Mapping Forest Climate-Sensitivity Belts in a Mountainous Region of Namyangju, South Korea, Using Satellite-Derived Thermal and Vegetation Phenological Variability

Abstract

Mountain forests play a key role in buffering local climate, yet their climate sensitivity is seldom mapped in a way that is directly usable for spatial planning. This study investigates how phenological thermal and vegetation variability are organized within the forested landscape of Namyangju, a mountainous region in central Korea, and derives spatial indicators of forest climate sensitivity. Using monthly, cloud-screened Landsat-8/9 land surface temperature (LST) and normalized difference vegetation index (NDVI) images over a recent multi-year period, we calculated phenological coefficients of variation for 34,123 forest grid cells and applied local clustering analysis to identify belts of high and low variability. Forest areas where LST and NDVI variability simultaneously occupied the upper tail of their distributions (top 5%/10%/20%) were interpreted as climate-sensitivity hotspots, whereas co-located coldspots were treated as microclimatic refugia. Across the mountainous terrain, sensitivity hotspots formed continuous belts along high-elevation ridges and steep, dissected slopes, while coldspots were concentrated in sheltered valley floors. Notably, the most sensitive belts were dominated by high-elevation conifer stands, despite the limited seasonal fluctuation typically expected in evergreen canopies. This pattern suggests that elevation strongly amplifies the coupling between thermal responsiveness and vegetation health, whereas valley-bottom forests act as stabilizers that maintain comparatively constant microclimatic and phenological conditions. We refer to these patterns as “forest climate-sensitivity belts,” which translate satellite observations into spatially explicit information on where climate-buffering functions are most vulnerable or resilient. Incorporating climate-sensitivity belts into forest plans and adaptation strategies can guide elevation-aware species selection in new afforestation, targeted restoration and fuel-load management in upland sensitivity zones, and the protection of valley refugia that support biodiversity, thermal buffering, and hydrological regulation. Because the framework relies on standard satellite products and transparent calculations, it can be updated as new imagery becomes available and transferred to other seasonal, mountainous regions, providing a practical basis for climate-resilient forest planning.

1. Introduction

Forests identified in landcover maps are increasingly recognized as a Nature-based Solution (NBS) that moderates climate through tree-dominated processes [1,2,3]. In particular, forest canopies buffer energy exchange between the atmosphere and the land surface and regulate surface temperature via radiative shielding and evapotranspiration, thereby playing a pivotal role in maintaining local climate stability [4,5,6]. Forests also store atmospheric carbon dioxide (CO₂) in biomass, reducing greenhouse gases (GHGs) and mitigating climate change. These functions position forests, within the NBS framework, as critical natural capital that delivers both mitigation and adaptation [7,8,9,10]. Sustaining these NBS benefits requires continuous, wall-to-wall monitoring and adaptive forest management [11,12,13]. However, approximately 63% of South Korea is forested, making comprehensive field surveys across the entire territory impractical [14,15]. This makes it difficult to distribute forest conditions and apply appropriate forest management methods accordingly. To overcome this constraint, the forest sector increasingly relies on remote sensing for synoptic, repeatable, and spatially explicit monitoring beyond sample-based assessments.

To operationalize monitoring of these functions, this study employs remote-sensing indicators that jointly capture canopy condition and surface thermal states—most notably the normalized difference vegetation index (NDVI) and land surface temperature (LST). NDVI is a canonical observation-derived index that quantifies vegetation activity and cover [16,17]. Generally, higher NDVI indicates denser and healthier vegetation, supporting observation-based monitoring [18,19,20]. Land Surface Temperature (LST) is derived from thermal infrared emissive properties to estimate surface temperature [21,22]. LST enables monitoring of thermal-related phenomena such as urban heat islands, drought, and crop or vegetation stress [23,24,25]. Both LST and NDVI can be obtained from satellite observations at commensurate acquisition times, for example, from the United States Geological Survey (USGS) Landsat series [26,27]. LST and NDVI are tightly coupled: in summer, areas with high NDVI typically exhibit lower LST because canopy shading and evapotranspiration depress surface temperatures, yielding a negative association; in winter, the relationship can invert and become positive [28,29,30,31,32]. These seasonal reversals, driven by acquisition timing and vegetation phenology, are a critical consideration for data use [18,33,34,35]. Especially in strongly seasonal climates such as South Korea, where phenological dynamics render the LST–NDVI association non-stationary throughout the year. To overcome this, previous studies analyzed LST and NDVI over a long period of time, but there is a limitation in that the resolution is low due to the use of MODIS satellites [36].

Accordingly, treating the LST–NDVI relationship as a single global correlation is inadequate; approaches that reflect spatiotemporal structure and variability are required. In forested landscapes, both variables exhibit pronounced spatial autocorrelation, with similar values clustering among neighboring locations [37,38]. Conventional regression models that ignore this spatial structure can yield biased or misleading inferences, and mean-based statistics alone are ill-suited to capture such variability patterns. To characterize the spatial distribution of LST and NDVI more precisely, spatial statistics, particularly hotspot analysis capable of identifying local clusters, are warranted [39,40]. While hotspot analysis is informative, it is typically univariate, and thus additional procedures are needed to incorporate temporal variation.

This study addresses those limitations by integrating hotspot analysis with time-series statistics to elucidate the spatiotemporal relationship between LST and NDVI and, in turn, to explore forests’ climate-buffering function. This study computes the annual coefficient of variation (CV) for annual time-series LST and NDVI and conducts hotspot analysis on these CV fields using Getis–Ord Gi* hotspot analysis [41,42]. The CV, a dimensionless ratio of the annual standard deviation to the mean, expresses fluctuation amplitude relative to the mean and thereby enables direct comparison of the relative variability of two variables with different units. This study then evaluates the coupling between thermal and greenness variability by regressing the Getis–Ord Gi* Z-score (GiZ) fields, first using ordinary least squares (OLS) and subsequently spatial regression models that account for spatial autocorrelation. In addition, we identify “management-priority sites” as locations that simultaneously fall within the upper 5% of both the GiZ distributions of LST and NDVI, and use these areas to derive policy implications for spatially explicit forest management. Beyond purely numeric summary statistics, this framework provides a basis for understanding, in space and time, how forests respond to climate variability and for designing targeted responses.

Our analysis offers decision-relevant evidence by analyzing LST–NDVI interactions across space and time and by objectively identifying forest areas where climate-mitigation and buffering functions are most pronounced. We assume that LST and NDVI share spatial clustering patterns because they are jointly shaped by common topographic and microclimatic forcing, and that when this clustering is explicitly quantified via GiZ, the resulting GiZ–GiZ relationships will exhibit stronger coupling than CV–CV relationships that consider only the magnitude of variability. By analyzing the co-clustering patterns of satellite-derived thermal and vegetation indicators within forested landscapes, this study reveals how forests modulate local microclimates and where these buffering functions are most vulnerable. In doing so, it underscores the importance of spatially explicit forest planning and management for sustaining climate-regulating ecosystem services in mountainous, seasonally variable regions such as South Korea.

2. Materials and Methods

2.1. Study Area

The study area is Namyangju, Gyeonggi-do, Republic of Korea (Figure 1). Across the Korean Peninsula, including Namyangju, the climate is humid continental with warm, wet summers and cold, dry winters. Mean annual air temperature is around 12–13 °C, with an annual temperature range of roughly 28 °C between the coldest (January) and warmest (August) months, and annual precipitation on the order of 1200–1400 mm, most of which falls during the summer monsoon. These large seasonal swings in temperature and moisture produce strong phenological variability in forest microclimate and markedly different phenological stages over the course of the year.

Figure 1. Study Area: Namyangju, Gyeonggi-do, Republic of Korea: (a) Administrative boundary; (b) Digital elevation model by SRTM (Shuttle Radar Topography Mission).

Situated on the eastern periphery of the Seoul Capital Region, Namyangju is an administrative district in which 67.28% of the total area, 30,822 ha, is forested, substantially higher than the provincial average of 50.23% (Figure 2) [43]. According to the microclass landcover map of the Ministry of Environment of Korea, which was produced from 1 m resolution aerial orthophotos with an overall classification accuracy of approximately 95%, the forests of Namyangju comprise approximately 20% coniferous stands, 71% broadleaved stands, and 9% mixed conifer–broadleaf stands (Figure 2a). This high forest fraction provides favorable conditions for evaluating the ecological and climatic functions that forests perform in urban-adjacent settings. Owing to its mosaic of urban fabric interspersed with forests, Namyangju is well-suited to observe how multiple forest functions, including climate regulation, mitigation of urban thermal environments, and the provision of ecosystem services, are connected to everyday living spaces.

Figure 2. Forest distribution of Namyangju city: (a) Forest distribution from microclass landcover map of Ministry of Environment: (b) Drone image of forest from April 2025.

Additionally, Namyangju features steep slopes and mountainous terrain spanning a wide elevational range, making it well-suited to analyze topographically driven spatial heterogeneity in vegetation and the attendant variation in land surface temperature. The area exhibits a mid-latitude continental climate with pronounced seasonal temperature contrasts, allowing observation of both summer heat and winter cold, which is advantageous for elucidating the seasonal relationship between LST and NDVI [44,45,46]. However, because the Republic of Korea—including the study area—is subject to the East Asian monsoon, summertime cloud cover is extensive, making it difficult to conduct year-long, monthly time-series remote-sensing analyses [47]. Among the relevant administrative units from which cloud-free, multi-temporal Landsat-8 and Landsat-9 imagery can be acquired to construct the time-series LST and NDVI inputs, including the neighboring, heavily forested jurisdictions of Pocheon and Yangpyeong, only Namyangju met the data-availability criteria for this study.

2.2. Datasets for Analysis

Both LST and NDVI are canonical observation-derived indices computed from satellite imagery. The two metrics, respectively, reflect the thermal state of the land surface and the vigor of vegetation activity. LST is retrieved from thermal infrared radiance, typically by inverting a radiative transfer/Planck function to brightness temperature and then converting to land surface temperature with emissivity correction [48]. NDVI, defined from the contrast between red and near-infrared reflectance, captures vegetation density and condition [49]. Meaningful interpretation is enabled by comparing these datasets at coincident acquisition times and harmonized spatial resolution. Accordingly, quantifying the LST–NDVI relationship requires image data acquired by the same (or cross-calibrated) sensor platform. In this context, we used imagery from the Landsat-8 and Landsat-9 missions provided by the USGS.

Because cloud contamination can severely degrade both thermal and optical signals and act as extreme outliers in variability metrics (e.g., standard deviation and coefficient of variation), we screened scenes using a strict criterion (0% scene-level cloud cover) and searched across 2020–2025 to construct a near-monthly sequence. To reduce potential bias from longer-term forest accumulation and growth changes, we limited the earliest search year to 2020. No cloud-free scene was available for July under this criterion, and the final dataset comprised 11 cloud-free scenes (Table 1).

Table 1. Cloud-free Landsat-8/9 scene selection used to construct the monthly LST–NDVI time series (2020–2025; Path 116/Row 034; July missing).

	Jan	Feb	Mar	Apr	May	Jun
Satellite	Landsat-9	Landsat-9	Landsat-9	Landsat-9	Landsat-8	Landsat-8
Date	01/26/2024	02/05/2022	03/14/2024	04/07/2024	05/09/2024	06/05/2025
	Jul	Aug	Sep	Oct	Nov	Dec
Satellite	Missing	Landsat-9	Landsat-8	Landsat-9	Landsat-9	Landsat-8
Date	-	08/29/2024	09/19/2020	10/24/2024	11/09/2024	12/19/2024

The two satellites carry nearly identical and cross-calibrated payloads, Operational Land Imager (OLI/OLI-2) and Thermal Infrared Sensor (TIRS/TIRS-2), yielding high data continuity and interoperability [50]. OLI supplies the visible and near-infrared bands used to generate NDVI, while TIRS provides the thermal bands needed to estimate LST. Spatial sampling is 30 m for OLI and 90 m TIRS, with each satellite on a 16-day repeat cycle; phased orbits provide ~8-day effective revisit when combined. We extracted scenes over Namyangju restricted to Path 116/Row 034 to ensure geometric consistency and assembled a near-monthly sequence spanning January–December. Because using more than one image in a given month would overweight specific periods and artificially inflate the variance and standard deviation when computing annual CV, we limited the series to approximately one scene per month and used multi-year acquisitions to fill month-level gaps.

2.3. Data Processing

LST was derived from the USGS Landsat 8/9 Collection-2 Level-2 Surface Temperature product (ST_B10). Rather than per-pixel QA masking, entire scenes exhibiting cloud, cloud shadow, cirrus, snow/ice, or water contamination over the study area (per QA bit flags) were excluded, and only cloud-free scenes were retained. The reason for excluding entire scenes is that LST is highly sensitive to cloud contamination, which produces spuriously low temperatures and can induce substantial errors when analyzing LST–NDVI correlations. Retained ST_B10 values were converted from scaled digital numbers to Kelvin and then to degrees Celsius (Equation (1)).

$$T_{°\mathrm{C}}=T_K-273.15$$

(1)

NDVI was derived from the USGS Landsat 8/9 Collection-2 Level-2 Surface Reflectance (L2 SR) product, using the red (Band 4) and NIR (Band 5) channels (Equation (2)). Because OLI and TIRS are co-acquired along the same orbit and acquisition time, we applied the same scene-level quality screening used for LST—excluding scenes flagged for cloud, cloud shadow, cirrus, snow/ice, or water—and retained only cloud-free observations. Consequently, each NDVI scene is contemporaneous with its corresponding LST scene. NDVI rasters were co-registered to the LST grid and resampled to 90 m (bilinear) for consistency across analyses (Figure 3). All LST and NDVI preprocessing was carried out in the Google Earth Engine (GEE) cloud platform using the JavaScript API, based on USGS Landsat 8/9 Collection-2 Level-2 Surface Temperature and Surface Reflectance products. For visual comparability across the full time series, the display range for LST was fixed to −8 to 40, and the display range for NDVI was fixed to 0 to 1 in Figure 3.

$$NDVI={\displaystyle \frac{NIR-RED}{NIR+RED}}$$

(2)

Figure 3. Generated monthly LST and NDVI images derived from Landsat 8/9.

To quantify the temporal stability and variability of LST and NDVI within forests, we spatially masked satellite scenes for Namyangju using the Ministry of Environment’s 2023 microclass landcover map and retained only areas classified as “forest.” (Figure 2) [51]. This restriction was intended to isolate and analyze climate–vegetation interactions specifically within forested landscapes.

Over the extracted forest mask, we generated a regular lattice of grid points at 90 m spacing, co-registered to the imagery and consistent with the native spatial resolution, yielding a total of 34,123 points. For each point, we extracted the complete time series of LST and NDVI values for the study period (forest area). The resulting per-point series were then used to compute the mean, standard deviation (SD), and CV for each variable.

The CV, defined as SD divided by the mean, is a dimensionless measure that enables direct comparison of relative variability across variables with different units and ranges (Equation (3)). Reliance on SD alone can over- or understate variability as a function of absolute magnitude; introducing the CV mitigates this bias and provides a more interpretable basis for comparing the relative stability or sensitivity of LST and NDVI within forest areas. We computed annual means, standard deviations, and coefficients of variation in Microsoft Excel 365 using built-in functions (e.g., AVERAGE, STDEV.P, and CV = SD/mean).

$$Coefficient of Variation={\displaystyle \frac{Standard Deviation}{Average}}$$

(3)

We next performed hotspot analysis on the CV fields of LST and NDVI separately, using the Getis–Ord Gi* statistic and its standardized Z-score to identify clusters of significantly high (hot) and low (cold) variability [41,42,52]. After testing multiple neighborhood distances in the hotspot procedure (270–540 m), we selected a distance band of 540 m as the operational scale, because it yielded the most stable and interpretable hotspot patterns and the strongest, yet robust, coupling between LST and NDVI variability in the subsequent regression analyses. Hotspot and coldspot patterns of intra-annual variability were identified in ArcGIS Pro (Esri, Redlands, CA, USA, v3.5.4) using the Hot Spot Analysis (Getis–Ord Gi*) tool. The analysis was applied to the LST CV and NDVI CV fields, yielding per-cell Getis–Ord Gi* statistics and their GiZ for both variables. These GiZ values were exported and subsequently used in the regression analyses.

To assess this spatial coupling, we regressed the NDVI GiZScores on the LST GiZScores using ordinary least squares (OLS). We then extended this analysis with spatial regression, fitting spatial autoregressive lag (SAR), spatial error (SEM), and spatial lag-of-X (SLX) models under k-nearest-neighbor spatial weights to account for residual spatial autocorrelation and to test the robustness of the estimated coupling. All regression analyses were carried out in Python (version 3.11) within a Jupyter Notebook environment. OLS models and spatial autoregressive models—SAR, SEM, and SLX—were estimated using the PySAL/spreg library, with k-nearest-neighbor spatial weights constructed via libpysal. Data handling and visualization were performed with pandas (v2.3.3), NumPy (v2.3.3), SciPy (v1.16.3), and Matplotlib (v3.10.7).

The data flow of the study is summarized in Figure 4. Based on the expectation that areas with high thermal variability also experience elevated vegetation variability, we anticipated a positive regression coefficient. This analysis evaluates the strength and significance of that relationship at spatial scales relevant to management, providing evidence on how the forest climate-buffering function varies across space. In doing so, it empirically links thermal sensitivity and vegetation stability and offers spatially explicit information to support climate-adaptation and forest-management decision-making.

Figure 4. Flow Chart of this study.

3. Results

3.1. Identifying Seasonal Change in CVs of LST and NDVI

To characterize variability across temporal scales, we first summarized the monthly mean patterns of normalized difference vegetation index (NDVI) and land surface temperature (LST) across all forest grid cells, reporting the monthly mean together with the inter-point dispersion (mean ± 1 standard deviation, shaded band; Figure 5a). We then summarized the coefficient of variation (CV) of LST and NDVI by season (spring, summer, fall, and winter), again reporting the mean CV with its spatial dispersion (mean ± 1 standard deviation, shaded band), and we additionally show the all-period mean CV as a dashed horizontal reference line to facilitate direct comparison between seasonal and overall conditions (Figure 5b,c).

Figure 5. Seasonal and monthly variability patterns of LST and NDVI: (a) Monthly mean NDVI and LST; (b) Seasonal mean LST CV; (c) Seasonal mean NDVI CV.

Monthly mean NDVI and LST follow expected seasonal cycles (Figure 5a): NDVI increases rapidly in late spring, remains high through early summer, and declines into winter, while LST rises to a late-summer peak and decreases toward winter. The shaded bands indicate that spatial heterogeneity is generally larger during transitional periods, particularly for NDVI during the green-up and senescence phases, consistent with heterogeneous canopy development and site conditions across complex terrain.

Seasonal CV patterns show distinct behavior for LST and NDVI (Figure 5b,c). Seasonal LST CV is relatively higher in spring and fall and lower in summer, with the lowest mean in winter (Figure 5b). In contrast, NDVI CV exhibits stronger seasonality, peaking in spring, dropping sharply in summer, and partially rebounding in fall, while remaining lower in winter (Figure 5c). For both variables, the dashed all-period mean lies above the seasonal means, indicating that CV computed over the full annual period is amplified by cross-season contrasts (e.g., warm–cold and green-up–dormancy transitions) in addition to within-season fluctuations. Together, these results show that thermal variability is most pronounced during transition seasons, whereas vegetation variability is most pronounced during spring green-up, and that annual-scale CV captures additional variability arising from seasonal turnover beyond any single-season subset.

Notably, although the monthly series was assembled by selecting representative scenes for each month across multiple years due to cloud constraints, the resulting NDVI and LST trajectories still reproduce the generally expected phenological seasonality (i.e., spring green-up, summer peak, and autumn–winter decline), supporting the interpretability of the summarized patterns.

3.2. Result of Getis–Ord Gi* Hotspot Analysis for CVs of LST and NDVI

The spatial distributions of hotspots in annual CV of LST and NDVI, derived via Getis–Ord Gi* hotspot analysis, are shown in Figure 6a,b. In both variables, the Getis–Ord Gi* maps of annual CV reveal a coherent spatial organization. Despite differences in the areal extent of clusters across confidence thresholds, LST and NDVI exhibit broadly congruent patterns: hotspots form continuous bands along high-elevation ridges and dissected mountain slopes, whereas coldspots dominate valley floors and low-lying terrain. The close correspondence with the digital elevation model (DEM) indicates that terrain imposes a first-order control on phenological variability (Figure 1b).

Figure 6. Getis–Ord Gi* hotspot maps of the annual coefficients of variation (CV) and management-priority sites: (a) Hot spot of LST–CV; (b) Hot spot of NDVI–CV; (c) Management-priority sites.

We delineated management-priority sites by selecting grid cells (evaluated at the top 5%, 10%, and 20% thresholds) that fell within the top 5%/10%/20% of Getis–Ord Gi Z-scores for LST–CV and simultaneously within the top 5%/10%/20% for NDVI–CV—i.e., locations where phenological thermal and greenness variability are each significantly clustered at or above the 95th/90th/80th percentile (Figure 6c). Operationally, we intersected the two hotspot layers to isolate zones of co-clustered variability at each threshold. Comparison with the DEM (Figure 1b) shows that these co-hotspot belts are preferentially situated in higher-elevation terrain, with elevation distributions shifted upward relative to the forest-wide baseline and spatial concentrations along upland ridges and dissected, steep slopes. The mean elevation of the co-hotspot belts increased with stricter thresholds, averaging 342 m at the 20% level, 367 m at the 10% level, and 447 m at the 5% level. This topographic alignment indicates that orographic controls—enhanced thermal contrasts, aspect-driven insolation, wind exposure, and thinner soils—coincide with stronger phenological amplitude and canopy structural heterogeneity, jointly amplifying both LST and NDVI variability. From a management standpoint, these high-elevation co-hotspots constitute priority surveillance and intervention zones for climate-adaptation actions. In mountainous zones, stronger seasonal thermal contrasts, aspect-driven insolation gradients, wind exposure, thinner soils, and intermittent water limitation jointly amplify LST variability; concomitantly, phenological amplitude and canopy structural heterogeneity enhance NDVI variability. Conversely, valley bottoms—benefiting from hydrologic buffering, gentler slopes, and more continuous canopy cover—show relatively muted variability in both fields.

A few local mismatches between the two maps suggest partial decoupling of thermal and greenness dynamics. For example, LST hotspots extend more broadly than NDVI hotspots in some ridge–shoulder complexes, consistent with thermal variability not always translating into proportional changes in vegetation greenness (e.g., where evergreen cover, shading, or soil moisture mitigate phenological swings). Taken together, the patterns support our hypothesis that spatial clustering of thermal variability tends to co-occur with clustering of vegetation variability, especially in complex terrain, and they identify upland belts as priority zones for monitoring climate–vegetation sensitivity.

3.3. Quantifying LST–NDVI Linkage Using Ordinary and Spatial Regression Models

We quantified the LST–NDVI linkage across 34,123 forest grid cells using two sets of regression models. First, we fit OLS specifications for (i) a variability model relating the coefficients of variation (CV–CV) and (ii) a hotspot model relating the Getis–Ord Gi* Z-scores (GiZ–GiZ), in each case regressing NDVI on LST. Both specifications yielded positive slopes, indicating that grid cells with larger phenological thermal variability or stronger thermal clustering also tend to exhibit larger vegetation variability or stronger vegetation clustering. OLS regressions showed a moderate positive linkage between LST and NDVI variability (R² = 0.28 for CV–CV), whereas the hotspot formulation yielded a substantially stronger association (R² = 0.42 for GiZ–GiZ), indicating that co-located clusters of high thermal variability tend to coincide more tightly with clusters of high vegetation variability than suggested by variability magnitudes alone. Consistently, the hotspot (GiZ–GiZ) model showed higher explanatory power than the CV–CV model (higher R² and correlation coefficients) (Table 2), implying that explicitly modeling spatial clustering captures the coupling between LST and NDVI more effectively than using variability magnitude alone. Rank-based (Spearman) correlations preserved the same ordering, indicating that this advantage of the hotspot formulation is not driven by a small number of extreme observations (Figure 7a,b).

Table 2. Summary statistics of ordinary least squares regressions relating LST to NDVI for the CV–CV and GiZ–GiZ formulations.

	R2	Pearson r	Spearman ρ
NDVI–CV~LST–CV	0.2841	0.5330	0.4475
NDVI–GiZ~LST–GiZ	0.4273	0.6537	0.5640

Figure 7. Comparison of LST–NDVI relationships: (a) CV vs. CV; (b) GiZ vs. GiZ; (c) Standardized graph.

To facilitate direct visual comparison of the two relationships, we standardized each variable to z-scores (mean 0, standard deviation 1) and overlaid the CV–CV and GiZ–GiZ pairs in a single panel with identical axes (Figure 7c). For visual clarity, we randomly subsampled 3000 grid cells from the full dataset when plotting the scatter points. Both axes are constrained to [−2, 5] and use a 1:1 scale; CV–CV is plotted in blue and GiZ–GiZ in orange, with OLS lines drawn over the point clouds. This normalization removes unit and scale effects, allowing the slopes to be interpreted purely as standardized effect sizes. Consistent with the regression results, the GiZ–GiZ line exhibits a steeper standardized slope than the CV–CV line, visually reinforcing that local clustering of thermal variability co-occurs more strongly with local clustering of vegetation variability.

To keep the regression narrative concise for a one-predictor setting, we treat spatial models (SAR/SEM/SLX) as robustness checks rather than standalone alternatives. These models differ from OLS only in that they explicitly account for spatial dependence (in the outcome, errors, or spatially lagged covariates), helping prevent inflated significance when nearby grid cells are not independent. Across k-nearest-neighbor weights, the GiZ–GiZ specification consistently retained higher pseudo-R² and lower errors than CV–CV, and an intermediate neighborhood (k ≈ 15) provided a practical balance between fit and residual spatial autocorrelation (Figure 8).

Figure 8. Sensitivity of spatial regression models to the number of nearest neighbors (k): (a) Pseudo-R² vs. k; (b) Residual Moran’s I vs. k; (c) RMSE vs. k.

We therefore use GiZ–GiZ OLS as the primary indicator of LST–NDVI spatial coupling, because GiZ standardizes local clustering relative to neighbors and is less sensitive to pixel-scale noise and broad-scale gradients than raw CV–CV.

For these reasons, we adopt the GiZ–GiZ OLS as our primary indicator and interpret it as a statistical characterization of the spatial association between thermal and greenness dynamics, while treating the spatial regression models (SAR, SEM, SLX) as robustness checks. Across k values, the GiZ–GiZ spatial specifications consistently retained higher pseudo-R² and lower errors than CV–CV, and k ≈ 15 provided a practical balance between fit and residual spatial autocorrelation (Figure 8). Thus, the stronger coupling observed for the hotspot formulation in OLS persists even after explicitly accounting for spatial dependence

4. Discussion

4.1. Implications of Results

This study demonstrates that spatiotemporal clustering of thermal variability (LST–CV) is closely mirrored by clustering of vegetation variability (NDVI–CV) across a mountainous, forest-dominated urban fringe. Seasonal summaries further show that LST–CV is elevated during transition seasons (spring and fall), whereas NDVI–CV peaks in spring and is suppressed during summer, reproducing a generally expected phenological pattern despite monthly inputs being compiled from multiple years under cloud constraints. The Getis–Ord Gi* results show that hotspots of phenological variability align with high-elevation ridges and dissected slopes, whereas coldspots concentrate in valley floors. This concordance with topography indicates that terrain imposes a first-order control on both thermal dynamics and phenology: stronger seasonal thermal contrasts, aspect-driven insolation gradients, wind exposure, and thinner soils in uplands amplify LST fluctuations; simultaneously, phenological amplitude and canopy structural heterogeneity increase NDVI variability. Conversely, hydrologically buffered lowlands with gentler slopes exhibit muted variability in both fields. Together, these patterns support the interpretation of forests as climate buffers whose effectiveness varies systematically with terrain and landscape position [53,54,55].

A key contribution of this study is the comparison between two ways of formalizing the LST–NDVI linkage using observation-based remote sensing data: a direct regression of variability magnitudes (CV–CV) and a regression of local clustering intensities (GiZ–GiZ) [29,56,57]. Although both specifications yielded positive and statistically meaningful associations, the hotspot formulation provided a substantially tighter fit (R² = 0.4273 vs. 0.2841) and stronger rank concordance. We attribute this improvement to the properties of GiZScores: as standardized local statistics that explicitly embed neighborhood structure via a spatial weights matrix, GiZScores benchmark each pixel against its neighbors, attenuating pixel-scale noise and reducing variance heterogeneity caused by spatial non-stationarity [58,59]. In practical terms, the GiZ–GiZ model targets the phenomenon of interest—the co-location of thermal and greenness hotspots/coldspots—more directly than CV–CV and therefore serves as a more appropriate indicator of spatial coupling in complex terrain [60,61].

These findings carry several implications for forest monitoring and management in urban-adjacent mountain regions. First, the co-clustering of high thermal and high greenness variability in uplands suggests that upland belts function as sensitivity zones where climate anomalies are more likely to propagate into pronounced phenological responses [62,63,64]. Second, because our GiZ approach identifies statistically concentrated belts rather than isolated pixels, it can inform operational prioritization—for example, targeting the top 5%/10%/20% of GiZScores as “management-priority sites” for intensified field checks, fuel-load management, microclimate restoration (e.g., windbreaks, understory moisture conservation), or the placement of in situ temperature/soil-moisture loggers. Third, the persistence of coldspots in valley floors highlights the role of hydrological buffering; safeguarding riparian corridors and cold-air drainage pathways may help maintain local thermal stability and phenological regularity within the broader urban matrix.

Methodologically, the study contributes a parsimonious yet spatially informed pipeline for coupling time-series variability metrics with local cluster statistics. Computing annual CVs provides a unitless, comparable measure of phenological fluctuation; applying Gi* to the CV fields transforms those fluctuations into statistically vetted clusters; and a simple OLS on GiZ scores yields an interpretable slope (co-clustering strength). To test robustness to residual spatial dependence, we additionally fit spatial regression models—SAR, SEM, and SLX—under k-nearest-neighbor weights. Spatial regression models fitted under k-nearest-neighbor weights further showed that the GiZ–GiZ specifications retain higher (pseudo) R² and lower prediction errors than their CV–CV counterparts across a range of k values, with k ≈ 15 offering a reasonable compromise between explanatory power and residual spatial autocorrelation. As k increases, the neighborhood expands and progressively smooths local structure, yielding only marginal gains in fit; therefore, k ≈ 15 was adopted as a pragmatic balance between model performance and preservation of local spatial patterns. In this sense, GiZ–GiZ OLS provides a robust yet transparent summary of spatial coupling, while the spatial models serve as robustness checks rather than heavily parameterized alternatives.

4.2. Policy Applications and Management Implications

From a policy perspective, integrating GiZ-based hotspot belts into municipal forest plans could help align routine surveillance with areas of greatest climate–vegetation sensitivity. Because the method is sensor-agnostic and computationally light, it is readily transferable to other counties in Gyeonggi-do and to other seasonal, mountainous regions where rapid, wall-to-wall screening is needed. More broadly, mapping the spatial concordance of thermal and phenological variability provides an observation-based line of evidence for climate buffering by forests, supporting nature-based solutions in local climate-adaptation and carbon-neutrality strategies [65,66].

The outputs have practical relevance for management prioritization. Hotspot belts delineate areas where forest temperature regulation and vegetation stability may be relatively vulnerable, whereas persistent coldspots indicate more stable conditions that may warrant safeguarding as microclimate-stabilizing zones. In our results, management-priority hotspots were concentrated in higher-elevation belts, suggesting that terrain-driven exposure can coincide with stronger coupled variability. These products can therefore support targeted field checks, monitoring deployment, and site-level interventions (e.g., restoration or fuel-load management) in sensitivity-prone uplands.

Finally, the standardized nature of the indicators enables consistent comparison across jurisdictions and update cycles. Repeating the same CV–Gi*–GiZ procedure on future image stacks can help track whether sensitivity belts shift or intensify over time, supporting adaptive planning based on comparable, reproducible spatial evidence.

4.3. Broader Applications for Nature-Related Risk and Spatial Planning

Beyond forest management, the co-clustering layers can be integrated with external spatial datasets (e.g., land-development footprints, disturbance maps, or infrastructure/asset layers) to screen where human activities intersect with sensitivity-prone forest belts. Such overlays can help identify potential exposure areas and prioritize monitoring or mitigation where disturbances coincide with high-variability zones [67].

The framework is also suitable for broader nature-related risk assessments that require spatially explicit evidence of ecosystem interface and potential impacts. For example, within the TNFD LEAP approach, the products can support the “Locate” and “Evaluate” steps by providing georeferenced screening indicators of where thermal and vegetation variability co-occur most strongly [68]. More generally, the method offers a repeatable way to translate satellite time series into actionable sensitivity belts for spatial planning and risk screening.

4.4. Limitations and Future Research Directions

This study has several limitations that should be acknowledged when interpreting the results. First, the temporal window is still relatively narrow. Although we used monthly composites over multiple years, the analysis ultimately relied on a limited number of cloud-free scenes and could not fully represent the summer monsoon period, when cloud cover is persistently high. As a result, the annual cycle of thermal and phenological variability, especially during peak heat and moisture stress, is only partially captured. This cloud-driven gap may also under-represent short-lived extreme thermal conditions, meaning that the estimated annual variability metrics can be conservative for summer extremes. Moreover, if persistent heatwaves or other extreme events were fully observed, the location and intensity of hotspot belts could shift; testing this sensitivity requires longer, more continuous multi-year time series. Longer, more continuous time series that explicitly include extreme years and key seasonal windows would allow a more robust characterization of forest climate sensitivity. Accordingly, the identified belts should be interpreted as observation-based, screening-level indicators for the analyzed time window rather than as definitive long-term climate sensitivity; confirming persistence will require multi-year series that include extreme summers and monsoon-season conditions. In addition, the present analysis lacks independent validation of the inferred microclimate patterns. Our results are derived from satellite-based land surface temperature and vegetation indices and therefore capture relative spatial organization (hotspot/coldspot belts) rather than direct in situ microclimate conditions within the canopy and near-surface air layer. Future work should validate and refine the identified belts using ground observations, such as networks of temperature/soil-moisture loggers or existing meteorological station data. Building on the belts identified here, follow-up studies can further analyze and quantify the microclimate impacts of these belts (e.g., local cooling intensity, moisture buffering, and heat-stress exposure) through field measurements and integrated analyses.

Second, the current framework emphasizes elevational gradients as a primary driver of sensitivity belts. Elevation is indeed a strong control on temperature regimes, but forest microclimates and vegetation dynamics are shaped by a much broader set of ecological and structural factors, including species composition (e.g., conifer versus broadleaf dominance), stand age and density, canopy layering, soil properties, moisture availability, and management history. However, these factors were not explicitly included as covariates in our regression models because the aim of this study was to quantify the joint spatial co-clustering between LST and NDVI variability as an operational screening indicator, rather than to attribute the coupling to specific drivers. Consequently, the observed associations should be interpreted as descriptive patterns of co-variability, and future work should incorporate terrain, structure, moisture, and disturbance variables in multivariate (spatial) models to improve attribution. In our results, the highest-sensitivity zones were concentrated in high-elevation conifer stands, even though conifers maintain foliage year-round and exhibit relatively small NDVI fluctuations. This suggests that altitude strongly amplifies thermal responsiveness and vegetation-health sensitivity, but it also highlights the need to disentangle the relative effects of topography, stand structure, and species traits. Future work should therefore integrate additional biophysical layers (e.g., stand inventory data, LiDAR-derived structure, soil and moisture maps) to identify which combinations of factors most strongly govern forest climate sensitivity.

Third, the choice of indicators and data products is still constrained. We relied on NDVI and Landsat-based LST proxies, which are well established but have known limitations in dense canopies and complex terrain. Complementary indices—such as EVI, red-edge metrics, or solar-induced chlorophyll fluorescence—and higher-resolution thermal or hyperspectral data could sharpen sensitivity detection and better capture stress responses in evergreen and mixed stands. Finally, our use of neighborhood-based statistics means that hotspot patterns remain partly dependent on the specification of the spatial weights and scale of analysis; systematic sensitivity testing across alternative neighborhood sizes and zoning schemes would help confirm the robustness of the identified belts.

Despite these limitations, the present framework demonstrates that combining time-series variability with standardized local clustering can yield an operational view of how forests modulate microclimate across mountainous landscapes. Building on this foundation, future research should (i) expand to longer and more diverse climate periods, including extreme events; (ii) couple remotely sensed sensitivity belts with ground-based measurements of microclimate, soil moisture, and forest health; and (iii) explicitly compare the influence of elevation, stand structure, species composition, and management regimes on thermal and phenological responsiveness. Such extensions would move beyond an elevation-dominated perspective toward a more complete ecological understanding of forest climate sensitivity, thereby strengthening the evidence base for spatially explicit, climate-resilient forest planning.

5. Conclusions

This study demonstrates that thermal and phenological variability within forests is not randomly distributed but follows clear and consistent patterns across the mountainous urban fringe of Namyangju. Phenological variability in land surface temperature and vegetation activity forms continuous belts along high-elevation ridges and dissected upper slopes, while low-variability coldspots concentrate in valley floors. These patterns highlight the first-order role of terrain in shaping forest microclimates and the amplitude of seasonal vegetation dynamics. In particular, the highest-sensitivity zones were concentrated in high-elevation conifer stands, indicating that altitude can strongly amplify the responsiveness of both temperature and vegetation health, even in evergreen forests where greenness indices typically show limited seasonal fluctuation. Conversely, persistent coldspots in sheltered valleys function as microclimatic refugia where thermal conditions and phenology remain comparatively stable.

By combining time-series variability metrics with local clustering, we derived spatially explicit “sensitivity belts” that translate pixel-level satellite observations into management-relevant information. These belts delineate where forest climate-regulating functions are most vulnerable and where they are most resilient. High-sensitivity zones along ridgelines and steep slopes can guide the prioritization of adaptive management, including species selection in new afforestation, elevation-aware planting schemes, restoration of degraded stands, and fuel-load or fire-risk reduction in areas where heat and vegetation dynamics respond strongly to climatic forcing. In contrast, coldspot belts in valley floors can be recognized as core ecosystem assets for maintaining microclimate stability, supporting biodiversity, and buffering surrounding human settlements against heat stress.

From a forest-planning perspective, the spatial patterns revealed here underscore the importance of treating climate sensitivity as a heterogeneous property of the landscape rather than a uniform attribute of “forest” as a whole. Integrating sensitivity belts into regional forest plans and climate-adaptation strategies would allow managers to align monitoring, protection, and investment with the most critical zones for maintaining climate-buffering services. Although the framework is built on satellite-derived indicators, its primary contribution is ecological and practical: it provides a tractable way to see where forests most effectively regulate local climate, where those functions are at risk, and how spatially explicit forest management can reinforce the resilience of mountainous forest landscapes under ongoing climate change.