in this sector includes ginseng cultivation, traditional irrigation ponds (dumbeong), as well as critical habitats for migratory bird species along the Imjingang River (Lee et al., 2022 ). Recent ecological assessments have documented that, within the DMZ, historical coniferous forests damaged during the Korean War have gradually transitioned to mature deciduous woodlands due to uninterrupted natural succession processes. In contrast, significant portions of the CCZ have been converted into agricultural or infrastructural developments due to ongoing human intervention (Kim & Lee., 2024). From 2001 to 2024, a comprehensive geospatial analysis identified 318 wildfire ignition points within the study area. Of these, 78 incidents were directly attributed to military live-fire exercises according to data compiled from the Korea Forest Service and the NASA FIRMS archives (Korea Forest Service, 2025 ; NASA FIRMS, 2025 ). The high frequency and spatial clustering of these military-origin ignitions within both the DMZ and CCZ underscore the heightened risk of transboundary wildfires in heavily militarized border zones (Brady, 2021 ). Demographic trends throughout the 2020s reveal rapid population growth in Paju, fueled by expansion from the Seoul metropolitan fringe and government-driven regional development initiatives (Kim et al., 2021 ). This urbanization has intensified the exposure of communities situated along the wildland–urban interface to wildfire hazards. Against this backdrop, the study area offers a salient case for examining wildfire risk in the context of intersecting military and civilian land uses, within a region that is simultaneously politically sensitive and ecologically significant (Kim et al., 2022 ). 3. Methods 3.1. Study framework This study introduces a spatially explicit framework for wildfire ignition risk assessment, specifically designed for MCIZs situated in restricted border areas. The framework integrates advanced machine learning and spatial regression approaches to comprehensively capture the interplay among climatic, topographic, anthropogenic, and institutional determinants of wildfire ignition in politically and ecologically sensitive contexts (see Fig. 2 ). Wildfire occurrence and its underlying drivers were modeled using three complementary analytical techniques. First, the RF model was employed to generate spatially explicit ignition probability maps using a 70:30 training-to-testing data split for robust validation. Second, the GAM was applied to investigate nonlinear relationships and threshold effects among key covariates such as relative humidity, wind speed, and vegetation indices. Third, the GWR approach was used to assess the spatial heterogeneity in the influence of predictors, allowing localized interpretation of ignition drivers across the DMZ–CCZ landscape. Variable selection was guided by Variance Inflation Factor (VIF) diagnostics, correlation matrices, and Random Forest (RF)-derived feature importance metrics, ensuring statistical rigor and model parsimony. Key outputs of the framework include wildfire ignition probability maps, composite spatial layers derived from the integrated modeling framework, and GWR-based local coefficient surfaces. Together, these outputs provide high-resolution, spatially differentiated insights into ignition dynamics and support data-driven wildfire risk assessment and spatial decision-making in sensitive borderland environments. 3.2. Variable Selection and Screening To quantitatively assess wildfire risk in the DMZ-adjacent landscape of Paju, we identified 11 key environmental drivers following criteria of theoretical relevance, regional context, data accessibility, and empirical support (Fig. A1 ). The selected variables were systematically organized into four core categories—climatic, topographic, land use/vegetation, and anthropogenic (civilian and military)—to adequately capture both biophysical and human-induced factors implicated in fire ignition and spread (Table 1 ). In a region characterized by a unique intersection of military activity, rural land use, and ecological succession, this multidimensional selection was essential for robust modeling. All variables were processed to a spatial resolution of 10 m × 10 m to ensure fine-scale representation of environmental heterogeneity, and matched to the wildfire inventory period (2001–2024). Climatic factors, in particular, were represented by minimum relative humidity and maximum wind speed—parameters selected to maximize the representation of conditions most conducive to ignition and rapid-fire spread. These data were sourced from the Korea Meteorological Administration and aligned with the study’s wildfire inventory period. Both relative humidity and wind speed are recognized as principal determinants of wildfire activity in fire-prone landscapes (Ying et al., 2021 ; Sutherland et al., 2023 ), with historical evidence from the DMZ showing that low humidity combined with episodic gusts has triggered rapidly spreading fires, especially when interacting with military ordnance (Park et al., 2019 ). Topographic variables—slope aspect, slope degree, elevation, and the topographic wetness index (TWI)—were derived from high-resolution Digital Elevation Models (DEM) obtained via the SRTM 1 Arc-Second Global dataset. These layers were selected to reflect current terrain conditions as of 2024, ensuring accurate representation of slope-mediated microclimatic and hydrological influences on fire behavior (Malik et al., 2021 ; Abbate et al., 2019 ; Heo et al., 2023 ). Topography critically mediates fuel continuity, solar radiation exposure, and soil moisture availability, all of which are important modulators of ignition probability and fire propagation in the complex terrain of the Korean borderlands. Land use and vegetation conditions were characterized using Sentinel-2 (2024) land cover classification and the Enhanced Vegetation Index (EVI) derived from Landsat 8 (2001–2024) imagery. The Sentinel-2 Land Use and Land Cover (LULC) layer provided an up-to-date depiction of land use patterns, while EVI—favored over NDVI for its superior sensitivity to canopy structure and resistance to saturation in dense vegetation—captured seasonal and interannual variation in vegetation greenness (Salavati et al., 2022 ; Malik et al., 2021 ). These datasets were crucial for distinguishing between military training grounds, agricultural fields, and peri-urban residential areas, each exhibiting different fuel structures and ignition susceptibilities. Anthropogenic variables included distance to the nearest road, distance to the nearest building, and proximity to live-fire military training zones, calculated from 2024 national infrastructure datasets (Ministry of Land, Infrastructure and Transport; National Geographic Information Institute) and Korea Forest Service military facility geospatial layers. Proximity to civilian infrastructure is associated with multiple ignition sources, including agricultural residue burning, recreational activities, and electrical faults (Heo et al., 2025; Naser & Kodur. 2025). In the study area, proximity to military training grounds has been documented as a major ignition driver, with incident records attributing approximately 24.5% of wildfires (78 out of 318 events) directly to military exercises and ordnance use (Kim & Lee. 2024; Lee et al. 2022 ). To mitigate multicollinearity and ensure statistical validity, all candidate variables underwent VIF and pairwise Pearson correlation screening (Fig. 3 , Table 2 ). Variables with VIF values below the conventional threshold of 5 were retained (max VIF = 5.44 for “Distance to Road”; VIF = 4.51 for “Distance to Building”), validating their inclusion. The correlation matrix indicated no severe interdependencies, with the highest correlation (r = 0.60) observed between road and building proximity—considered conceptually and spatially distinct within the regional context. Table 1 Selected list of environmental drivers, data sources, period, and references used for fire risk modeling. Category Factor Source Period Reference Climatic Min Relative Humidity Korea Meteorological Administration ( https://data.kma.go.kr ) 2001–2024 Ying et al., 2021 Max Wind Speed Sutherland et al., 2023 ; Heo et al., 2025 Topographic Slope Aspect DEM - Heo et al., 2023 ; Ángel et al., 2023 Digital Elevation Model SRTM 1 Arc-Second Global ( https://earthexplorer.usgs.gov/ 2024 Malik et al., 2021 ; Salavati et al., 2022 Slope Degree DEM - Abbate et al., 2019 ; Asori et al., 2020 Topographic Wetness Index DEM - Fang et al., 2018 ; Nasiri et al., 2022 Land Use and Vegetation Land Use and Land Cover Sentinal-2 (ESA via ESRI Platform) 2024 Salavati et al., 2022 ; Donovan et al., 2020 Enhanced Vegetation Index Landsat 8 ( https://earthexplorer.usgs.gov/ ) 2001–2024 Malik et al., 2021 ; Costa-Saura et al., 2022 Civilian Distance to Road Korea Ministry of Land, Infrastructure and Transport ( https://www.molit.go.kr/english/intro.do ) 2024 Cao et al., 2021 ; Heo et al., 2024 Distance to Building Infrastructure National Geographic Information Institute ( https://ngii.go.kr ) 2024 Naser & Kodur. 2025; Papathoma-Köhle et al., 2022 Military Proximity to Live-Fire Exercises Korea Forest Service ( https://fqgis.forest.go.kr ) - Kim & Lee. 2024; Lee et al. 2022 Table 2 Results of VIF analysis for 11 targeted variables. No. Feature VIF 1 Relative Humidity (%) 1.10 2 Wind Speed (m/s) 1.43 3 Slope Aspect 1.04 4 Slope Degree 1.64 5 Topographic Wetness Index 1.02 6 Elevation 1.82 7 Distance to Live-Fire Exercises (m) 3.24 8 Distance to Building Infrastructure (m) 4.51 9 Land Use and Land Cover 1.14 10 Distance to Road (m) 5.44 11 Enhanced Vegetation Index 1.47 3.3. Modeling Approaches To characterize wildfire risk arising from military–civilian interactions within the DMZ-adjacent region of Paju, we implemented a hybrid modeling framework that integrates three complementary analytical techniques: the GAM, RF, and GWR. Each approach was strategically selected to address specific dimensions of wildfire risk, encompassing global nonlinear patterns, spatial non-stationarity, and intricate interactions among environmental and anthropogenic drivers. The GAM is a flexible, semi-parametric regression framework designed to model potential non-linear associations between wildfire occurrence and predictor variables. GAM utilizes smooth spline functions to represent covariate effects, facilitating the detection of ecological thresholds and non-monotonic responses within ignition dynamics (Sagrario et al., 2025; Ye et al., 2020 ). In our analysis, GAM was deployed to elucidate the influence of key variables—including relative humidity, vegetation status, and distance to military training zones—on wildfire probability. This approach is particularly suited to the DMZ–CCZ context, where environmental gradients are modulated by restricted access, complex hydrological conditions, and episodic military disturbances. The GWR method accommodates spatial non-stationarity by allowing local variation in regression coefficients, thus capturing the possibility that the relationship between wildfire occurrence and its predictors may change across the heterogeneous landscape (Pahlavani et al., 2024 ; Schag et al., 2022 ). In the context of the Paju border region, GWR is essential for disentangling spatially-varying risk patterns, such as those found in agricultural zones versus areas adjacent to military infrastructure. Through the estimation of location-specific parameter estimates, GWR offers critical insights for the design of place-based fire management and risk mitigation strategies. The RF algorithm is an ensemble-based machine learning technique that aggregates multiple decision trees built on bootstrapped samples of the data. Owing to its robustness to noisy inputs, multicollinearity, and its capacity to handle high-dimensional, interacting predictors, RF is well-suited for modeling complex wildfire ignition processes (Heo et al., 2023 ). In this study, RF was instrumental in quantifying the relative importance of each environmental and anthropogenic variable—most notably, proximity to critical infrastructure and military facilities. By jointly employing GAM, RF, and GWR, we sought to balance interpretability, predictive performance, and spatial specificity. GAM enables global interpretability through partial dependence plots, highlighting nonlinear threshold effects of predictors such as wind speed or vegetation indices. RF complements this by modeling complex, higher-order interactions and generating robust variable importance rankings. GWR further enhances spatial fidelity by explicitly modeling spatial heterogeneities in predictor influence, a feature essential for nuanced fire risk governance in a transboundary, militarized setting (Pahlavani et al., 2024 ; Cao., 2020). All models were independently trained on a commong set of pre-screened variables defined in Section 3.2. Model performance was rigorously assessed using standard evaluation metrics: F1-score, Area Under the ROC Curve (AUC) and Root Mean Squared Error (RMSE). These metrics offer a comprehensive appraisal of both classification accuracy and probabilistic modeling, and align with established practices in recent wildfire risk research (Heo et al., 2025; Wang. & Wang., 2020). 4. Results 4.1. Model Validation and Performance Assessment The predictive performance of the RF, GAM, and GWR approaches was assessed using a comprehensive set of classification and regression metrics (Table 3 ). Evaluation criteria included overall accuracy, AUC, RMSE, as well as class-specific precision, recall, and F1-score statistics. Both RF and GWR models achieved the highest overall accuracy at 0.86, while the GAM model registered a value of 0.82. RF and GWR also shared the highest AUC values at 0.91, indicating strong model performance in distinguishing between wildfire and non-wildfire events. The GAM yielded an AUC of 0.86. Class-specific performance for wildfire occurrence (Class 1) revealed the highest F1-score with RF (0.67), followed by GWR (0.57) and GAM (0.44). Class 0 (no fire) predictions were robust across all models, with F1-scores above 0.90 and the highest achieved by GWR (0.92). These validation outcomes demonstrate that all three modeling approaches performed satisfactorily based on the selected evaluation metrics, with RF and GWR exhibiting consistently strong results in both classification and regression-based assessments. Table 3 RF, GAM and GWR Model performance metrics for wildfire occurrence prediction. Metric RF GAM GWR Accuracy 0.86 0.82 0.86 AUC Score 0.91 0.86 0.91 RMSE 0.41 0.34 0.31 Precision (Class 0) 0.87 0.85 0.87 Precision (Class 1) 0.81 0.56 0.77 Recall (Class 0) 0.95 0.93 0.97 Recall (Class 1) 0.57 0.36 0.45 F1-Score (Class 0) 0.91 0.89 0.92 F1-Score (Class 1) 0.67 0.44 0.57 Class 0: No wildfire occurrence; Class 1: Wildfire occurrence; AUC Score is based on the ROC curve. 4.2. Nonlinear Effects of Environmental Drivers The GAM identified minimum relative humidity (RH) and maximum wind speed (WS) as the most influential predictors of wildfire occurrence (p < 0.05), with both variables displaying statistically significant nonlinear effects (Fig. 5 ). The partial dependence plot for RH exhibited a bell-shaped relationship, with the peak modeled fire probability between approximately 13.8% and 14.0%, followed by a pronounced decline outside this interval. WS demonstrated a non-monotonic association, with wildfire probability increasing up to around 13.5–14.0 m/s before showing a decrease at elevated WS values. Slope aspect (ASP) showed modest variation, with amplified fire probabilities on west- to northwest-facing slopes. Slope degree (SLO) displayed a U-shaped relationship, with augmented probabilities at both low ( 30°) gradients. The TWI exhibited a double-peak pattern, with increased probabilities in both low-TWI areas (approximately − 8) and high-TWI areas (above − 2), and reduced probabilities in the intermediate range. DEM indicated elevated fire probability in low-lying terrain (< 100 m), with a secondary increase above ~ 1,000 m. For anthropogenic proximity variables, distance to military live-fire wildfires (MILITARY) followed an inverted-U pattern, peaking at intermediate distances of approximately 7–8 km and decreasing at shorter and longer distances. Distance to buildings (BUILDING) showed amplified probabilities within 0–2 km, declining steadily with increasing distance. Distance to roads (ROAD) exhibited a generally increasing trend, with a marked rise beyond 6 km. Among environmental variables, LULC was highly significant (p < 0.001), with greater wildfire probabilities associated with rangeland, bare ground, and cropland, and lower probabilities for water bodies and flooded vegetation. EVI reached maximum fire probability at intermediate values (0.3–0.4), with diminished probabilities for both sparse ( 0.6) vegetation cover. 4.3. Spatial Wildfire Risk Prediction The RF model demonstrated robust predictive capacity for wildfire ignition probability across the study area, revealing clear spatial patterns in risk distribution (Fig. 6 A). Continuous RF-derived probabilities were classified using the Natural Jenks method into five ordinal risk categories: very low (0–0.05), low (0.05–0.17), moderate (0.17–0.32), high (0.32–0.51), and very high (0.51–1.00). Mapping of these classes indicated that wildfire risk is predominantly concentrated along the region’s western and northern margins, notably in zones adjacent to urban interfaces and major road infrastructure. Area statistics (Table A1 ) showed that approximately 88.88% of the region was assigned to the very low-risk category, while high and very high-risk areas comprised 1.61% and 0.73% of the total area, respectively. Despite their limited spatial extent, these higher-risk zones corresponded closely with areas of intensified human–environmental interactions. To visualize these priority areas, Fig. 6 B highlights the aggregated high and very high-risk zones, and Figs. 6 B- 1 and 6 B- 2 provide detailed views of representative hotspots. 4.4. Spatial Heterogeneity of Wildfire Risk Drivers The GWR model was employed to characterize the spatial variability in the effects of 11 key wildfire risk drivers. Summary statistics of the local coefficient estimates are presented in Table 4 , and their geographic distributions are visualized in Fig. 7 . WS was associated with greater spatial variation in effect (range: −4.87 to 3.05; mean = 0.02), although the mean effect was close to zero. Positive WS associations were prevalent in the northern and northeastern regions (Fig. 7 A). Analysis of the GWR outputs indicated pronounced spatial heterogeneity in the influence of both climatic and anthropogenic variables. RH exhibited the greatest spatial variability among all predictors, with local coefficients ranging from − 4.04 to 4.74 (mean = − 0.21). The majority of the study area—particularly in central and northern sectors—displayed a negative association between RH and fire occurrence, indicating increased ignition probability under low-humidity conditions (Fig. 7 B). Topographic factors, including slope degree and aspect, showed relatively minor and stable coefficients near zero, suggesting minimal explanatory contribution to spatial ignition variation in this context. The TWI revealed consistently negative coefficients overall (mean = − 0.003), with the strongest effects observed in the southeastern region, aligning with areas of drier terrain (Fig. 7 F). Anthropogenic features such as distance to live-fire exercises, buildings, and roads exhibited modest mean effects (mean coefficients: ~0.00005 to 0.0001). Notable spatial variation was evident near urban and semi-urban boundaries, especially in the northeast and central zones, where the relationship between human proximity and wildfire occurrence was stronger (Figs. 7 G–I). Vegetation and land cover variables demonstrated greater spatial influence. The EVI showed a positive association with wildfire probability across most of the study area (mean = 0.17; peak up to 0.85) (Fig. 7 K). Similarly, LULC had moderate positive coefficients (mean = 0.007), with prominent effects along forest–agriculture transitions in southern and central boundary regions (Fig. 7 J). Table 4 Summary of GWR Parameter Estimate. (RH: Min Relative Humidity; WS: Max Wind Speed, ASP: Slope Aspect, SLO: Slope Degree, TWI: Topographic Wetness Index, DEM: Digital Elevation Model, Military: Distance to Live-Fire Exercises, BUILDING: Distance to Building Infrastructure, LULC: Land Use and Land Cover, ROAD: Distance to Road, EVI: Enhanced Vegetation Index) Variable Mean Median Std Min Max RH -0.21 -0.23 1.27 -4.04 4.74 WS 0.02 0.21 1.13 -4.87 3.05 ASP 0.0002 0.0001 0.0004 -0.0005 0.001 SLO -0.00000001 -0.00000001 0.00000004 -0.00000012 0.00000007 TWI -0.003 -0.001 0.006 -0.015 0.013 DEM 0.001 0.001 0.002 -0.003 0.006 MILITARY 0.00005 0.00003 0.0001 -0.0005 0.0004 BUILDING 0.0001 0.0001 0.0001 -0.0003 0.0004 LULC 0.007 0.005 0.009 -0.014 0.034 ROAD 0.00004 0.00004 0.0002 -0.0004 0.001 EVI 0.17 0.18 0.22 -0.55 0.85 5. Discussion 5.1. Primary Drivers and Their Spatial Variation The RF and GWR models exhibited strong predictive performance (AUC = 0.91), highlighting proximity to military installations, low RH, elevated WS, and built infrastructure as the primary determinants of wildfire ignition. These findings corroborate previous research indicating that military training activities and urban–rural anthropogenic factors jointly influence wildfire ignition across DMZ–CCZ regions (Anton & Lawrence, 2016 ; Laushman et al., 2020 ). The GAM analysis elucidated nonlinear hazard thresholds: RH demonstrated peak ignition probability within a narrow 13.8–14.0% window, a condition likely optimizing fuel dryness and ignition success, while WS showed a similar peak at 13.5–14.0 m/s, after which increased turbulence may suppress ignition stability (Richards et al., 2022). Analysis of spatial proximity revealed that ignition risk for military training zones followed an inverted-U shape, peaking at 7–8 km—closely aligned with typical operational training radii and potentially linked to detection or suppression response delays. Roads and buildings showed increased risk at 0–2 km, reflecting human-caused ignition patterns observed in peri-urban settings (Laushman et al., 2020 ). GWR results further underscored spatial heterogeneity; military proximity had the strongest effect in eastern and northern sectors alongside meteorological effects that varied considerably across the study area, indicating localized operational contexts and environmental conditions drive ignition patterns (Kim & Lee, 2024 ; Coutaz, 2018 ). 5.2. Integrated Modeling for Spatial Precision and Policy Relevance By integrating RF, GAM, and GWR, this study overcomes the constraints of single-model approaches, particularly relevant for regions with limited data access and complex socio-political overlays. RF was instrumental in providing robust variable ranking, GAM quantified critical nonlinear ignition thresholds for predictive interpretation, and GWR revealed micro-scale geographic variations that are essential for spatially precise wildfire risk assessment (Rodrigues & de la Riva, 2014; Wu et al., 2018 ). The integration of these outputs within an integrated modeling framework enabled the development of weighted and spatially normalized ignition susceptibility maps — a substantial methodological improvement over earlier wildfire risk assessments in the DMZ that relied primarily on coarse statistics or post-fire remote sensing (Lee et al., 2022 ; Dillon et al., 2015 ; Sakellariou et al., 2022 ). By translating probabilistic model outputs into spatially interpretable risk representations, the study enhances the practical applicability of predictive modeling for geographically adaptive wildfire prevention and management, a priority frequently emphasized in international disaster risk reduction literature (Mastrorillo et al., 2024 ). 5.3. Implications and Limitations In comparison with previous wildfire risk assessments (Li et al., 2025 ; Zhang et al., 2022 ), the principal novelty of this study lies in its integration of multiple analytical techniques within a unified framework and its dedicated focus on the underexamined MCIZs—landscapes characterized by both high ignition potential and complex governance constraints (Murillo-Sandoval et al., 2018 ; Brady, 2021 ; Park et al., 2019 ). The integration of RF, GAM, and GWR enables the simultaneous consideration of predictive accuracy, nonlinear threshold detection, and spatial heterogeneity, thereby providing a more comprehensive understanding of ignition risk than single-model approaches. The findings highlight the need for differentiated wildfire prevention strategies that reflect the spatial and environmental variability of ignition drivers. For instance, meteorological factors such as relative humidity and wind speed exhibited sharp nonlinear thresholds associated with peak ignition probabilities, suggesting the necessity of localized early warning protocols under these critical conditions. Moreover, the spatially heterogeneous effects of military proximity underscore the importance of managing ignition risk through both institutional coordination and spatially adaptive monitoring, particularly in access-restricted or operationally sensitive zones. Despite these advances, several limitations remain. The temporal resolution of the meteorological datasets may be insufficient to capture short-term fluctuations that influence ignition dynamics, and the transferability of the integrated modeling framework to other geographic or political contexts remains to be empirically validated through cross-regional applications. 6. Conclusion This study advances wildfire ignition risk assessment in the Korean DMZ and adjacent MCIZs by integrating RF, GAM, and GWR into a unified analytical framework. The integrated approach effectively identified the principal ignition drivers while capturing the nonlinear thresholds and spatial heterogeneity essential for predictive accuracy in access-restricted environments. By linking high-resolution predictive modeling with an integrated modeling framework, the study provides a scientifically rigorous and practically applicable approach for wildfire risk assessment in politically sensitive border regions and offers a transferable foundation for similar assessments in other restricted or data-scarce landscapes worldwide. Declarations Conflict of Interest Statement The authors declare that there is no conflict of interest regarding the publication of this manuscript. Data Access Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. Ethics Statement This study does not involve human participants or animals and therefore did not require ethics approval. All data used were publicly available and secondary in nature. Author Contribution Sujung Heo conceptualized the study, developed the methodology, conducted the data analysis, and drafted the manuscript.Sujung Ahn supervised the research and contributed to study design, policy interpretation, and manuscript revision as the corresponding author.Song Hee Han, Sung Eun Cha, and Mi Na Jang contributed to data curation, validation, and visualization.Hyunsu Kim and Sung Cheol Jung assisted in spatial data preprocessing and statistical verification.Minjeong Heo contributed to the literature review and editing.Junsoo Kim provided technical advice on spatial modeling and international case comparison.All authors discussed the results and reviewed the final manuscript. Acknowledgement This research was supported by the National Institute of Forest Science (Project No. FE0500-2025-02-2025). References Abbate, A., Longoni, L., Ivanov, V. I., & Papini, M. (2019). Wildfire impacts on slope stability triggering in mountain areas. Geosciences, 9(10), 417. Andrianarivony, H. S., & Akhloufi, M. A. (2024). Machine learning and deep learning for wildfire spread prediction: A review. Fire, 7(12), 482. Ángel Javaloyes, M., Pendás-Recondo, E., & Sánchez, M. (2023). A general model for wildfire propagation with wind and slope. SIAM Journal on Applied Algebra and Geometry, 7(2), 414–439. Anton, C. E., & Lawrence, C. (2016). Does place attachment predict wildfire mitigation and preparedness? A comparison of wildland–urban interface and rural communities. Environmental management, 57(1), 148–162. Asori, M., Emmanuel, D., & Dumedah, G. (2020). Wildfire hazard and risk modelling in the northern regions of Ghana using GIS-based multi-criteria decision making analysis. Journal of Environment and Earth Science, 10(11). Bhowmik, R. T., Jung, Y. S., Aguilera, J. A., Prunicki, M., & Nadeau, K. (2023). A multi-modal wildfire prediction and early-warning system based on a novel machine learning framework. Journal of environmental management, 341, 117908. Bisenic, N. (2022). A Scenario Based Fire Susceptibility Approach for Remote Sensing Platform Comparison: Los Angeles County Area, Southern California, USA (Doctoral dissertation, University of Southern California). Bowman, D. M., Williamson, G. J., Abatzoglou, J. T., Kolden, C. A., Cochrane, M. A., & Smith, A. M. (2017). Human exposure and sensitivity to globally extreme wildfire events. Nature ecology & evolution, 1(3), 0058. Brady, L. M. (2021). From war zone to biosphere reserve: the Korean DMZ as a Scientific Landscape. Notes and Records, 75(2), 189–205. Cao, L., Elliot, W., & Long, J. W. (2021). Spatial simulation of forest road effects on hydrology and soil erosion after a wildfire. Hydrological Processes, 35(6), e14139. Cao, Q. (2020). Exploring Spatially Varying Relationships between Forest Fire and Environmental Factors in Fujian, China (Doctoral dissertation, State University of New York College of Environmental Science and Forestry). Chen, B., Wu, S., Jin, Y., Song, Y., Wu, C., Venevsky, S., … Gong, P. (2024). Wildfire risk for global wildland–urban interface areas. Nature Sustainability, 7(4), 474–484. Coskuner, K. A. (2022). Assessing the performance of MODIS and VIIRS active fire products in the monitoring of wildfires: a case study in Turkey. iForest-Biogeosciences and Forestry, 15(2), 85. Costa-Saura, J. M., Bacciu, V., Ribotta, C., Spano, D., Massaiu, A., & Sirca, C. (2022). Predicting and mapping potential fire severity for risk analysis at regional level using Google Earth engine. Remote Sensing, 14(19), 4812. Coutaz, G. (2018). South Korea: Vulnerability and Adaptation to Natural Disasters. In Coping with Disaster Risk Management in Northeast Asia: Economic and Financial Preparedness in China, Taiwan, Japan and South Korea (pp. 111–132). Emerald Publishing Limited. Detmer, A. R., Ward, E. J., Hunsicker, M. E., Andrews, K. S., Conrad, M., Ferriss, B. E., … Samhouri, J. F. (2025). Evaluating the robustness of generalized additive models as a tool for threshold detection in variable environments. Ecosphere, 16(3), e70117. Dillon, G., Menakis, J., & Fay, F. (2015). Wildland fire potential: A tool for assessing wildfire risk and fuels management needs. In In: Keane, Robert E.; Jolly, Matt; Parsons, Russell; Riley, Karin. Proceedings of the large wildland fires conference; May 19–23, 2014; Missoula, MT. Proc. RMRS-P-73. Fort Collins, CO: US Department of Agriculture, Forest Service, Rocky Mountain Research Station. p. 60–76. (Vol. 73, pp. 60–76). Donovan, V. M., Wonkka, C. L., Wedin, D. A., & Twidwell, D. (2020). Land-use type as a driver of large wildfire occurrence in the US Great Plains. Remote Sensing, 12(11), 1869. Fang, L., Yang, J., White, M., & Liu, Z. (2018). Predicting potential fire severity using vegetation, topography and surface moisture availability in a Eurasian boreal forest landscape. Forests, 9(3), 130. FIRMS. (2025). Fire Information for Resource Management System. https://firms.modaps.eosdis.nasa.gov/ . Heo, S., & Lee, D. K. (2025). Assessing disasters in East Kalimantan: machine learning approaches for sustainable urban development. Environmental Research Letters, 20(10), 104003. Heo, S., Park, S., & Lee, D. K. (2023). Multi-hazard exposure mapping under climate crisis using random forest algorithm for the Kalimantan Islands, Indonesia. Scientific Reports, 13(1), 13472. Heo, S., Sohn, W., Park, S., & Lee, D. K. (2024). Multi-hazard assessment for flood and Landslide risk in Kalimantan and Sumatra: Implications for Nusantara, Indonesia's new capital. Heliyon, 10(18). Kim, H., Seo, H., Kim, S., Kim, H., & Ko, M. (2022). Fish diversity and ichthyofauna of areas adjacent to the Demilitarized Zone in South Korea. Diversity, 14(12), 1011. Kim, J. H., Park, S., Kim, S. H., & Lee, E. J. (2021). Long-Term Land Cover Changes in the Western Part of the Korean Demilitarized Zone. Land, 10(7), 708. https://doi.org/10.3390/land10070708 Kim, J., & Lee, S. O. (2024). Fragmented, Materialized, Militarized Geopolitics of Wildfires in the Inter-Korean Border. Geoforum, 155, 104077. Kim, K. G. (2013). History and Evolution of the DMZ Ecosystem. In The Demilitarized Zone (DMZ) of Korea: Protection, Conservation and Restoration of a Unique Ecosystem (pp. 1–28). Berlin, Heidelberg: Springer Berlin Heidelberg. Korea Forest Service. (2025). https://www.forest.go.kr/kfsweb/kfi/kfs/frfr/selectFrfrStatsArea.do Laushman, K. M., Munson, S. M., & Villarreal, M. L. (2020). Wildfire risk and hazardous fuel reduction treatments along the US-Mexico border: a review of the science (1986–2019). Air, Soil and Water Research, 13, 1178622120950272. Lee, S., Song, J. S., Lee, C. W., & Ko, B. (2022). The Study of DMZ Wildfire Damage Area Detection Method Using Sentinel-2 Satellite Images. Korean Journal of Remote Sensing, 38(5_1), 545–557. Li, C., Su, Z., Ni, R., Wang, G., Ouyang, Y., Zeng, A., & Guo, F. (2025). Integrated spatial generalized additive modeling for forest fire prediction: a case study in Fujian Province, China. Journal of Forestry Research, 36(1), 30. Malik, A., Rao, M. R., Puppala, N., Koouri, P., Thota, V. A. K., Liu, Q., … Gao, J. (2021). Data-driven wildfire risk prediction in northern California. Atmosphere, 12(1), 109. Marcos, B., Gonçalves, J., Alcaraz-Segura, D., Cunha, M., & Honrado, J. P. (2021). A framework for multi-dimensional assessment of wildfire disturbance severity from remotely sensed ecosystem functioning attributes. Remote Sensing, 13(4), 780. Mastrorillo, M., Scartozzi, C. M., Pacillo, G., Menza, G., Desai, B., Maviza, G., … de Dinechin, F. (2024). Towards a common vision for climate change, security and migration in the Mediterranean. Montoya Alvis, J., Orozco Mendoza, G. L., & Zartha Sossa, J. W. (2025). Extreme Fire Events in Wildland–Urban Interface Areas: A Review of the Literature Concerning Determinants for Risk Governance. Sustainability, 17(10), 4505. Moreira, F., Ascoli, D., Safford, H., Adams, M. A., Moreno, J. M., Pereira, J. M., … Fernandes, P. M. (2020). Wildfire management in Mediterranean-type regions: paradigm change needed. Environmental Research Letters, 15(1), 011001. Murillo-Sandoval, P. J., Hilker, T., Krawchuk, M. A., & Van Den Hoek, J. (2018). Detecting and attributing drivers of forest disturbance in the Colombian andes using landsat time-series. Forests, 9(5), 269. Murray, A. T., Baik, J., Figueroa, V. E., Rini, D., Moritz, M. A., Roberts, D. A., … Jones, C. (2023). Developing effective wildfire risk mitigation plans for the wildland urban interface. International Journal of Applied Earth Observation and Geoinformation, 124, 103531. Naser, M. Z., & Kodur, V. (2025). Vulnerability of structures and infrastructure to wildfires: a perspective into assessment and mitigation strategies. Natural Hazards, 1–21. Nasiri, V., Sadeghi, S. M. M., Bagherabadi, R., Moradi, F., Deljouei, A., & Borz, S. A. (2022). Modeling wildfire risk in western Iran based on the integration of AHP and GIS. Environmental monitoring and assessment, 194(9), 644. Nikolakis, W., & Roberts, E. (2022). Wildfire governance in a changing world: Insights for policy learning and policy transfer. Risk, Hazards & Crisis in Public Policy, 13(2), 144–164. Pahlavani, P., Raei, A., Bigdeli, B., & Ghorbanzadeh, O. (2024). Identifying influential spatial drivers of forest fires through geographically and temporally weighted regression coupled with a continuous invasive weed optimization algorithm. Fire, 7(1), 33. Papathoma-Köhle, M., Schlögl, M., Garlichs, C., Diakakis, M., Mavroulis, S., & Fuchs, S. (2022). A wildfire vulnerability index for buildings. Scientific Reports, 12(1), 6378. Park, J., Sim, W., Park, J., & Lee, J. (2019). Object-based Land Cover Change Detection and Landscape Structure Analysis of Demilitarized Zone in Korea. Sensors & Materials, 31. Punzo, G., Castellano, R., & Bruno, E. (2022). Using geographically weighted regressions to explore spatial heterogeneity of land use influencing factors in Campania (Southern Italy). Land Use Policy, 112, 105853. Sagrario, M. D. L. Á. G., Musazzi, S., Córdoba, F. E., Mendiolar, M., & Lami, A. (2020). Inferring the occurrence of regime shifts in a shallow lake during the last 250 years based on multiple indicators. Ecological Indicators, 117, 106536. Sakellariou, S., Sfougaris, A., Christopoulou, O., & Tampekis, S. (2022). Integrated wildfire risk assessment of natural and anthropogenic ecosystems based on simulation modeling and remotely sensed data fusion. International Journal of Disaster Risk Reduction, 78, 103129. Salavati, G., Saniei, E., Ghaderpour, E., & Hassan, Q. K. (2022). Wildfire risk forecasting using weights of evidence and statistical index models. Sustainability, 14(7), 3881. Schag, G. M., Stow, D. A., Riggan, P. J., & Nara, A. (2022). Spatial-statistical analysis of landscape-level wildfire rate of spread. Remote Sensing, 14(16), 3980. Sutherland, D., Rashid, M. A., Hilton, J. E., & Moinuddin, K. A. (2023). Implementation of spatially-varying wind adjustment factor for wildfire simulations. Environmental Modelling & Software, 163, 105660. Thapa, S. B. (2024). Integrated wildfire risk management: Measuring risk perceptions, simulating fire severity maps, and visualizing fire risk in the California Wildland-Urban Interface (Doctoral dissertation, University of California, Merced). Wang, S. S. C., & Wang, Y. (2020). Quantifying the effects of environmental factors on wildfire burned area in the south central US using integrated machine learning techniques. Atmospheric Chemistry and Physics, 20(18), 11065–11087. Wu, Z., He, H. S., Fang, L., Liang, Y., & Parsons, R. A. (2018). Wind speed and relative humidity influence spatial patterns of burn severity in boreal forests of northeastern China. Annals of Forest Science, 75(3), 66. Ye, T., Liu, W., Mu, Q., Zong, S., Li, Y., & Shi, P. (2020). Quantifying livestock vulnerability to snow disasters in the Tibetan Plateau: Comparing different modeling techniques for prediction. International Journal of Disaster Risk Reduction, 48, 101578. Ying, L., Cheng, H., Shen, Z., Guan, P., Luo, C., & Peng, X. (2021). Relative humidity and agricultural activities dominate wildfire ignitions in Yunnan, Southwest China: Patterns, thresholds, and implications. Agricultural and Forest Meteorology, 307, 108540. Zhang, Z., Yang, S., Wang, G., Wang, W., Xia, H., Sun, S., & Guo, F. (2022). Evaluation of geographically weighted logistic model and mixed effect model in forest fire prediction in northeast China. Frontiers in Forests and Global Change, 5, 1040408. Additional Declarations No competing interests reported. Supplementary Files Appendix.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7872076","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":536045535,"identity":"5f02cc7d-f09a-41f2-a50f-c1ede7390cd1","order_by":0,"name":"Sujung Heo","email":"","orcid":"","institution":"National Institute of Forest Science","correspondingAuthor":false,"prefix":"","firstName":"Sujung","middleName":"","lastName":"Heo","suffix":""},{"id":536045537,"identity":"c524d279-4678-46f2-a063-16b26f53d731","order_by":1,"name":"Sujung Ahn","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1UlEQVRIiWNgGAWjYDACCRDBA8TsbUgCxGnhOUaSFjAjjUgt/LN7D3/mkTmcJx/5LPFxRYUdg+TsAwQsuXMuTZqH53Cx4e20w4ZnziQzSPMlELDmRo4ZM1BL4sbZ6W2SjW3MDHI8BHTI38gx/gzWMvM4UMu/esJaDG7kGIAcljhfgu2YZGPDYQZpQloMgQ6TnMOTnriBJy3ZsOHYcR7JHgJa5IAO+/C2xzpxfvsxw4cNNdVyEmcIaAEDRqDBBgcgbELOgoEfwHBoIFLtKBgFo2AUjDwAAJzgPsIkLo0lAAAAAElFTkSuQmCC","orcid":"","institution":"National Institute of Forest Science","correspondingAuthor":true,"prefix":"","firstName":"Sujung","middleName":"","lastName":"Ahn","suffix":""},{"id":536045538,"identity":"962a1baf-8ffd-4f76-ab3f-b091a5155ebd","order_by":2,"name":"Song Hee Han","email":"","orcid":"","institution":"National Institute of Forest Science","correspondingAuthor":false,"prefix":"","firstName":"Song","middleName":"Hee","lastName":"Han","suffix":""},{"id":536045540,"identity":"205ae14e-8ec3-44f2-b45f-41b5fb4a222f","order_by":3,"name":"Sung Eun Cha","email":"","orcid":"","institution":"National Institute of Forest Science","correspondingAuthor":false,"prefix":"","firstName":"Sung","middleName":"Eun","lastName":"Cha","suffix":""},{"id":536045541,"identity":"235c8cfe-467d-46d0-8c74-5dd21569ce61","order_by":4,"name":"Mi Na Jang","email":"","orcid":"","institution":"National Institute of Forest Science","correspondingAuthor":false,"prefix":"","firstName":"Mi","middleName":"Na","lastName":"Jang","suffix":""},{"id":536045542,"identity":"76b34adc-78e8-407d-9019-9afc2668a059","order_by":5,"name":"Hyunsu Kim","email":"","orcid":"","institution":"National Institute of Forest Science","correspondingAuthor":false,"prefix":"","firstName":"Hyunsu","middleName":"","lastName":"Kim","suffix":""},{"id":536045543,"identity":"e0a6b431-c5db-4456-8ac2-a69ae1273639","order_by":6,"name":"Sung Cheol Jung","email":"","orcid":"","institution":"National Institute of Forest Science","correspondingAuthor":false,"prefix":"","firstName":"Sung","middleName":"Cheol","lastName":"Jung","suffix":""},{"id":536045544,"identity":"a8b27715-4ea7-4be2-8353-d1b10596c543","order_by":7,"name":"Minjeong Heo","email":"","orcid":"","institution":"Korea University","correspondingAuthor":false,"prefix":"","firstName":"Minjeong","middleName":"","lastName":"Heo","suffix":""},{"id":536045545,"identity":"7410953b-8eef-47a4-96ba-59a92ea0192f","order_by":8,"name":"Junsoo Kim","email":"","orcid":"","institution":"Martin Luther University Halle-Wittenberg","correspondingAuthor":false,"prefix":"","firstName":"Junsoo","middleName":"","lastName":"Kim","suffix":""}],"badges":[],"createdAt":"2025-10-16 00:38:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7872076/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7872076/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":94783025,"identity":"258a2b03-9fdf-4f53-a263-44706a69075e","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7363797,"visible":true,"origin":"","legend":"","description":"","filename":"MSSJHEO251016.docx","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/e713a83a826545e59bfc8bb8.docx"},{"id":94824299,"identity":"80308e0e-ddf7-44e4-9b2e-7de1b9b9e287","added_by":"auto","created_at":"2025-10-31 06:48:47","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10169,"visible":true,"origin":"","legend":"","description":"","filename":"13bd9916ece543eb9e5b90e685bc30c5.json","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/cddba7afd9cbf6c7e28cf306.json"},{"id":94783026,"identity":"06745b5d-1396-4687-bce1-298388280bca","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":142338,"visible":true,"origin":"","legend":"","description":"","filename":"13bd9916ece543eb9e5b90e685bc30c51enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/aec1028ffda937c88f02550b.xml"},{"id":94825283,"identity":"485d1812-7101-4404-aeaa-8c49a862e549","added_by":"auto","created_at":"2025-10-31 06:50:04","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":497499,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/f04c3ff580bb1652a2cd328f.png"},{"id":94824718,"identity":"ac660588-64ec-42b4-9e85-9ce34ec8c2d4","added_by":"auto","created_at":"2025-10-31 06:49:14","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":705891,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/28449eef8e5ddc3bb5f557e1.jpeg"},{"id":94783029,"identity":"acdbc602-bda1-4a15-baab-f23cc271dddb","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":348898,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/bc3f95bed2ddab107ec0d245.png"},{"id":94783033,"identity":"cfd76829-94e9-40e7-88cd-a380a8d2de2a","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":163567,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/6212cc0361ad03da2726b246.png"},{"id":94783031,"identity":"b168658b-8290-4bcd-834a-c74e5b3c6374","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":82226,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/d158bf281eb504bc3a30136e.png"},{"id":94825793,"identity":"4ace403f-2c3d-4afa-8e34-5f14b1227eee","added_by":"auto","created_at":"2025-10-31 06:50:45","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":220126,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/d1c16522cb030d5fac03868a.png"},{"id":94783040,"identity":"13cd46c3-768d-4be1-81d6-8949359cb2ca","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2319926,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/86b681d995d94b571e3f300e.png"},{"id":94825529,"identity":"b4a68b21-ffde-48ff-a53b-bf6a1161e7e6","added_by":"auto","created_at":"2025-10-31 06:50:23","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1098773,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/9bad9d27da5f10af74c0f7f1.jpeg"},{"id":94783035,"identity":"0c393e8b-8dea-478a-8e67-aaf176541cbe","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"jpeg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":426238,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/6e556d86089c72584fbb78f3.jpeg"},{"id":94783036,"identity":"e04aafe3-5393-4782-b63a-1bf07c529a0f","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"jpeg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1703537,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/4afe9f1e9320fd2a453a9a19.jpeg"},{"id":94824446,"identity":"fddbb044-7b2a-402b-9f36-692c9d5e29ad","added_by":"auto","created_at":"2025-10-31 06:49:00","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":91604,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/1ca07fd3892c6aa8d3316694.png"},{"id":94824900,"identity":"299dd2e3-dafc-4f40-8177-11ef94d1c0a5","added_by":"auto","created_at":"2025-10-31 06:49:32","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":186636,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/b5cc3ee322eacaff0776b0e5.png"},{"id":94783039,"identity":"0c6262c2-90f2-4097-a036-783c6e0e3b30","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":65904,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/0a985a84f5d98401a1e7b722.png"},{"id":94783032,"identity":"8f25d310-d2b0-405d-b3a0-991549506e78","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":31655,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/89a49baf5d27cf8550a689df.png"},{"id":94825606,"identity":"4997c6d2-5b6e-40ed-81a2-e58e80a372a3","added_by":"auto","created_at":"2025-10-31 06:50:29","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17733,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/6c76a3a281be573bd7a571f2.png"},{"id":94824765,"identity":"3743ebe2-1fdc-4449-8d89-8bd764d439d5","added_by":"auto","created_at":"2025-10-31 06:49:17","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":57332,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/9e757dd2873b2bcad8299e56.png"},{"id":94783045,"identity":"fc90ea44-9945-4c35-98cd-4315856f8b40","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":272647,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/ae4cae5fcd1c4c89a36929f4.png"},{"id":94783048,"identity":"27dd7bdc-805a-4e92-a7e2-589e5f3fb642","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":229466,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/ffc6f06535ec98766cc902a3.png"},{"id":94783050,"identity":"41d47bc4-33d1-4ce8-bfe6-6c67571193f4","added_by":"auto","created_at":"2025-10-30 16:06:03","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":90539,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/e1d61ca08067345f40468815.png"},{"id":94783044,"identity":"756ae898-1461-4653-a47f-22e07bc32f61","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":484502,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/767dba2c69a6c088ec48f38d.png"},{"id":94783047,"identity":"93ccabce-3355-4dbf-94da-277f9cfa0977","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"xml","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":140840,"visible":true,"origin":"","legend":"","description":"","filename":"13bd9916ece543eb9e5b90e685bc30c51structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/bb8c23513abce1abf38764cd.xml"},{"id":94825305,"identity":"9ea961f8-95cc-4bf0-872c-899a171f2d71","added_by":"auto","created_at":"2025-10-31 06:50:06","extension":"html","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":148694,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/3901d7268655c74cb60fcdfc.html"},{"id":94825113,"identity":"2e198f03-d180-4838-aa25-4f610c5fda93","added_by":"auto","created_at":"2025-10-31 06:49:53","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":228986,"visible":true,"origin":"","legend":"\u003cp\u003eLocation of the study area and spatial distribution of wildfires (2001–2024)\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/0d498dd06d4eaf48637cb9b1.png"},{"id":94783020,"identity":"c6da8f4d-1f71-4d22-8395-65737e21f6ab","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":447613,"visible":true,"origin":"","legend":"\u003cp\u003eStudy framework.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/763d011e828a043a2852c07d.png"},{"id":94783018,"identity":"eb952e4b-a104-4c40-a003-7d02b87a86be","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":90059,"visible":true,"origin":"","legend":"\u003cp\u003eFinal set of 11 variables selected for wildfire risk modeling. (RH: Min Relative Humidity; WS: Max Wind Speed, ASP: Slope Aspect, SLO: Slope Degree, TWI: Topographic Wetness Index, DEM: Digital Elevation Model, Military: Distance to Live-Fire Exercises, BUILDING: Distance to Building Infrastructure, LULC: Land Use and Land Cover, ROAD: Distance to Road, EVI: Enhanced Vegetation Index)\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/ea5c8dd852909eb431c8f5d3.png"},{"id":94824894,"identity":"7f374b64-21be-48c2-84f9-84bf818085f4","added_by":"auto","created_at":"2025-10-31 06:49:32","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":69252,"visible":true,"origin":"","legend":"\u003cp\u003eGini feature importance ranked by mean AUC drop. (RH: Min Relative Humidity; WS: Max Wind Speed, ASP: Slope Aspect, SLO: Slope Degree, TWI: Topographic Wetness Index, DEM: Digital Elevation Model, Military: Distance to Live-Fire Exercises, BUILDING: Distance to Building Infrastructure, LULC: Land Use and Land Cover, ROAD: Distance to Road, EVI: Enhanced Vegetation Index)\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/2fef5c619aada73456f23313.png"},{"id":94825314,"identity":"99520cfc-b823-4271-9b7a-5309f18bcad9","added_by":"auto","created_at":"2025-10-31 06:50:06","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":205929,"visible":true,"origin":"","legend":"\u003cp\u003ePartial dependence plots of selected environmental drivers affecting wildfire occurrence, as estimated by GAM. (RH: Min Relative Humidity; WS: Max Wind Speed, ASP: Slope Aspect, SLO: Slope Degree, TWI: Topographic Wetness Index, DEM: Digital Elevation Model, Military: Distance to Live-Fire Exercises, BUILDING: Distance to Building Infrastructure, LULC: Land Use and Land Cover, ROAD: Distance to Road, EVI: Enhanced Vegetation Index)\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/adcc8cacfd358f92d5022e76.png"},{"id":94783024,"identity":"8baeef73-9550-416d-add7-438094b84169","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1774422,"visible":true,"origin":"","legend":"\u003cp\u003eSpatial distribution of wildfire risk prediction and localized high-risk zones based on RF model. (A) Predicted wildfire probability map across the study area using RF model, classified into five risk levels using the Natural Jenks method: Very low, Low, Moderate, High, and Very high. (B) Delineation of high-risk (High + Very high) zones extracted from the full probability map. (B-1) Zoomed-in view of a high-risk area in the northern region. (B-2) Zoomed-in view of a high-risk area near the central-southern boundary.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/e2c2a8a4e4d7f4a144ee83fa.png"},{"id":94825104,"identity":"d79a67f4-c2e7-4202-a1d0-1842d32ae36d","added_by":"auto","created_at":"2025-10-31 06:49:51","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":2764911,"visible":true,"origin":"","legend":"\u003cp\u003eSpatial distribution of local coefficients estimated by the GWR model for key wildfire predictors.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/1d52b1c8d6ab02900f9a2e80.png"},{"id":100365648,"identity":"eea434d9-4ee0-49c3-8ab7-14ea4e447e22","added_by":"auto","created_at":"2026-01-16 07:55:29","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6363055,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/e66925f4-35ef-4c84-9c7b-bbe56008a23f.pdf"},{"id":94783022,"identity":"1942e077-d51f-4090-b101-28058f3dc160","added_by":"auto","created_at":"2025-10-30 16:06:02","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2039184,"visible":true,"origin":"","legend":"","description":"","filename":"Appendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-7872076/v1/5035b58c713a69df716e2842.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Integrated Modeling of Wildfire Ignition Risk in the Military–Civilian Interface of the Korean DMZ","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eWildfires have increasingly emerged as a critical global threat, exacerbated by the intersection of climate change, land-use transformations, and the expansion of human activity into fire-prone landscapes (Bowman et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Moreira et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). While substantial advances have been made in wildfire modeling and mitigation, much of the research has focused on either densely populated wildland\u0026ndash;urban interface (WUI) zones or remote forested regions (Murray et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Chen et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Transitional areas such as military\u0026ndash;civilian interface zones (MCIZs), however, remain substantially underexamined despite their distinct geographies of risk, which are shaped by restricted access, concentrated military activity, and adjacent civilian development (Kim et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThis research gap is particularly pronounced in Republic of Korea (\u0026ldquo;Korea\u0026rdquo;), especially within the western Demilitarized Zone (DMZ) and Civilian Control Zone (CCZ), where live-fire military exercises, agricultural residue burning, and climate-sensitive vegetation dynamics converge to elevate wildfire ignition risk (Kim \u0026amp; Lee, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). From 2001 to 2024, national fire statistics and data from the Fire Information for Resource Management System (FIRMS) indicate that over 95.65% of wildfires in these regions are attributed to anthropogenic ignition sources, with military activities (56.52%) and civilian open burning (20.29%) representing the primary causes (Korea Forest Service, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; FIRMS, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Nevertheless, systematic spatial modeling of wildfire risk and rigorous evaluation of mitigations remain limited for these sensitive zones (Lee et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eRecent advancements in data science and spatial analysis, including machine learning, spatial regression, and geospatial modeling, provide promising avenues to capture the multi-dimensional and spatially heterogeneous nature of wildfire ignition risk (Marcos et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Bisenic, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Andrianarivony \u0026amp; Akhloufi, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Random Forest (RF) models have demonstrated strong performance in ignition risk prediction and variable selection (Heo et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), Generalized Additive Models (GAM) facilitate robust exploration of nonlinear thresholds and ignition probabilities (Detmer et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), while Geographically Weighted Regression (GWR) enables the assessment of spatial heterogeneity across complex landscapes (Punzo et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). However, few studies have integrated these methodologies to jointly address predictive accuracy, interpretability, and spatial non-stationarity\u0026mdash;particularly within the context of access-restricted military landscapes (Kim et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Kim \u0026amp; Lee, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eTo address these gaps, this study proposes a spatially explicit wildfire ignition risk assessment framework specifically tailored to MCIZs. By integrating RF, GAM, and GWR, the framework systematically identifies the principal climatic, topographic, anthropogenic, and military drivers of ignition, quantifies their nonlinear thresholds, and evaluates their spatial heterogeneity across the DMZ\u0026ndash;CCZ landscape. Building on these complementary outputs, an integrated modeling framework was developed to synthesize probabilistic model results into spatially interpretable ignition susceptibility maps, providing a unified representation of wildfire risk patterns. The research is guided by three key questions: (1) Which factors exert the strongest influence on wildfire ignition in MCIZs? (2) Under what threshold conditions (e.g., low relative humidity, high wind speeds, or proximity to military training areas) does ignition risk sharply increase? (3) How do these effects vary geographically, revealing spatial patterns of ignition vulnerability?\u003c/p\u003e"},{"header":"2. Study area","content":"\u003cp\u003eThe present study focuses on the northern region of Paju City, Korea, which is characterized by its direct proximity to the DMZ and the CCZ (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). This locale constitutes a complex socio-ecological frontier, where zones of military activity\u0026mdash;including installations, live-fire training grounds, and restricted buffer strips\u0026mdash;converge with areas dedicated to civilian settlements, agriculture, and transportation infrastructure. Following the Korean Armistice Agreement of 1953, the DMZ has remained under stringent military supervision with civilian access strictly prohibited, while the adjacent CCZ has permitted limited agricultural activity and residential development, resulting in divergent land use practices across relatively short distances (Kim et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe western sector of Paju\u0026rsquo;s CCZ, contiguous with the DMZ, exemplifies the dynamic interface between military infrastructure and rural livelihoods. The land use mosaic in this sector includes ginseng cultivation, traditional irrigation ponds (dumbeong), as well as critical habitats for migratory bird species along the Imjingang River (Lee et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Recent ecological assessments have documented that, within the DMZ, historical coniferous forests damaged during the Korean War have gradually transitioned to mature deciduous woodlands due to uninterrupted natural succession processes. In contrast, significant portions of the CCZ have been converted into agricultural or infrastructural developments due to ongoing human intervention (Kim \u0026amp; Lee., 2024).\u003c/p\u003e\u003cp\u003eFrom 2001 to 2024, a comprehensive geospatial analysis identified 318 wildfire ignition points within the study area. Of these, 78 incidents were directly attributed to military live-fire exercises according to data compiled from the Korea Forest Service and the NASA FIRMS archives (Korea Forest Service, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; NASA FIRMS, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). The high frequency and spatial clustering of these military-origin ignitions within both the DMZ and CCZ underscore the heightened risk of transboundary wildfires in heavily militarized border zones (Brady, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eDemographic trends throughout the 2020s reveal rapid population growth in Paju, fueled by expansion from the Seoul metropolitan fringe and government-driven regional development initiatives (Kim et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). This urbanization has intensified the exposure of communities situated along the wildland\u0026ndash;urban interface to wildfire hazards. Against this backdrop, the study area offers a salient case for examining wildfire risk in the context of intersecting military and civilian land uses, within a region that is simultaneously politically sensitive and ecologically significant (Kim et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"3. Methods","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e3.1. Study framework\u003c/h2\u003e\u003cp\u003eThis study introduces a spatially explicit framework for wildfire ignition risk assessment, specifically designed for MCIZs situated in restricted border areas. The framework integrates advanced machine learning and spatial regression approaches to comprehensively capture the interplay among climatic, topographic, anthropogenic, and institutional determinants of wildfire ignition in politically and ecologically sensitive contexts (see Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eWildfire occurrence and its underlying drivers were modeled using three complementary analytical techniques. First, the RF model was employed to generate spatially explicit ignition probability maps using a 70:30 training-to-testing data split for robust validation. Second, the GAM was applied to investigate nonlinear relationships and threshold effects among key covariates such as relative humidity, wind speed, and vegetation indices. Third, the GWR approach was used to assess the spatial heterogeneity in the influence of predictors, allowing localized interpretation of ignition drivers across the DMZ\u0026ndash;CCZ landscape. Variable selection was guided by Variance Inflation Factor (VIF) diagnostics, correlation matrices, and Random Forest (RF)-derived feature importance metrics, ensuring statistical rigor and model parsimony.\u003c/p\u003e\u003cp\u003eKey outputs of the framework include wildfire ignition probability maps, composite spatial layers derived from the integrated modeling framework, and GWR-based local coefficient surfaces. Together, these outputs provide high-resolution, spatially differentiated insights into ignition dynamics and support data-driven wildfire risk assessment and spatial decision-making in sensitive borderland environments.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e3.2. Variable Selection and Screening\u003c/h2\u003e\u003cp\u003eTo quantitatively assess wildfire risk in the DMZ-adjacent landscape of Paju, we identified 11 key environmental drivers following criteria of theoretical relevance, regional context, data accessibility, and empirical support (Fig. \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003eA1\u003c/span\u003e). The selected variables were systematically organized into four core categories\u0026mdash;climatic, topographic, land use/vegetation, and anthropogenic (civilian and military)\u0026mdash;to adequately capture both biophysical and human-induced factors implicated in fire ignition and spread (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). In a region characterized by a unique intersection of military activity, rural land use, and ecological succession, this multidimensional selection was essential for robust modeling.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eAll variables were processed to a spatial resolution of 10 m \u0026times; 10 m to ensure fine-scale representation of environmental heterogeneity, and matched to the wildfire inventory period (2001\u0026ndash;2024). Climatic factors, in particular, were represented by minimum relative humidity and maximum wind speed\u0026mdash;parameters selected to maximize the representation of conditions most conducive to ignition and rapid-fire spread. These data were sourced from the Korea Meteorological Administration and aligned with the study\u0026rsquo;s wildfire inventory period. Both relative humidity and wind speed are recognized as principal determinants of wildfire activity in fire-prone landscapes (Ying et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Sutherland et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), with historical evidence from the DMZ showing that low humidity combined with episodic gusts has triggered rapidly spreading fires, especially when interacting with military ordnance (Park et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eTopographic variables\u0026mdash;slope aspect, slope degree, elevation, and the topographic wetness index (TWI)\u0026mdash;were derived from high-resolution Digital Elevation Models (DEM) obtained via the SRTM 1 Arc-Second Global dataset. These layers were selected to reflect current terrain conditions as of 2024, ensuring accurate representation of slope-mediated microclimatic and hydrological influences on fire behavior (Malik et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Abbate et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Heo et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Topography critically mediates fuel continuity, solar radiation exposure, and soil moisture availability, all of which are important modulators of ignition probability and fire propagation in the complex terrain of the Korean borderlands.\u003c/p\u003e\u003cp\u003eLand use and vegetation conditions were characterized using Sentinel-2 (2024) land cover classification and the Enhanced Vegetation Index (EVI) derived from Landsat 8 (2001\u0026ndash;2024) imagery. The Sentinel-2 Land Use and Land Cover (LULC) layer provided an up-to-date depiction of land use patterns, while EVI\u0026mdash;favored over NDVI for its superior sensitivity to canopy structure and resistance to saturation in dense vegetation\u0026mdash;captured seasonal and interannual variation in vegetation greenness (Salavati et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Malik et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). These datasets were crucial for distinguishing between military training grounds, agricultural fields, and peri-urban residential areas, each exhibiting different fuel structures and ignition susceptibilities.\u003c/p\u003e\u003cp\u003eAnthropogenic variables included distance to the nearest road, distance to the nearest building, and proximity to live-fire military training zones, calculated from 2024 national infrastructure datasets (Ministry of Land, Infrastructure and Transport; National Geographic Information Institute) and Korea Forest Service military facility geospatial layers. Proximity to civilian infrastructure is associated with multiple ignition sources, including agricultural residue burning, recreational activities, and electrical faults (Heo et al., 2025; Naser \u0026amp; Kodur. 2025). In the study area, proximity to military training grounds has been documented as a major ignition driver, with incident records attributing approximately 24.5% of wildfires (78 out of 318 events) directly to military exercises and ordnance use (Kim \u0026amp; Lee. 2024; Lee et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eTo mitigate multicollinearity and ensure statistical validity, all candidate variables underwent VIF and pairwise Pearson correlation screening (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003e, Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Variables with VIF values below the conventional threshold of 5 were retained (max VIF\u0026thinsp;=\u0026thinsp;5.44 for \u0026ldquo;Distance to Road\u0026rdquo;; VIF\u0026thinsp;=\u0026thinsp;4.51 for \u0026ldquo;Distance to Building\u0026rdquo;), validating their inclusion. The correlation matrix indicated no severe interdependencies, with the highest correlation (r\u0026thinsp;=\u0026thinsp;0.60) observed between road and building proximity\u0026mdash;considered conceptually and spatially distinct within the regional context.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eSelected list of environmental drivers, data sources, period, and references used for fire risk modeling.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCategory\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFactor\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSource\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePeriod\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eReference\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eClimatic\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMin Relative Humidity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eKorea Meteorological Administration (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://data.kma.go.kr\u003c/span\u003e\u003cspan address=\"https://data.kma.go.kr\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003e2001\u0026ndash;2024\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYing et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2021\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMax Wind Speed\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eSutherland et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Heo et al., 2025\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003eTopographic\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSlope Aspect\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDEM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eHeo et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; \u0026Aacute;ngel et al., 2023\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDigital Elevation Model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSRTM 1 Arc-Second Global (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://earthexplorer.usgs.gov/\u003c/span\u003e\u003cspan address=\"https://earthexplorer.usgs.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2024\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMalik et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Salavati et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSlope Degree\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDEM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eAbbate et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Asori et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTopographic Wetness Index\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDEM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eFang et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Nasiri et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eLand Use and Vegetation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLand Use and Land Cover\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSentinal-2 (ESA via ESRI Platform)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2024\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eSalavati et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Donovan et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2020\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnhanced Vegetation Index\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLandsat 8 (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://earthexplorer.usgs.gov/\u003c/span\u003e\u003cspan address=\"https://earthexplorer.usgs.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2001\u0026ndash;2024\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMalik et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Costa-Saura et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eCivilian\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDistance to Road\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKorea Ministry of Land, Infrastructure and Transport (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.molit.go.kr/english/intro.do\u003c/span\u003e\u003cspan address=\"https://www.molit.go.kr/english/intro.do\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2024\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCao et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Heo et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDistance to Building Infrastructure\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNational Geographic Information Institute (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ngii.go.kr\u003c/span\u003e\u003cspan address=\"https://ngii.go.kr\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2024\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNaser \u0026amp; Kodur. 2025; Papathoma-K\u0026ouml;hle et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMilitary\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eProximity to Live-Fire Exercises\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKorea Forest Service (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://fqgis.forest.go.kr\u003c/span\u003e\u003cspan address=\"https://fqgis.forest.go.kr\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eKim \u0026amp; Lee. 2024; Lee et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eResults of VIF analysis for 11 targeted variables.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo.\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFeature\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eVIF\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRelative Humidity (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.10\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eWind Speed (m/s)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.43\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSlope Aspect\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.04\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSlope Degree\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.64\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTopographic Wetness Index\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.02\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eElevation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.82\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDistance to Live-Fire Exercises (m)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e3.24\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDistance to Building Infrastructure (m)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e4.51\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLand Use and Land Cover\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.14\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDistance to Road (m)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e5.44\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEnhanced Vegetation Index\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.47\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e3.3. Modeling Approaches\u003c/h2\u003e\u003cp\u003eTo characterize wildfire risk arising from military\u0026ndash;civilian interactions within the DMZ-adjacent region of Paju, we implemented a hybrid modeling framework that integrates three complementary analytical techniques: the GAM, RF, and GWR. Each approach was strategically selected to address specific dimensions of wildfire risk, encompassing global nonlinear patterns, spatial non-stationarity, and intricate interactions among environmental and anthropogenic drivers.\u003c/p\u003e\u003cp\u003eThe GAM is a flexible, semi-parametric regression framework designed to model potential non-linear associations between wildfire occurrence and predictor variables. GAM utilizes smooth spline functions to represent covariate effects, facilitating the detection of ecological thresholds and non-monotonic responses within ignition dynamics (Sagrario et al., 2025; Ye et al., \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). In our analysis, GAM was deployed to elucidate the influence of key variables\u0026mdash;including relative humidity, vegetation status, and distance to military training zones\u0026mdash;on wildfire probability. This approach is particularly suited to the DMZ\u0026ndash;CCZ context, where environmental gradients are modulated by restricted access, complex hydrological conditions, and episodic military disturbances.\u003c/p\u003e\u003cp\u003eThe GWR method accommodates spatial non-stationarity by allowing local variation in regression coefficients, thus capturing the possibility that the relationship between wildfire occurrence and its predictors may change across the heterogeneous landscape (Pahlavani et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Schag et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In the context of the Paju border region, GWR is essential for disentangling spatially-varying risk patterns, such as those found in agricultural zones versus areas adjacent to military infrastructure. Through the estimation of location-specific parameter estimates, GWR offers critical insights for the design of place-based fire management and risk mitigation strategies.\u003c/p\u003e\u003cp\u003eThe RF algorithm is an ensemble-based machine learning technique that aggregates multiple decision trees built on bootstrapped samples of the data. Owing to its robustness to noisy inputs, multicollinearity, and its capacity to handle high-dimensional, interacting predictors, RF is well-suited for modeling complex wildfire ignition processes (Heo et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In this study, RF was instrumental in quantifying the relative importance of each environmental and anthropogenic variable\u0026mdash;most notably, proximity to critical infrastructure and military facilities.\u003c/p\u003e\u003cp\u003eBy jointly employing GAM, RF, and GWR, we sought to balance interpretability, predictive performance, and spatial specificity. GAM enables global interpretability through partial dependence plots, highlighting nonlinear threshold effects of predictors such as wind speed or vegetation indices. RF complements this by modeling complex, higher-order interactions and generating robust variable importance rankings. GWR further enhances spatial fidelity by explicitly modeling spatial heterogeneities in predictor influence, a feature essential for nuanced fire risk governance in a transboundary, militarized setting (Pahlavani et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Cao., 2020).\u003c/p\u003e\u003cp\u003eAll models were independently trained on a commong set of pre-screened variables defined in Section 3.2. Model performance was rigorously assessed using standard evaluation metrics: F1-score, Area Under the ROC Curve (AUC) and Root Mean Squared Error (RMSE). These metrics offer a comprehensive appraisal of both classification accuracy and probabilistic modeling, and align with established practices in recent wildfire risk research (Heo et al., 2025; Wang. \u0026amp; Wang., 2020).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Results","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\n\u003ch2\u003e4.1. Model Validation and Performance Assessment\u003c/h2\u003e\n\u003cp\u003eThe predictive performance of the RF, GAM, and GWR approaches was assessed using a comprehensive set of classification and regression metrics (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). Evaluation criteria included overall accuracy, AUC, RMSE, as well as class-specific precision, recall, and F1-score statistics.\u003c/p\u003e\n\u003cp\u003eBoth RF and GWR models achieved the highest overall accuracy at 0.86, while the GAM model registered a value of 0.82. RF and GWR also shared the highest AUC values at 0.91, indicating strong model performance in distinguishing between wildfire and non-wildfire events. The GAM yielded an AUC of 0.86.\u003c/p\u003e\n\u003cp\u003eClass-specific performance for wildfire occurrence (Class 1) revealed the highest F1-score with RF (0.67), followed by GWR (0.57) and GAM (0.44). Class 0 (no fire) predictions were robust across all models, with F1-scores above 0.90 and the highest achieved by GWR (0.92).\u003c/p\u003e\n\u003cp\u003eThese validation outcomes demonstrate that all three modeling approaches performed satisfactorily based on the selected evaluation metrics, with RF and GWR exhibiting consistently strong results in both classification and regression-based assessments.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003ctable id=\"Tab3\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eRF, GAM and GWR Model performance metrics for wildfire occurrence prediction.\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eMetric\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eRF\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eGAM\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eGWR\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAccuracy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.86\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.82\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.86\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAUC Score\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.91\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.86\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.91\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRMSE\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.41\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.34\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.31\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePrecision (Class 0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.87\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.85\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.87\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePrecision (Class 1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.81\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.56\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.77\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRecall (Class 0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.95\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.93\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.97\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRecall (Class 1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.57\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.36\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.45\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eF1-Score (Class 0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.91\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.89\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.92\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eF1-Score (Class 1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.67\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.44\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.57\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eClass 0: No wildfire occurrence; Class 1: Wildfire occurrence; AUC Score is based on the ROC curve.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\n\u003ch2\u003e4.2. Nonlinear Effects of Environmental Drivers\u003c/h2\u003e\n\u003cp\u003eThe GAM identified minimum relative humidity (RH) and maximum wind speed (WS) as the most influential predictors of wildfire occurrence (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), with both variables displaying statistically significant nonlinear effects (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e). The partial dependence plot for RH exhibited a bell-shaped relationship, with the peak modeled fire probability between approximately 13.8% and 14.0%, followed by a pronounced decline outside this interval. WS demonstrated a non-monotonic association, with wildfire probability increasing up to around 13.5\u0026ndash;14.0 m/s before showing a decrease at elevated WS values. Slope aspect (ASP) showed modest variation, with amplified fire probabilities on west- to northwest-facing slopes. Slope degree (SLO) displayed a U-shaped relationship, with augmented probabilities at both low (\u0026lt;\u0026thinsp;5\u0026deg;) and steep (\u0026gt;\u0026thinsp;30\u0026deg;) gradients. The TWI exhibited a double-peak pattern, with increased probabilities in both low-TWI areas (approximately \u0026minus;\u0026thinsp;8) and high-TWI areas (above \u0026minus;\u0026thinsp;2), and reduced probabilities in the intermediate range. DEM indicated elevated fire probability in low-lying terrain (\u0026lt;\u0026thinsp;100 m), with a secondary increase above ~\u0026thinsp;1,000 m. For anthropogenic proximity variables, distance to military live-fire wildfires (MILITARY) followed an inverted-U pattern, peaking at intermediate distances of approximately 7\u0026ndash;8 km and decreasing at shorter and longer distances. Distance to buildings (BUILDING) showed amplified probabilities within 0\u0026ndash;2 km, declining steadily with increasing distance. Distance to roads (ROAD) exhibited a generally increasing trend, with a marked rise beyond 6 km. Among environmental variables, LULC was highly significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), with greater wildfire probabilities associated with rangeland, bare ground, and cropland, and lower probabilities for water bodies and flooded vegetation. EVI reached maximum fire probability at intermediate values (0.3\u0026ndash;0.4), with diminished probabilities for both sparse (\u0026lt;\u0026thinsp;0.2) and dense (\u0026gt;\u0026thinsp;0.6) vegetation cover.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n\u003ch2\u003e4.3. Spatial Wildfire Risk Prediction\u003c/h2\u003e\n\u003cp\u003eThe RF model demonstrated robust predictive capacity for wildfire ignition probability across the study area, revealing clear spatial patterns in risk distribution (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003eA). Continuous RF-derived probabilities were classified using the Natural Jenks method into five ordinal risk categories: very low (0\u0026ndash;0.05), low (0.05\u0026ndash;0.17), moderate (0.17\u0026ndash;0.32), high (0.32\u0026ndash;0.51), and very high (0.51\u0026ndash;1.00). Mapping of these classes indicated that wildfire risk is predominantly concentrated along the region\u0026rsquo;s western and northern margins, notably in zones adjacent to urban interfaces and major road infrastructure. Area statistics (Table \u003cspan class=\"InternalRef\"\u003eA1\u003c/span\u003e) showed that approximately 88.88% of the region was assigned to the very low-risk category, while high and very high-risk areas comprised 1.61% and 0.73% of the total area, respectively. Despite their limited spatial extent, these higher-risk zones corresponded closely with areas of intensified human\u0026ndash;environmental interactions. To visualize these priority areas, Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003eB highlights the aggregated high and very high-risk zones, and Figs.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003eB-\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e and \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003eB-\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e provide detailed views of representative hotspots.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n\u003ch2\u003e4.4. Spatial Heterogeneity of Wildfire Risk Drivers\u003c/h2\u003e\n\u003cp\u003eThe GWR model was employed to characterize the spatial variability in the effects of 11 key wildfire risk drivers. Summary statistics of the local coefficient estimates are presented in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e, and their geographic distributions are visualized in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eWS was associated with greater spatial variation in effect (range: \u0026minus;4.87 to 3.05; mean\u0026thinsp;=\u0026thinsp;0.02), although the mean effect was close to zero. Positive WS associations were prevalent in the northern and northeastern regions (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003eA).\u003c/p\u003e\n\u003cp\u003eAnalysis of the GWR outputs indicated pronounced spatial heterogeneity in the influence of both climatic and anthropogenic variables. RH exhibited the greatest spatial variability among all predictors, with local coefficients ranging from \u0026minus;\u0026thinsp;4.04 to 4.74 (mean\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.21). The majority of the study area\u0026mdash;particularly in central and northern sectors\u0026mdash;displayed a negative association between RH and fire occurrence, indicating increased ignition probability under low-humidity conditions (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003eB).\u003c/p\u003e\n\u003cp\u003eTopographic factors, including slope degree and aspect, showed relatively minor and stable coefficients near zero, suggesting minimal explanatory contribution to spatial ignition variation in this context. The TWI revealed consistently negative coefficients overall (mean\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.003), with the strongest effects observed in the southeastern region, aligning with areas of drier terrain (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003eF).\u003c/p\u003e\n\u003cp\u003eAnthropogenic features such as distance to live-fire exercises, buildings, and roads exhibited modest mean effects (mean coefficients: ~0.00005 to 0.0001). Notable spatial variation was evident near urban and semi-urban boundaries, especially in the northeast and central zones, where the relationship between human proximity and wildfire occurrence was stronger (Figs.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003eG\u0026ndash;I).\u003c/p\u003e\n\u003cp\u003eVegetation and land cover variables demonstrated greater spatial influence. The EVI showed a positive association with wildfire probability across most of the study area (mean\u0026thinsp;=\u0026thinsp;0.17; peak up to 0.85) (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003eK). Similarly, LULC had moderate positive coefficients (mean\u0026thinsp;=\u0026thinsp;0.007), with prominent effects along forest\u0026ndash;agriculture transitions in southern and central boundary regions (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003eJ).\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab5\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eSummary of GWR Parameter Estimate. (RH: Min Relative Humidity; WS: Max Wind Speed, ASP: Slope Aspect, SLO: Slope Degree, TWI: Topographic Wetness Index, DEM: Digital Elevation Model, Military: Distance to Live-Fire Exercises, BUILDING: Distance to Building Infrastructure, LULC: Land Use and Land Cover, ROAD: Distance to Road, EVI: Enhanced Vegetation Index)\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eVariable\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eMean\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eMedian\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eStd\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eMin\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eMax\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRH\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.21\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.23\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1.27\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-4.04\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e4.74\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eWS\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.02\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.21\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1.13\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-4.87\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e3.05\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eASP\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.0002\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.0001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.0004\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.0005\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSLO\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.00000001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.00000001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.00000004\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.00000012\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.00000007\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTWI\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.003\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.006\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.015\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.013\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDEM\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.002\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.003\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.006\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eMILITARY\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.00005\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.00003\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.0001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.0005\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.0004\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBUILDING\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.0001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.0001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.0001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.0003\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.0004\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eLULC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.007\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.005\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.009\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.014\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.034\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eROAD\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.00004\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.00004\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.0002\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.0004\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eEVI\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.17\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.18\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.22\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e-0.55\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.85\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003c/div\u003e"},{"header":"5. Discussion","content":"\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e5.1. Primary Drivers and Their Spatial Variation\u003c/h2\u003e\u003cp\u003eThe RF and GWR models exhibited strong predictive performance (AUC\u0026thinsp;=\u0026thinsp;0.91), highlighting proximity to military installations, low RH, elevated WS, and built infrastructure as the primary determinants of wildfire ignition. These findings corroborate previous research indicating that military training activities and urban\u0026ndash;rural anthropogenic factors jointly influence wildfire ignition across DMZ\u0026ndash;CCZ regions (Anton \u0026amp; Lawrence, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Laushman et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). The GAM analysis elucidated nonlinear hazard thresholds: RH demonstrated peak ignition probability within a narrow 13.8\u0026ndash;14.0% window, a condition likely optimizing fuel dryness and ignition success, while WS showed a similar peak at 13.5\u0026ndash;14.0 m/s, after which increased turbulence may suppress ignition stability (Richards et al., 2022). Analysis of spatial proximity revealed that ignition risk for military training zones followed an inverted-U shape, peaking at 7\u0026ndash;8 km\u0026mdash;closely aligned with typical operational training radii and potentially linked to detection or suppression response delays. Roads and buildings showed increased risk at 0\u0026ndash;2 km, reflecting human-caused ignition patterns observed in peri-urban settings (Laushman et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). GWR results further underscored spatial heterogeneity; military proximity had the strongest effect in eastern and northern sectors alongside meteorological effects that varied considerably across the study area, indicating localized operational contexts and environmental conditions drive ignition patterns (Kim \u0026amp; Lee, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Coutaz, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e5.2. Integrated Modeling for Spatial Precision and Policy Relevance\u003c/h2\u003e\u003cp\u003eBy integrating RF, GAM, and GWR, this study overcomes the constraints of single-model approaches, particularly relevant for regions with limited data access and complex socio-political overlays. RF was instrumental in providing robust variable ranking, GAM quantified critical nonlinear ignition thresholds for predictive interpretation, and GWR revealed micro-scale geographic variations that are essential for spatially precise wildfire risk assessment (Rodrigues \u0026amp; de la Riva, 2014; Wu et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The integration of these outputs within an integrated modeling framework enabled the development of weighted and spatially normalized ignition susceptibility maps \u0026mdash; a substantial methodological improvement over earlier wildfire risk assessments in the DMZ that relied primarily on coarse statistics or post-fire remote sensing (Lee et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Dillon et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Sakellariou et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). By translating probabilistic model outputs into spatially interpretable risk representations, the study enhances the practical applicability of predictive modeling for geographically adaptive wildfire prevention and management, a priority frequently emphasized in international disaster risk reduction literature (Mastrorillo et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003e5.3. Implications and Limitations\u003c/h2\u003e\u003cp\u003eIn comparison with previous wildfire risk assessments (Li et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), the principal novelty of this study lies in its integration of multiple analytical techniques within a unified framework and its dedicated focus on the underexamined MCIZs\u0026mdash;landscapes characterized by both high ignition potential and complex governance constraints (Murillo-Sandoval et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Brady, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Park et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). The integration of RF, GAM, and GWR enables the simultaneous consideration of predictive accuracy, nonlinear threshold detection, and spatial heterogeneity, thereby providing a more comprehensive understanding of ignition risk than single-model approaches.\u003c/p\u003e\u003cp\u003eThe findings highlight the need for differentiated wildfire prevention strategies that reflect the spatial and environmental variability of ignition drivers. For instance, meteorological factors such as relative humidity and wind speed exhibited sharp nonlinear thresholds associated with peak ignition probabilities, suggesting the necessity of localized early warning protocols under these critical conditions. Moreover, the spatially heterogeneous effects of military proximity underscore the importance of managing ignition risk through both institutional coordination and spatially adaptive monitoring, particularly in access-restricted or operationally sensitive zones.\u003c/p\u003e\u003cp\u003eDespite these advances, several limitations remain. The temporal resolution of the meteorological datasets may be insufficient to capture short-term fluctuations that influence ignition dynamics, and the transferability of the integrated modeling framework to other geographic or political contexts remains to be empirically validated through cross-regional applications.\u003c/p\u003e\u003c/div\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eThis study advances wildfire ignition risk assessment in the Korean DMZ and adjacent MCIZs by integrating RF, GAM, and GWR into a unified analytical framework. The integrated approach effectively identified the principal ignition drivers while capturing the nonlinear thresholds and spatial heterogeneity essential for predictive accuracy in access-restricted environments. By linking high-resolution predictive modeling with an integrated modeling framework, the study provides a scientifically rigorous and practically applicable approach for wildfire risk assessment in politically sensitive border regions and offers a transferable foundation for similar assessments in other restricted or data-scarce landscapes worldwide.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eConflict of Interest Statement\u003c/h2\u003e\u003cp\u003eThe authors declare that there is no conflict of interest regarding the publication of this manuscript.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003ch2\u003eData Access Statement\u003c/h2\u003e\u003cp\u003eThe data that support the findings of this study are available from the corresponding author upon reasonable request.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eEthics Statement\u003c/strong\u003e\u003cp\u003eThis study does not involve human participants or animals and therefore did not require ethics approval. All data used were publicly available and secondary in nature.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eSujung Heo conceptualized the study, developed the methodology, conducted the data analysis, and drafted the manuscript.Sujung Ahn supervised the research and contributed to study design, policy interpretation, and manuscript revision as the corresponding author.Song Hee Han, Sung Eun Cha, and Mi Na Jang contributed to data curation, validation, and visualization.Hyunsu Kim and Sung Cheol Jung assisted in spatial data preprocessing and statistical verification.Minjeong Heo contributed to the literature review and editing.Junsoo Kim provided technical advice on spatial modeling and international case comparison.All authors discussed the results and reviewed the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThis research was supported by the National Institute of Forest Science (Project No. FE0500-2025-02-2025).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAbbate, A., Longoni, L., Ivanov, V. I., \u0026amp; Papini, M. (2019). Wildfire impacts on slope stability triggering in mountain areas. Geosciences, 9(10), 417.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAndrianarivony, H. S., \u0026amp; Akhloufi, M. A. (2024). Machine learning and deep learning for wildfire spread prediction: A review. Fire, 7(12), 482.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e\u0026Aacute;ngel Javaloyes, M., Pend\u0026aacute;s-Recondo, E., \u0026amp; S\u0026aacute;nchez, M. (2023). A general model for wildfire propagation with wind and slope. SIAM Journal on Applied Algebra and Geometry, 7(2), 414\u0026ndash;439.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAnton, C. E., \u0026amp; Lawrence, C. (2016). Does place attachment predict wildfire mitigation and preparedness? A comparison of wildland\u0026ndash;urban interface and rural communities. Environmental management, 57(1), 148\u0026ndash;162.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAsori, M., Emmanuel, D., \u0026amp; Dumedah, G. (2020). Wildfire hazard and risk modelling in the northern regions of Ghana using GIS-based multi-criteria decision making analysis. Journal of Environment and Earth Science, 10(11).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBhowmik, R. T., Jung, Y. S., Aguilera, J. A., Prunicki, M., \u0026amp; Nadeau, K. (2023). A multi-modal wildfire prediction and early-warning system based on a novel machine learning framework. Journal of environmental management, 341, 117908.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBisenic, N. (2022). A Scenario Based Fire Susceptibility Approach for Remote Sensing Platform Comparison: Los Angeles County Area, Southern California, USA (Doctoral dissertation, University of Southern California).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBowman, D. M., Williamson, G. J., Abatzoglou, J. T., Kolden, C. A., Cochrane, M. A., \u0026amp; Smith, A. M. (2017). Human exposure and sensitivity to globally extreme wildfire events. Nature ecology \u0026amp; evolution, 1(3), 0058.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrady, L. M. (2021). From war zone to biosphere reserve: the Korean DMZ as a Scientific Landscape. Notes and Records, 75(2), 189\u0026ndash;205.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCao, L., Elliot, W., \u0026amp; Long, J. W. (2021). Spatial simulation of forest road effects on hydrology and soil erosion after a wildfire. Hydrological Processes, 35(6), e14139.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCao, Q. (2020). Exploring Spatially Varying Relationships between Forest Fire and Environmental Factors in Fujian, China (Doctoral dissertation, State University of New York College of Environmental Science and Forestry).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen, B., Wu, S., Jin, Y., Song, Y., Wu, C., Venevsky, S., \u0026hellip; Gong, P. (2024). Wildfire risk for global wildland\u0026ndash;urban interface areas. Nature Sustainability, 7(4), 474\u0026ndash;484.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCoskuner, K. A. (2022). Assessing the performance of MODIS and VIIRS active fire products in the monitoring of wildfires: a case study in Turkey. iForest-Biogeosciences and Forestry, 15(2), 85.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCosta-Saura, J. M., Bacciu, V., Ribotta, C., Spano, D., Massaiu, A., \u0026amp; Sirca, C. (2022). Predicting and mapping potential fire severity for risk analysis at regional level using Google Earth engine. Remote Sensing, 14(19), 4812.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCoutaz, G. (2018). South Korea: Vulnerability and Adaptation to Natural Disasters. In Coping with Disaster Risk Management in Northeast Asia: Economic and Financial Preparedness in China, Taiwan, Japan and South Korea (pp. 111\u0026ndash;132). Emerald Publishing Limited.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDetmer, A. R., Ward, E. J., Hunsicker, M. E., Andrews, K. S., Conrad, M., Ferriss, B. E., \u0026hellip; Samhouri, J. F. (2025). Evaluating the robustness of generalized additive models as a tool for threshold detection in variable environments. Ecosphere, 16(3), e70117.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDillon, G., Menakis, J., \u0026amp; Fay, F. (2015). Wildland fire potential: A tool for assessing wildfire risk and fuels management needs. In In: Keane, Robert E.; Jolly, Matt; Parsons, Russell; Riley, Karin. Proceedings of the large wildland fires conference; May 19\u0026ndash;23, 2014; Missoula, MT. Proc. RMRS-P-73. Fort Collins, CO: US Department of Agriculture, Forest Service, Rocky Mountain Research Station. p. 60\u0026ndash;76. (Vol. 73, pp. 60\u0026ndash;76).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDonovan, V. M., Wonkka, C. L., Wedin, D. A., \u0026amp; Twidwell, D. (2020). Land-use type as a driver of large wildfire occurrence in the US Great Plains. Remote Sensing, 12(11), 1869.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFang, L., Yang, J., White, M., \u0026amp; Liu, Z. (2018). Predicting potential fire severity using vegetation, topography and surface moisture availability in a Eurasian boreal forest landscape. Forests, 9(3), 130.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFIRMS. (2025). Fire Information for Resource Management System. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://firms.modaps.eosdis.nasa.gov/\u003c/span\u003e\u003cspan address=\"https://firms.modaps.eosdis.nasa.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHeo, S., \u0026amp; Lee, D. K. (2025). Assessing disasters in East Kalimantan: machine learning approaches for sustainable urban development. Environmental Research Letters, 20(10), 104003.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHeo, S., Park, S., \u0026amp; Lee, D. K. (2023). Multi-hazard exposure mapping under climate crisis using random forest algorithm for the Kalimantan Islands, Indonesia. Scientific Reports, 13(1), 13472.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHeo, S., Sohn, W., Park, S., \u0026amp; Lee, D. K. (2024). Multi-hazard assessment for flood and Landslide risk in Kalimantan and Sumatra: Implications for Nusantara, Indonesia's new capital. Heliyon, 10(18).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim, H., Seo, H., Kim, S., Kim, H., \u0026amp; Ko, M. (2022). Fish diversity and ichthyofauna of areas adjacent to the Demilitarized Zone in South Korea. Diversity, 14(12), 1011.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim, J. H., Park, S., Kim, S. H., \u0026amp; Lee, E. J. (2021). Long-Term Land Cover Changes in the Western Part of the Korean Demilitarized Zone. Land, 10(7), 708. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/land10070708\u003c/span\u003e\u003cspan address=\"10.3390/land10070708\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim, J., \u0026amp; Lee, S. O. (2024). Fragmented, Materialized, Militarized Geopolitics of Wildfires in the Inter-Korean Border. Geoforum, 155, 104077.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim, K. G. (2013). History and Evolution of the DMZ Ecosystem. In The Demilitarized Zone (DMZ) of Korea: Protection, Conservation and Restoration of a Unique Ecosystem (pp. 1\u0026ndash;28). Berlin, Heidelberg: Springer Berlin Heidelberg.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKorea Forest Service. (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.forest.go.kr/kfsweb/kfi/kfs/frfr/selectFrfrStatsArea.do\u003c/span\u003e\u003cspan address=\"https://www.forest.go.kr/kfsweb/kfi/kfs/frfr/selectFrfrStatsArea.do\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLaushman, K. M., Munson, S. M., \u0026amp; Villarreal, M. L. (2020). Wildfire risk and hazardous fuel reduction treatments along the US-Mexico border: a review of the science (1986\u0026ndash;2019). Air, Soil and Water Research, 13, 1178622120950272.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLee, S., Song, J. S., Lee, C. W., \u0026amp; Ko, B. (2022). The Study of DMZ Wildfire Damage Area Detection Method Using Sentinel-2 Satellite Images. Korean Journal of Remote Sensing, 38(5_1), 545\u0026ndash;557.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi, C., Su, Z., Ni, R., Wang, G., Ouyang, Y., Zeng, A., \u0026amp; Guo, F. (2025). Integrated spatial generalized additive modeling for forest fire prediction: a case study in Fujian Province, China. Journal of Forestry Research, 36(1), 30.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMalik, A., Rao, M. R., Puppala, N., Koouri, P., Thota, V. A. K., Liu, Q., \u0026hellip; Gao, J. (2021). Data-driven wildfire risk prediction in northern California. Atmosphere, 12(1), 109.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMarcos, B., Gon\u0026ccedil;alves, J., Alcaraz-Segura, D., Cunha, M., \u0026amp; Honrado, J. P. (2021). A framework for multi-dimensional assessment of wildfire disturbance severity from remotely sensed ecosystem functioning attributes. Remote Sensing, 13(4), 780.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMastrorillo, M., Scartozzi, C. M., Pacillo, G., Menza, G., Desai, B., Maviza, G., \u0026hellip; de Dinechin, F. (2024). Towards a common vision for climate change, security and migration in the Mediterranean.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMontoya Alvis, J., Orozco Mendoza, G. L., \u0026amp; Zartha Sossa, J. W. (2025). Extreme Fire Events in Wildland\u0026ndash;Urban Interface Areas: A Review of the Literature Concerning Determinants for Risk Governance. Sustainability, 17(10), 4505.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMoreira, F., Ascoli, D., Safford, H., Adams, M. A., Moreno, J. M., Pereira, J. M., \u0026hellip; Fernandes, P. M. (2020). Wildfire management in Mediterranean-type regions: paradigm change needed. Environmental Research Letters, 15(1), 011001.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMurillo-Sandoval, P. J., Hilker, T., Krawchuk, M. A., \u0026amp; Van Den Hoek, J. (2018). Detecting and attributing drivers of forest disturbance in the Colombian andes using landsat time-series. Forests, 9(5), 269.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMurray, A. T., Baik, J., Figueroa, V. E., Rini, D., Moritz, M. A., Roberts, D. A., \u0026hellip; Jones, C. (2023). Developing effective wildfire risk mitigation plans for the wildland urban interface. International Journal of Applied Earth Observation and Geoinformation, 124, 103531.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNaser, M. Z., \u0026amp; Kodur, V. (2025). Vulnerability of structures and infrastructure to wildfires: a perspective into assessment and mitigation strategies. Natural Hazards, 1\u0026ndash;21.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNasiri, V., Sadeghi, S. M. M., Bagherabadi, R., Moradi, F., Deljouei, A., \u0026amp; Borz, S. A. (2022). Modeling wildfire risk in western Iran based on the integration of AHP and GIS. Environmental monitoring and assessment, 194(9), 644.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNikolakis, W., \u0026amp; Roberts, E. (2022). Wildfire governance in a changing world: Insights for policy learning and policy transfer. Risk, Hazards \u0026amp; Crisis in Public Policy, 13(2), 144\u0026ndash;164.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePahlavani, P., Raei, A., Bigdeli, B., \u0026amp; Ghorbanzadeh, O. (2024). Identifying influential spatial drivers of forest fires through geographically and temporally weighted regression coupled with a continuous invasive weed optimization algorithm. Fire, 7(1), 33.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePapathoma-K\u0026ouml;hle, M., Schl\u0026ouml;gl, M., Garlichs, C., Diakakis, M., Mavroulis, S., \u0026amp; Fuchs, S. (2022). A wildfire vulnerability index for buildings. Scientific Reports, 12(1), 6378.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePark, J., Sim, W., Park, J., \u0026amp; Lee, J. (2019). Object-based Land Cover Change Detection and Landscape Structure Analysis of Demilitarized Zone in Korea. Sensors \u0026amp; Materials, 31.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePunzo, G., Castellano, R., \u0026amp; Bruno, E. (2022). Using geographically weighted regressions to explore spatial heterogeneity of land use influencing factors in Campania (Southern Italy). Land Use Policy, 112, 105853.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSagrario, M. D. L. \u0026Aacute;. G., Musazzi, S., C\u0026oacute;rdoba, F. E., Mendiolar, M., \u0026amp; Lami, A. (2020). Inferring the occurrence of regime shifts in a shallow lake during the last 250 years based on multiple indicators. Ecological Indicators, 117, 106536.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSakellariou, S., Sfougaris, A., Christopoulou, O., \u0026amp; Tampekis, S. (2022). Integrated wildfire risk assessment of natural and anthropogenic ecosystems based on simulation modeling and remotely sensed data fusion. International Journal of Disaster Risk Reduction, 78, 103129.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSalavati, G., Saniei, E., Ghaderpour, E., \u0026amp; Hassan, Q. K. (2022). Wildfire risk forecasting using weights of evidence and statistical index models. Sustainability, 14(7), 3881.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchag, G. M., Stow, D. A., Riggan, P. J., \u0026amp; Nara, A. (2022). Spatial-statistical analysis of landscape-level wildfire rate of spread. Remote Sensing, 14(16), 3980.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSutherland, D., Rashid, M. A., Hilton, J. E., \u0026amp; Moinuddin, K. A. (2023). Implementation of spatially-varying wind adjustment factor for wildfire simulations. Environmental Modelling \u0026amp; Software, 163, 105660.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eThapa, S. B. (2024). Integrated wildfire risk management: Measuring risk perceptions, simulating fire severity maps, and visualizing fire risk in the California Wildland-Urban Interface (Doctoral dissertation, University of California, Merced).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang, S. S. C., \u0026amp; Wang, Y. (2020). Quantifying the effects of environmental factors on wildfire burned area in the south central US using integrated machine learning techniques. Atmospheric Chemistry and Physics, 20(18), 11065\u0026ndash;11087.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWu, Z., He, H. S., Fang, L., Liang, Y., \u0026amp; Parsons, R. A. (2018). Wind speed and relative humidity influence spatial patterns of burn severity in boreal forests of northeastern China. Annals of Forest Science, 75(3), 66.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYe, T., Liu, W., Mu, Q., Zong, S., Li, Y., \u0026amp; Shi, P. (2020). Quantifying livestock vulnerability to snow disasters in the Tibetan Plateau: Comparing different modeling techniques for prediction. International Journal of Disaster Risk Reduction, 48, 101578.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYing, L., Cheng, H., Shen, Z., Guan, P., Luo, C., \u0026amp; Peng, X. (2021). Relative humidity and agricultural activities dominate wildfire ignitions in Yunnan, Southwest China: Patterns, thresholds, and implications. Agricultural and Forest Meteorology, 307, 108540.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang, Z., Yang, S., Wang, G., Wang, W., Xia, H., Sun, S., \u0026amp; Guo, F. (2022). Evaluation of geographically weighted logistic model and mixed effect model in forest fire prediction in northeast China. Frontiers in Forests and Global Change, 5, 1040408.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Wildfire ignition risk, Integrated modeling framework, Random Forest (RF), Generalized Additive Model (GAM), Geographically Weighted Regression (GWR), Spatial heterogeneity, Military–civilian interface zones (MCIZs), Demilitarized Zone (DMZ)","lastPublishedDoi":"10.21203/rs.3.rs-7872076/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7872076/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMilitary\u0026ndash;civilian interface zones (MCIZs) adjacent to the Korean Demilitarized Zone (DMZ) represent complex wildfire environments shaped by restricted access, intensive military activities, and adjacent civilian land use. This study develops a spatially explicit wildfire ignition risk assessment framework for the DMZ and Civilian Control Zone (CCZ) in Paju, South Korea, by integrating Random Forest (RF), Generalized Additive Models (GAM), and Geographically Weighted Regression (GWR). A dataset of 318 wildfire ignition events (2001\u0026ndash;2024), including 78 associated with military activities, was analyzed. The RF model achieved high predictive accuracy (AUC\u0026thinsp;=\u0026thinsp;0.81), identifying proximity to military training zones, relative humidity, wind speed, and proximity to built infrastructure as dominant ignition drivers. GAM revealed narrow nonlinear thresholds\u0026mdash;relative humidity at 13.8\u0026ndash;14.0% and wind speed at 13.5\u0026ndash;14.0 m/s\u0026mdash;corresponding to peak ignition probabilities. GWR demonstrated pronounced spatial heterogeneity, with military proximity exerting stronger influence in eastern and northern sectors, while meteorological effects varied geographically. Based on these outputs, an integrated modeling framework was established to synthesize probabilistic model results into spatially explicit ignition susceptibility maps. The proposed approach provides a scientifically rigorous and operationally applicable method for quantifying ignition risk in politically sensitive, access-restricted landscapes, offering valuable insights for adaptive wildfire prevention and spatially informed governance of transboundary fire risk.\u003c/p\u003e","manuscriptTitle":"Integrated Modeling of Wildfire Ignition Risk in the Military–Civilian Interface of the Korean DMZ","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-30 16:05:57","doi":"10.21203/rs.3.rs-7872076/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"11f3b4de-849f-40db-8c32-261a783661e2","owner":[],"postedDate":"October 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-01-13T05:38:57+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-30 16:05:57","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7872076","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7872076","identity":"rs-7872076","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}