Spatial prioritization of vegetation-based climate adaptation pathways in tropical megacity Jakarta, Indonesia

Abstract

Rapid urbanization in tropical megacities intensifies surface warming, yet spatial heat diagnostics frequently lack alignment with administratively actionable adaptation planning. This study develops and validates a regression-informed spatial prioritization framework for vegetation-based climate adaptation using Jakarta, the capital city of Indonesia, as a case study. Harmonized pixel-level layers of Land Surface Temperature (LST), vegetation deficit, built-up intensity, and population density were derived from multisource satellite and reanalysis datasets and integrated through cross-sectional ordinary least squares regression. Contribution-based and equal redistribution scenarios, combined with threshold relaxation tests, were applied to evaluate prioritization stability. High-LST clusters persistently coincide with dense built-up corridors, and regression results indicate that built-up intensity exhibits the strongest standardized association with surface temperature, while vegetation deficit retains an independent contribution. Priority tiers remain structurally consistent across alternative weighting and threshold configurations, demonstrating robustness of administratively aggregated heat cores. The empirical correspondence between LST and Universal Thermal Climate Index supports the planning relevance of surface-based diagnostics in dense tropical environments. By linking spatial analytics to governance-relevant intervention tiers, the proposed framework bridges the gap between urban heat assessment and phased adaptation planning. The approach offers a transferable model for evidence-informed vegetation-based adaptation in rapidly urbanizing tropical cities.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-50579-y.

Introduction

Urbanization modifies surface–atmosphere energy exchanges, frequently generating warmer conditions in cities relative to their rural surroundings, a phenomenon classically described as the Urban Heat Island (UHI)^1–6. Traditionally, UHI intensity is defined as the synchronous air-temperature difference between urban and rural reference sites under comparable meteorological conditions. While this metric has been foundational in urban climate science, it does not fully represent the spatial complexity of heat exposure within cities nor the effectiveness of adaptation measures^3,7,8. Intra-urban thermal variability, influenced by land-cover heterogeneity, impervious materials, vegetation structure, anthropogenic heat emissions, and urban morphology, can equal or exceed the urban–rural contrast itself^9–11. Recent scholarship further argues that reducing UHI intensity is not synonymous with reducing urban heat risk; mitigation and adaptation efforts must prioritize lowering the absolute thermal burden experienced by residents rather than solely modifying relative urban–rural gradients^12–14. These concerns are particularly salient in humid tropical megacities, where high background temperatures, monsoonal dynamics, and rapid land transformation amplify heat exposure and intensify localized thermal accumulation^15–17. In such environments, understanding the spatial configuration of urban heat is essential for climate adaptation planning.

Urban heat dynamics in tropical regions require explicit consideration rather than extrapolation from temperate-climate studies¹⁸. In the Köppen–Geiger classification, tropical climates (type A) include rainforest (Af), monsoon (Am), and savannah (Aw/As) regimes, distinguished primarily by precipitation seasonality^19,20. These tropical environments are characterized by persistently high temperatures, elevated humidity, strong solar radiation, and complex land–atmosphere interactions. Recent synthesis¹⁸ further indicates that UHI effects in tropical cities may persist throughout the diurnal cycle and interact strongly with convection and precipitation processes. In Jakarta, long-term observational records demonstrate a temperature increase of approximately 1.6 °C per century²¹, exceeding the global average, along with intensification of extreme precipitation and significant shifts in diurnal rainfall patterns associated with rapid urbanization. These findings highlight that tropical urban climates are shaped by coupled thermal and hydrometeorological processes, reinforcing the need for spatially explicit and policy-relevant UHI assessments.

The manifestation of urban heat depends strongly on the metric and scale of analysis. Canopy-layer air temperature contrasts capture classical UHI intensity, whereas satellite-derived land surface temperature (LST) provides spatially continuous representations of surface heat patterns^22,23. Although surface and air temperatures often co-vary, they are not interchangeable, as surface thermal responses are directly shaped by land-cover composition, material properties, and radiative exchange processes^3,10. Moreover, human thermal stress reflects combined effects of temperature, humidity, radiation, and airflow, complicating interpretation of single heat metrics^8,24,25. High-resolution and hyperlocal analyses demonstrate that heat exposure environments vary substantially within short spatial distances and that vegetated and built-up areas differentially influence experienced thermal conditions^26–29. These findings reinforce the importance of moving beyond singular measures of UHI intensity toward spatially integrated assessments that consider ecological structure, built morphology, and exposure dynamics^12,13,30. Such considerations are particularly relevant in Jakarta, where long-term observational studies indicate sustained warming trends and changing precipitation dynamics associated with urbanization²¹, and where empirical analyses have documented spatial heterogeneity in LST and heat stress across administrative districts^31–33.

Urban vegetation plays a central role in moderating surface and near-surface temperatures through evapotranspiration, shading, and alteration of surface energy fluxes^1,3,4. Numerous empirical and meta-analytical studies report consistent negative associations between vegetation indices and surface temperature across climatic regions^9,23,34–36. Neighborhood-scale modeling further indicates that achieving minimum tree cover thresholds can substantially reduce heatwave burden under projected warming scenarios³⁷. However, cooling performance is influenced not only by vegetation quantity but also by spatial configuration, canopy traits, and interactions with impervious surfaces^28,29,38. Landscape heterogeneity and patch structure shape the propagation of cooling effects⁹, while green and blue infrastructure modeling demonstrates that strategic spatial distribution enhances thermal regulation and climate resilience^39–41. These insights suggest that vegetation-based cooling is inherently spatial and context-dependent, requiring integration of ecological design principles with urban morphological constraints when formulating climate adaptation responses in dense tropical settings.

Despite these advances, much of the urban heat literature, globally and in rapidly urbanizing megacities, remains focused on heat detection, spatial mapping, and statistical association^6,22,23. Studies in Southeast Asian cities have documented links between landscape composition and surface temperature dynamics^16,42. In Jakarta, prior research has mapped LST distributions, analyzed vegetation–temperature relationships, and examined land-cover transformation effects^31,33,43. Heat stress analyses further reveal spatial disparities associated with El Niño variability and urban expansion³². Concurrently, policy analyses indicate persistent gaps between planned and realized green open space targets⁴⁴. However, comparatively fewer studies have translated such diagnostic insights into operational spatial prioritization frameworks that integrate thermal intensity, vegetation condition, built-environment constraints, and population exposure within a unified adaptation-oriented model. While green network planning and urban morphology studies have advanced understanding of configuration effects^29,30, explicit integration into climate adaptation decision-support tools remains limited, particularly in humid tropical megacities characterized by high demographic density and land scarcity. Recent tropical-climate synthesis shows that the evidence base remains uneven, with a strong concentration of studies in a limited number of large metropolitan areas, while neighborhood-scale, multi-parameter, and locally validated analyses remain comparatively scarce, particularly in Southeast Asian cities where monsoon dynamics and rapid urbanization shaping complex heat–climate relationships¹⁸.

This study develops a spatially explicit prioritization model for vegetation-based climate adaptation in the tropical megacity of Jakarta. The analysis integrates surface thermal intensity, vegetation deficit, built-environment characteristics, and population exposure into a composite framework that delineates differentiated adaptation zones. By synthesizing ecological, morphological, and demographic indicators, the study operationalizes intra-urban thermal heterogeneity into implementable vegetation-based adaptation pathways that align climate diagnostics with spatial planning feasibility^24,45. Rather than proposing generalized greening targets or network optimization alone, the framework supports targeted interventions under land and infrastructural constraints, extending prior UHI mapping, landscape heterogeneity research, and green infrastructure scholarship toward governance-relevant spatial decision support^7,46,47. Although applied to Jakarta, the methodological structure is transferable to other tropical megacities where surface urban heat intensifies rapidly under socioeconomic and morphological pressures^17,48. In doing so, the study advances urban climate scholarship by linking heat science, vegetation structure, spatial configuration, and adaptation-oriented planning within a coherent analytical framework.

Methods

Study area and climatic context

The study was conducted in Jakarta, Indonesia (6°10′–6°22′ S; 106°40′–106°58′ E), one of the largest tropical megacities in Southeast Asia (Fig. 1). The city covers approximately 662 km² and consists of five municipalities (Central, West, North, South, and East Jakarta) along with the Thousand Islands Regency. Situated on a low-lying coastal plain with elevations ranging from − 2 to 50 m above sea level, Jakarta is classified as having a Köppen–Geiger tropical monsoon climate (Am). This climate type is characterized by persistently high annual temperatures (mean 26–33 °C), elevated humidity, and strong seasonal rainfall variability driven by monsoon circulation^19,20. This classification is consistent with long-term observations showing distinct wet (November–April) and dry (May–October) seasons^21,49, as well as pronounced diurnal and seasonal variability in rainfall and temperature patterns²¹.

Fig. 1.

Administrative boundaries of Jakarta, Indonesia, showing the five municipalities, Kepulauan Seribu (one thousand islands) and major land-cover characteristics. The map was created by the authors using spatial data obtained from the Government of DKI Jakarta and processed in QGIS Desktop version 3.34.15 (QGIS Association, https://www.qgis.org.) The detailed information on Municipalities and District in DKI Jakarta are in Supplementary Method S1.

Over the past four decades, rapid urban expansion has transformed vegetated land into impervious surfaces, with built-up areas now exceeding 80% of the mainland urban footprint⁵⁰. This land-cover conversion has intensified surface heat accumulation and altered local energy balances, contributing to persistent UHI conditions documented across multiple districts^32,33. Previous climatological analyses indicate increasing surface and near-surface temperature trends, particularly in highly compact western and northern zones, where vegetation deficits coincide with high population density^21,45.

Jakarta thus represents a critical empirical case of a tropical megacity facing compound pressures from urban densification, ecological fragmentation, and climatic stress. Its spatial heterogeneity in vegetation cover, impervious intensity, and demographic exposure provides a suitable context for evaluating structured, evidence-based prioritization of vegetation-based climate adaptation strategies.

Data sources and preprocessing

Thermal and environmental indicators were derived from multisource satellite and gridded datasets summarized in Table 1. LST was retrieved from Landsat 8 Collection 2 Tier 1 (Thermal Infrared Sensor, Band 10) processed in Google Earth Engine (GEE). Imagery covering June 2024–June 2025 was filtered using a scene-level cloud cover threshold (< 20%) and composited using a median aggregation approach to reduce residual cloud contamination and short-term variability. The median composite provides a spatially stable representation of surface thermal conditions rather than a single-scene snapshot. Surface temperature was calculated using a radiative transfer formulation with NDVI-based emissivity correction, following established remote sensing procedures^22,29. Detailed equations and parameter specifications are provided in Supplementary Methods S1.

Table 1.

Datasets employed, spatial resolution, temporal coverage, and institutional sources used in this study.

Dataset	Variable derived	Spatial resolution	Temporal coverage	Source	Purpose in analysis
Landsat 8 Collection 2 Level-2 (TIRS Band 10)	Land Surface Temperature (LST)	30 m	June 2024–June 2025	USGS Earth Explorer	Surface thermal intensity assessment
Sentinel-2 Level-2 A (Bands 4, 8, 11)	Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-up Index (NDBI), Vegetation Deficit Index (VDI)	10–20 m (resampled to 30 m)	June 2024–June 2025	Copernicus Open Access Hub	Vegetation and built-up characterization
ERA5 Reanalysis	Universal Thermal Climate Index (UTCI) inputs (air temperature, humidity, wind speed, radiation)	~ 0.1° (~ 11 km) (downscaled to 30 m grid alignment)	June 2024–June 2025	ECMWF	Human heat stress consistency analysis
Population Census Data	Population Density (rasterized)	Administrative unit resampled to 30 m grid	Latest census release (2024)	BPS DKI Jakarta (2024)	Exposure assessment
Administrative Boundaries	Municipal delineation	Vector boundaries	Current administrative units (2024)	Government of DKI Jakarta	Aggregation and modal classification

Vegetation and built-up indicators were derived from Sentinel-2 MSI imagery (COPERNICUS/S2_SR_HARMONIZED) processed in GEE over the same temporal window (June 2024–June 2025). Scenes were filtered using cloudy pixel percentage < 20% and QA60 cloud masking, and a median composite was generated. NDVI and NDBI were calculated from reflectance bands, and the VDI was derived as a directional transformation of NDVI to align with heat-amplifying indicators. Meteorological variables derived from ERA5 (Table 1), including air temperature, dew point temperature, wind speed, and radiation fluxes, were harmonized to the 30 m analysis grid to ensure spatial comparability with satellite-derived indicators. Population density data from the latest BPS census were rasterized to the same spatial resolution to represent exposure, while administrative boundaries were used for district-level aggregation and modal classification.

Thermal–environmental indicators

Vegetation greenness and impervious-surface intensity were quantified using the NDVI and NDBI:

Where: SWIR = reflectance of Short-Wave Infrared (Band 11 of Sentinel-2), NIR = reflectance of Near-Infrared (for NDBI: Band 8 of Sentinel-2, and NDVI: Band 5 of Landsat 8), Red = reflectance of red wavelength (Band 4 of Landsat 8).

For NDBI, negative values typically correspond to vegetation or non-built surfaces. Because NDBI is interpreted here strictly as a built-up intensity indicator, NDBI values below zero were set to zero prior to modeling. This transformation restricts the predictor to represent only impervious or built-up influence, preventing vegetation-dominated pixels from contributing negative built-up signals. For NDVI, values below zero generally represent water bodies, shadows, or non-vegetated surfaces rather than vegetation presence. Since the analysis focuses on vegetation-related cooling effects, NDVI values below zero were truncated to zero prior to further transformation. VDI was then derived as:

This transformation ensures that higher values consistently indicate lower vegetation cover and therefore greater potential contribution to surface heating. The inversion also aligns the directionality of vegetation-related effects with other heat-amplifying predictors in subsequent modeling stages.

Population exposure was incorporated using the most recent administrative census dataset from BPS DKI Jakarta⁵⁰. Population counts were rasterized and harmonized to a 30 m spatial grid using area-weighted allocation. All spatial layers were projected to a common coordinate reference system and resampled to 30 m resolution using bilinear interpolation for continuous variables.To evaluate consistency between surface thermal burden and human heat stress conditions, the UTCI was calculated using ERA5 meteorological inputs (air temperature, relative humidity, wind speed, and radiation components). UTCI was analyzed separately from LST and no direct physical conversion between the two metrics was assumed^8,13.

UTCI estimation

UTCI was computed from ERA5-Land Hourly reanalysis data (ECMWF/ERA5_LAND/HOURLY) produced by the European Centre for Medium-Range Weather Forecasts and aggregated to mean conditions over the study period (June 2024–June 2025). Meteorological inputs included air temperature and humidity (derived from dew point) at 2 m, wind speed calculated from 10 m zonal and meridional components assuming open terrain conditions⁵¹, and surface radiation fluxes. Mean Radiant Temperature (Tmrt) was estimated from downward shortwave and longwave radiation using a Stefan–Boltzmann-based approximation with an absorption coefficient of 0.7, following established approaches^52,53. Due to data limitations, shortwave radiation was assumed isotropic, longwave radiation homogeneous, and urban shading and morphological effects were not explicitly represented.

All variables were spatially aligned to the 30 m analysis grid used for satellite-derived indicators to ensure consistency across datasets. Given the native spatial resolution of ERA5-Land (~ 0.1°), this alignment represents a statistical downscaling approach, and UTCI values should be interpreted as spatially distributed approximations of macro-scale atmospheric conditions rather than fine-scale urban microclimates. UTCI was calculated using the standard polynomial formulation⁵⁴ implemented in the pythermalcomfort library. Detailed equations, variable derivations, and computational procedures are provided in Supplementary Methods S2. The resulting UTCI values were interpreted using established thermal stress categories and used to assess the consistency between surface thermal conditions and human-experienced heat stress across the study area.

Spatial prioritization framework

A composite spatial prioritization framework was developed to delineate areas requiring vegetation-based climate adaptation by integrating thermal intensity, vegetation deficit, built-up concentration, and population exposure. All indicators (LST, VDI, NDBI, and population density) were normalized to a 0–1 range using min–max scaling to ensure comparability across variables with different units and distributions. Normalization was conducted at the pixel level across the entire study domain. The composite Priority Index (PI) was calculated as a weighted linear combination of normalized indicators:

where represents scenario-specific weighting coefficients. The inclusion of multiple indicators reflects the multi-dimensional nature of urban heat exposure, which emerges from the interaction between surface thermal conditions, vegetation scarcity, impervious surface concentration, and demographic vulnerability^12,14.

Robustness was assessed and single-parameter dominance was avoided by evaluating four weighting scenarios: (1) equal weighting; (2) moderate thermal emphasis; (3) strong thermal emphasis; and (4) high thermal emphasis. In thermal-emphasis scenarios, residual weight was redistributed proportionally among non-thermal indicators. This sensitivity analysis was designed to examine spatial stability of priority zones rather than to prescribe normative policy weighting^7,13. Detailed weight allocation matrices are provided in Supplementary Table S2.

Priority classes were determined using percentile-based thresholds applied to the composite index distribution. Pixels were categorized as Very High (> 75th percentile), High (50th–75th percentile), Moderate (25th–50th percentile), and Low (< 25th percentile). Percentile classification enables relative prioritization under heterogeneous urban conditions and avoids arbitrary fixed-value cutoffs. This framework (Fig. 2) operationalizes vegetation-based adaptation pathways as spatially differentiated planning categories derived from multi-indicator integration. The approach does not simulate canopy expansion, corridor design, or microclimatic cooling magnitude, but instead identifies relative priority zones for strategic intervention.

Fig. 2.

Methodological workflow of the spatial prioritization framework, illustrating data processing, normalization, composite index construction, weighting scenarios, and classification into priority zones.

Statistical modeling and robustness assessment

Cross-sectional global Ordinary Least Squares (OLS) regression model was applied using pixel-based observations to quantify spatial associations between land-cover characteristics and surface thermal conditions. The dependent variable was LST, while independent variables included VDI, NDBI, and rescaled population density. The model evaluates how intra-urban variation in vegetation scarcity, impervious surface concentration, and demographic exposure is statistically associated with spatial heterogeneity in LST under comparable seasonal and meteorological conditions. The regression specification is expressed as:

where denotes pixel location, is the intercept, and is the error term. Coefficients were estimated using robust standard errors to account for potential heteroscedasticity. Statistical significance was assessed at the 95% confidence level (α = 0.05), and model explanatory power was evaluated using R² and adjusted R². To enhance interpretability, vegetation effects were also expressed per 0.1 increment in NDVI.

Consistency between surface thermal burden and human heat stress indicators assessed by estimated a parallel OLS model using UTCI as the dependent variable with the same predictor set. An additional regression was conducted to evaluate the linear association between LST and UTCI. These analyses were designed to identify relative statistical associations across spatial observations rather than to simulate causal cooling magnitudes or forecast temperature reductions under hypothetical vegetation expansion scenarios.

Given the pixel-based structure of the dataset, spatial dependence among neighboring observations may exist. However, the regression framework is employed here for comparative driver assessment and prioritization support rather than spatial econometric inference. Full regression outputs and diagnostic statistics are provided in Supplementary Tables S3–S5.

Priority thresholds and adaptation pathway interpretation

Composite Priority Index (PI) values were classified using percentile-based thresholds to identify relative intervention urgency within the urban system. Pixels exceeding the 75th percentile were categorized as Very High Priority, followed by High (50th–75th percentile), Moderate (25th–50th percentile), and Low (< 25th percentile). This distribution-based approach enables adaptive prioritization under heterogeneous urban conditions and avoids fixed-value thresholds that may not generalize across spatial contexts.

Additional intervention cut-offs (PI > 0.75, > 0.65, > 0.55, and > 0.45) were evaluated to quantify changes in total area requiring vegetation-based adaptation under progressively conservative criteria to examine planning sensitivity. Pixel counts were converted to area units to facilitate interpretation for urban planning purposes.

For administrative applicability, pixel-level classifications were aggregated to municipal boundaries using modal classification. Each administrative unit was assigned the priority category most frequently occurring within its jurisdiction, with conservative tie-breaking applied toward higher priority levels to avoid underestimation of thermal vulnerability.

In this study, vegetation-based adaptation pathways refer to spatially prioritized zones where greening interventions, such as tree canopy enhancement, park expansion, or vegetated surface retrofitting, are strategically directed based on integrated thermal, land-cover, and exposure indicators. The framework does not simulate canopy expansion scenarios, corridor optimization, or microclimatic cooling magnitudes. Rather, it provides a structured spatial basis for evidence-informed planning under current observed conditions. Detailed district-level summaries and sensitivity comparisons are provided in Supplementary Tables S6–S7.

Result

Spatial patterns of surface thermal intensity and heat stress

Clear spatial gradients in surface thermal intensity emerge across the metropolitan landscape. Elevated LST is consistently concentrated within the dense urban cores of central, western, and northern Jakarta, where impervious cover dominates and vegetative continuity is limited (Fig. 3a). These zones form contiguous thermal clusters rather than isolated hotspots, indicating structurally embedded heat amplification within the built fabric. In contrast, comparatively lower LST values are observed in southern and peri-urban areas characterized by greater vegetation presence and more heterogeneous land cover.

Fig. 3.

Thermal baseline and key spatial co-locations of LST (a), UTCI (b), NDBI (c), and Vegetation Deficit (d). Note: The legend are different for each panel.

Patterns of human-perceived thermal stress, represented by the UTCI, display strong spatial coherence with the LST distribution (Fig. 3b). Areas identified as surface thermal intensification zones correspond closely with elevated UTCI values, confirming that surface heating is reflected in physiologically relevant heat exposure conditions. Although atmospheric processes introduce local variability, the spatial alignment between high LST and high UTCI demonstrates that surface thermal structure provides a meaningful approximation of urban-scale heat stress in Jakarta’s densely built districts.

VDI further delineates this gradient (Fig. 3c). Districts exhibiting high vegetation deficit coincide spatially with the most pronounced LST clusters, while areas with more continuous vegetative cover exhibit attenuated surface temperatures. Built-up intensity, measured through NDBI (Fig. 3d), reinforces this structural contrast: high NDBI values concentrate within the same central corridors where elevated LST and UTCI are observed. The co-location of high LST, high UTCI, high vegetation deficit, and strong built-up intensity reveals a consistent thermal–environmental configuration rather than fragmented anomalies.

Collectively, these spatial patterns establish a coherent thermal hierarchy across Jakarta’s urban landscape. Surface thermal intensity is not randomly distributed but aligns systematically with vegetation scarcity and built-up concentration. This spatial coherence provides the empirical basis for the regression-based weighting schemes and subsequent prioritization analyses, directly addressing the study objective of identifying vegetation-based adaptation pathways within warming urban environments.

Statistical drivers of surface thermal intensity and heat stress

Pixel-level OLS regression applied to the spatially gridded dataset reveals statistically robust associations between surface thermal intensity and urban environmental indicators. The LST model exhibits strong explanatory capacity, with all predictors reaching statistical significance (Supplementary Table S2). NDBI displays the largest standardized coefficient, indicating the strongest relative association with elevated surface temperature. VDI is also positively and associated with LST, while population density contributes positively with comparatively smaller standardized magnitude (Fig. 4a).

Fig. 4.

Empirical grounding for prioritization. Standardized coefficients bar chart for LST model (a), LST–UTCI scatter/regression (b), and Vegetation deficit association plots with UTCI (c) and LST (d).

The standardized coefficient structure quantitatively confirms the spatial patterns as demonstrated in the preceding spatial analysis of vegetation deficit and thermal exposure patterns. Areas characterized by higher built-up intensity correspond to higher surface temperatures, while vegetation deficit remains independently associated with LST after accounting for built-up intensity and population density. Although population density is positively associated with LST, its standardized effect remains lower than that of built-up intensity and vegetation deficit, indicating that structural land-cover characteristics exert a stronger association with surface heating than demographic concentration alone.

The OLS model for UTCI (Supplementary Table S3) exhibits a parallel directional structure. Built-up intensity remains positively associated with higher UTCI values, and vegetation deficit and population density continue to show statistically significant relationships. However, the standardized magnitudes of these predictors are generally smaller than in the LST model, reflecting the additional influence of atmospheric variables embedded in the UTCI formulation.

The empirical linkage between LST and UTCI is further confirmed through pixel-level regression (Fig. 4B), which demonstrates a strong positive linear relationship between surface temperature and human-perceived heat stress. This relationship indicates that spatial variation in surface thermal intensity corresponds closely with physiologically relevant exposure conditions across the urban landscape. The estimated regression structure provides the quantitative basis for the weighting schemes employed in the subsequent prioritization analysis. The standardized contributions derived from the LST model inform the regression-contribution redistribution scenario, while the equal redistribution scenario serves as a comparative benchmark. The consistency of predictor directionality across both LST and UTCI models supports the internal coherence of the prioritization framework.

Spatial prioritization under alternative weighting schemes

The composite Priority Index was evaluated under alternative weighting configurations to assess how emphasizing surface thermal intensity relative to vegetation deficit, built-up intensity, and population density influences spatial targeting. Figure 5 presents the resulting priority maps under equal weighting and progressively increasing LST emphasis. Under equal weighting, high and very high priority zones concentrate within central and western urban corridors, with moderate-priority areas extending into adjacent districts.

Fig. 5.

Priority Index under weighting scenarios based on Contribution-based redistribution.

Increasing the weight assigned to LST produces incremental intensification within existing thermal clusters. Under moderate LST emphasis, high-priority classifications expand primarily within the same central districts identified in the baseline scenario. Further increases in LST weight accentuate these areas but do not introduce spatially distinct hotspots beyond the established urban cores.

The regression-contribution redistribution scenario yields a comparable geographic configuration, although the concentration of very high priority zones becomes more pronounced in districts characterized by strong built-up intensity. This pattern reflects the dominant standardized contribution of NDBI in the LST model (Supplementary Table S2), which shapes the redistribution of non-LST weights. Despite this adjustment, the principal priority corridors remain consistent across redistribution approaches.

Differences among weighting schemes are expressed primarily through shifts in classification intensity rather than through relocation of priority geography. Adjusting the relative contribution of LST modifies the proportion of pixels assigned to higher priority classes, yet the spatial configuration of dominant thermal corridors remains largely unchanged. Redistribution informed by regression-derived contributions accentuates areas characterized by strong built-up intensity, whereas equal redistribution produces a more balanced expansion; however, both approaches converge in identifying the same central urban districts as primary intervention zones. The prioritization framework therefore exhibits sensitivity in magnitude while preserving stability in spatial ordering.

Threshold sensitivity and administrative stability

The influence of classification thresholds on spatial coverage was assessed by applying progressively relaxed Priority Index cutoffs (> 0.75, > 0.65, > 0.55, and > 0.45) under both weighting schemes, and the resulting changes in areal extent are reflected in the corresponding pixel counts (Fig. 6) for both contribution-based redistribution (Fig. 6a) and equal redistribution (Fig. 6b). Reducing the threshold systematically increases the areal extent of high-priority classifications (Supplementary Tables S7a–S7b), consistent with the monotonic rise in pixel totals across cutoffs shown in Fig. 6. Under the most restrictive cutoff (> 0.75), priority zones are confined to a limited number of central districts exhibiting elevated surface temperature and built-up intensity. Lowering the threshold to > 0.65 expands the classified area primarily within adjacent urban corridors rather than generating spatially discontinuous zones. Further relaxation to > 0.55 and > 0.45 increases overall coverage while maintaining the continuity of previously identified clusters. Relaxation was extended to > 0.45 to capture contiguous upper-moderate priority areas that remain spatially coherent and thus more actionable for corridor-based interventions, while still excluding broadly low-priority pixels.

Fig. 6.

Pixel count for each threshold shown for both Contribution based redistribution (a) and Equal distribution (b).

Comparison between regression-contribution redistribution and equal redistribution indicates that threshold relaxation produces expansion within largely overlapping administrative boundaries, as indicated by the broadly comparable threshold-by-threshold pixel patterns between panels (Fig. 6a–b). Variations between schemes are reflected mainly in the extent of classified area rather than in the identity of districts selected. The districts exceeding higher thresholds remain largely consistent across redistribution approaches.

The progressive relaxation of thresholds reveals a pattern of spatial consolidation rather than structural reconfiguration. Districts identified under the strictest cutoff persist as core priority zones as thresholds are lowered, while additional areas emerge predominantly along contiguous urban corridors exhibiting similar thermal–environmental characteristics. This expansion reflects incremental inclusion of adjacent high-intensity pixels rather than redefinition of priority geography. The administrative hierarchy derived from the stability matrix (Supplementary Table S9) therefore represents graded extension of established hotspots rather than volatility in district ranking across scenarios. These empirically derived prioritization patterns provide the analytical foundation for the governance implications in the Discussion section.

Discussion

Surface vegetation configuration and urban thermal patterns

Building upon the empirically validated prioritization structure presented in the Result section, the spatial analysis reveals a pronounced association between surface vegetation configuration and the thermal heterogeneity observed across urban districts of Jakarta. As illustrated in Fig. 3a, areas characterized by low spectrally derived greenness and high built-up intensity correspond to elevated LST, whereas zones with more continuous vegetation coverage display comparatively moderated thermal signatures. These spatial contrasts are further supported by regression results (Fig. 4a), which indicate a statistically consistent negative relationship between NDVI and LST. This pattern reflects the combined influence of reduced evapotranspiration, diminished shading, and increased sensible heat flux in densely constructed environments, consistent with established urban climatology theory^9,12,22.

Vegetation in this study is represented through surface-based spectral indicators rather than explicit three-dimensional canopy structure. Although this representation does not resolve vertical morphology or aerodynamic processes, it captures spatial heterogeneity in greenness intensity and distribution at the metropolitan scale. The negative regression coefficients reported in Fig. 4a and Supplementary Table S4 confirm that planar vegetation configuration serves as a meaningful proxy for differentiating relative surface thermal intensity across dense urban fabrics. Comparable investigations have demonstrated that landscape heterogeneity and vegetation continuity significantly influence urban thermal gradients when evaluated through remotely sensed two-dimensional metrics^55,56.

The empirical relationship between LST and the UTCI further strengthens interpretive robustness. As shown in Fig. 3b and detailed in Supplementary Table S4, spatial correspondence between surface temperature and human heat stress indicators indicates that concentrated LST hotspots coincide with zones of elevated exposure risk. While LST represents radiative surface conditions and UTCI integrates atmospheric parameters relevant to physiological stress, their alignment supports the use of LST as a screening variable for city-scale prioritization, provided that interpretation emphasizes relative spatial differentiation rather than absolute heat-stress thresholds^10,27.

The results also demonstrate that thermal moderation is influenced not solely by total vegetation extent but by its spatial configuration. Districts where vegetation is fragmented into isolated patches exhibit higher LST values compared to districts where green spaces are more continuous or connected (Figs. 3a and 4a). This observation aligns with connectivity-oriented green infrastructure frameworks, which emphasize that spatial arrangement amplifies cooling potential beyond simple areal metrics^29,57. Within Jakarta, clusters of elevated LST are concentrated in compact commercial and high-density residential corridors, whereas southern and eastern districts characterized by broader and more connected green spaces display comparatively lower surface temperatures.

These findings clarify that surface vegetation configuration, measured through spectrally derived indicators, provides a sufficiently sensitive and operationally scalable representation for identifying priority intervention areas at the megacity scale. The study does not attempt to simulate neighborhood-scale microclimate dynamics; rather, it advances a spatial prioritization framework designed to guide policy-level allocation of greening interventions. By demonstrating that planar vegetation metrics robustly correspond with surface thermal intensity and exposure gradients (Fig. 4a; Supplementary Table S4), the analysis supports the integration of remote sensing–based screening tools into metropolitan adaptation planning for tropical megacities.

Spatial prioritization as an adaptation governance instrument

Beyond identifying thermal–vegetative associations, the principal contribution of this study lies in demonstrating that spatial prioritization of vegetation-based pathways can be operationalized as a stable decision-support framework for metropolitan adaptation planning. The composite index integrating LST, NDVI, NDBI, and population density produces geographically consistent priority zones across multiple threshold and weighting scenarios (Fig. 5). As shown in Supplementary Table S3a, variations in threshold levels primarily influence the spatial extent of classified priority areas rather than their geographic location. Administrative districts identified as very high or high priority under conservative thresholds remain prioritized under expanded thresholds, indicating structural robustness of the prioritization scheme.

This stability has direct implications for urban governance. In contexts where financial resources are constrained, intervention can be initiated in the most thermally vulnerable and densely exposed zones (Fig. 5). As budget al.locations increase, threshold adjustment enables systematic expansion into adjacent areas without altering the underlying spatial hierarchy documented in Supplementary Table S3a. The framework therefore supports incremental scaling rather than binary classification, allowing adaptation measures to be aligned with fiscal capacity while maintaining strategic coherence. Such scalability addresses a persistent gap in urban heat mitigation planning, where prioritization often lacks quantitative justification or spatial continuity^12,29.

The comparison between contribution-based and equal redistribution of non-LST predictors further reinforces methodological reliability. Adjusting predictor weights modifies the magnitude of composite index values and the proportional coverage of priority classes; however, it does not substantially alter the ordering of core hotspot districts (Supplementary Table S3a). This finding indicates that LST functions as a dominant structural driver at the city scale, while vegetation deficit, built-up intensity, and population exposure refine spatial differentiation within high-temperature corridors. The preservation of spatial ordering across redistribution scenarios confirms that the prioritization framework is not sensitive to moderate variations in weighting assumptions.

From a governance perspective, this robustness is essential. Urban adaptation strategies frequently operate under uncertainty regarding indicator weighting, data availability, and policy emphasis. A prioritization tool that maintains geographic consistency under alternative parameter configurations enhances institutional confidence and facilitates cross-departmental coordination. In the context of Jakarta, where urban planning, environmental management, and public works agencies operate within distinct regulatory mandates, a transparent and stable spatial framework provides a common analytical reference for intervention targeting.

The integration of demographic exposure into the composite index differentiates this approach from purely biophysical heat mapping. High-priority zones identified in Fig. 5 correspond to districts where elevated LST intersects with concentrated population density, reinforcing the alignment between biophysical risk and human exposure. This coupling strengthens the equity dimension of the framework and positions it as a policy-relevant instrument rather than a descriptive climatological exercise.

The conceptual implication extends beyond Jakarta. Tropical megacities share structural features including high imperviousness, fragmented green infrastructure, and dense residential corridors. A spatially robust, threshold-scalable prioritization framework offers a transferable methodological template for identifying intervention zones under comparable climatic and morphological conditions. By demonstrating empirical stability across weighting scenarios and administrative aggregation (Fig. 5; Supplementary Table S3a), the study advances a replicable model for embedding vegetation-based adaptation within metropolitan governance systems.

Translating spatial heat diagnostics into governance-relevant adaptation pathways

The prioritization outcomes demonstrate that spatial heat risk in Jakarta is structurally concentrated within compact central and western urban corridors rather than diffusely distributed across the metropolitan area (Fig. 5). The consistency of high-priority classifications across weighting scenarios and threshold relaxations (Fig. 6) indicates that the identified districts represent stable thermal cores rather than artefacts of parameter selection. This stability is further reinforced by the district-level robustness matrix (Supplementary Table S9), where Tier 1 districts remain persistently classified under both redistribution schemes and multiple intervention thresholds. In this tiering system, Tier 1 denotes the most consistently prioritized districts, those that emerge under the strictest cutoffs and continue to be selected as thresholds are relaxed, and Tiers 2–4 reflect progressively broader inclusion with lower persistence across scenarios.

Such spatial persistence suggests that vegetation-based adaptation in tropical megacities should not be conceptualized as uniform greening expansion but as structurally differentiated intervention. The regression results (Supplementary Tables S2–S3) show that built-up intensity exerts the strongest standardized association with surface temperature, while vegetation deficit retains an independent and statistically significant contribution. These findings align with multidimensional urban green space analyses demonstrating that spatial configuration and impervious concentration significantly shape surface thermal heterogeneity⁵⁸. However, unlike configuration-optimization studies, the present framework translates these associations into zoning-relevant priority tiers grounded in administrative feasibility.

The empirical association between LST and UTCI (Fig. 4b; Supplementary Fig. S1) further indicates that surface thermal intensity provides a meaningful approximation of physiologically relevant heat exposure in dense tropical districts. This correspondence strengthens the planning relevance of surface-based diagnostics, particularly in data-constrained urban environments where long-term canopy-layer monitoring is limited. Recent global assessments show that urban heat island effects interact with broader warming and heatwave dynamics, amplifying risk in climate-sensitive regions⁵⁹. Although this study does not model extreme-event amplification, the concentration of high-priority zones within compact urban districts suggests that such areas may face compounded exposure under future climate intensification.

Importantly, the stability of priority districts across contribution-based and equal redistribution scenarios (Supplementary Tables S8a–S8b; S9) demonstrates that the framework is robust to normative weighting assumptions. Redistribution informed by regression-derived contributions accentuates central built-up districts, whereas equal redistribution yields more spatially dispersed classifications; yet both approaches converge on the same Tier 1 and Tier 2 administrative units. This convergence supports the use of the framework as a decision-support instrument rather than a parameter-sensitive modeling exercise.

The policy implications extend to micro-scale exposure environments. Recent high-resolution investigations show that vegetated and built-up surfaces differentially shape localized heat exposure, particularly in transit corridors and pedestrian nodes^26,27. The prioritization outputs of this study similarly highlight dense transport-adjacent districts as persistent high-risk zones (Fig. 5). This spatial alignment reinforces the relevance of corridor-based greening strategies street and corridor-level shading interventions as practical adaptation measures within compact tropical megacities.

Compared to index-based hazard aggregation approaches⁶⁰, the present study advances a translational step by embedding regression-informed weighting into administrative prioritization logic. Rather than solely quantifying urban heat burden, the framework operationalizes intra-urban heterogeneity into tiered intervention zones that can be aligned with municipal development and spatial planning instruments, namely the Long-Term Regional Development Plan (in bahasa Indonesia named Rencana Pembangunan Jangka Panjang Daerah, RPJPD), Provincial Spatial Plan (Rencana Tata Ruang Wilayah, RTRW), and Detailed Spatial Plan (Rencana Detail Tata Ruang, RDTR). This integration is conceptualized in Fig. 7 as a Science–Policy–Adaptation Continuum, linking spatial diagnostics (Figs. 3 and 4), prioritization outputs (Figs. 5 and 6), and governance pathways.

Fig. 7.

Science–Policy–Adaptation Continuum for vegetation-based climate pathways in tropical megacities. Spatial maps, heat-exposure graphics, threshold-calibrated priority visualizations, and final figure composition were prepared by the authors. Selected illustrative icons were generated using AI-assisted design tools.

The contribution of the study therefore lies in demonstrating that remotely sensed LST diagnostics, vegetation deficit metrics, and exposure indicators can be systematically synthesized into a stable and administratively interpretable prioritization matrix under humid tropical conditions. While three-dimensional microclimate modeling and species-specific canopy simulations would refine cooling magnitude estimates, the current approach establishes a replicable spatial foundation for evidence-informed vegetation-based adaptation in rapidly urbanizing megacities.

Limitations and future research directions

Building on the study’s core contribution, namely, a transparent and transferable framework that translates spatial heat diagnostics into administratively interpretable priority tiers for vegetation-based adaptation (Fig. 7), interpretation of the findings should also consider the methodological boundaries of the current analysis. Several limitations should therefore be acknowledged in interpreting the findings and in guiding future extensions of the framework.

First, the analysis relies on two-dimensional surface diagnostics derived from remotely sensed indicators and gridded meteorological reanalysis. While LST and the UTCI exhibit strong spatial correspondence (Fig. 4; Supplementary Fig. S1), the framework does not incorporate three-dimensional urban morphology or detailed canopy structure. Urban geometry, vertical stratification, and aerodynamic interactions can substantially influence microclimatic cooling performance, particularly in compact districts with complex building forms^10,13. Consequently, the present study identifies spatial priority zones but does not quantify localized cooling magnitudes under specific canopy expansion or design scenarios. Integration with microclimate simulation platforms (e.g., ENVI-met or CFD-based models) would enable future studies to evaluate species-level or configuration-specific cooling performance within the identified priority districts.

Second, the prioritization model is based on cross-sectional regression using composite surfaces representing June 2024–June 2025 conditions. Although this composite approach reduces short-term variability and enhances spatial stability, it does not explicitly model extreme heat events, diurnal variation, or interannual climate variability. Evidence from global urban assessments indicates that urban heat island intensity interacts with broader warming and heatwave dynamics, amplifying risk in climate-sensitive regions^59,60. Future research should therefore incorporate event-based analyses and climate projections to assess how priority zones respond under compound thermal stress conditions.

Third, the regression framework evaluates statistical association rather than causal inference. The standardized coefficients (Supplementary Tables S2–S3) provide interpretable comparative weights for prioritization but do not isolate mechanistic surface–atmosphere energy balance processes. The contribution-based redistribution scenarios are designed to support transparent sensitivity testing rather than to estimate marginal cooling effects of specific interventions. Advanced modeling approaches, including spatial econometrics or explainable machine-learning techniques, could refine attribution while preserving interpretability for policy use⁵⁸.

Fourth, vegetation indicators are represented through NDVI-derived greenness and VDI. These metrics capture horizontal vegetation abundance but do not differentiate canopy height, leaf area index, species composition, or evapotranspiration efficiency. Empirical studies demonstrate that vegetation configuration and built-up morphology jointly shape urban heat exposure^26,27. However, the present study intentionally refrains from prescribing species-specific interventions because such prescriptions would require ecological trait datasets and process-based simulations beyond the scope of the current design. Future work could integrate tree inventory data, canopy morphology metrics, and physiological parameters to evaluate how vegetation structure modifies cooling performance within Tier 1 and Tier 2 districts.

Fifth, while population density is incorporated as an exposure proxy, the framework does not explicitly include socioeconomic vulnerability indicators such as age distribution, income stratification, or health sensitivity. Urban heat risk emerges from the interaction between hazard, exposure, and vulnerability. Expanding the prioritization model to include social vulnerability layers would strengthen its equity-oriented dimension and align the framework with emerging evidence-based adaptation guidelines for urban resilience planning⁶¹.

Despite these limitations, the framework demonstrates that multi-indicator spatial integration can produce stable, administratively interpretable prioritization zones under tropical megacity conditions. The robustness of district rankings across weighting and threshold scenarios suggests that the identified Tier 1 districts represent structurally embedded thermal cores rather than artifacts of modeling assumptions. This stability provides a foundation for phased adaptation implementation, where future refinements, such as microclimate simulation, species-level optimization, vulnerability integration, and event-based analysis, can be layered onto an already validated spatial backbone. Future research should pursue three complementary directions:

1. coupling spatial prioritization with three-dimensional microclimate modeling to estimate intervention magnitude;

2. integrating extreme-event diagnostics and climate projections to assess compound heat amplification; and.

3. incorporating socioeconomic vulnerability indicators to strengthen equity-sensitive adaptation planning.

With this extensions, the Science–Policy–Adaptation Continuum can further evolve from a prioritization instrument into a dynamic, iterative urban climate governance framework applicable across rapidly urbanizing tropical regions.

Conclusion

This study developed and validated a spatial prioritization framework for vegetation-based climate adaptation in a rapidly urbanizing tropical megacity. By integrating remotely sensed LST, vegetation deficit indicators, built-up intensity, and population exposure into a harmonized pixel-level regression and weighting framework, the analysis identified administratively interpretable priority tiers that remain stable across redistribution scenarios and threshold relaxations. This directly addresses the translational gap identified in existing urban heat research, where spatial diagnostics frequently lack alignment with administratively actionable adaptation planning units.

The empirical correspondence between LST and UTCI supports the planning relevance of surface-based thermal diagnostics in dense tropical environments, while regression-informed weighting revealed that built-up concentration exerts the strongest standardized association with surface temperature, with vegetation deficit retaining an independent and spatially coherent contribution. The persistence of Tier 1 districts under alternative weighting and threshold configurations confirms that these zones represent structurally embedded thermal cores rather than artifacts of methodological design.

Beyond diagnostic mapping, the study advances a policy-operational contribution. Because priority rankings remain stable across scenario tests, municipal authorities may adopt phased implementation strategies, initiating vegetation enhancement in Tier 1 districts under constrained budgets and expanding toward Tier 2 zones as fiscal capacity permits. This provides a flexible, budget-responsive adaptation pathway grounded in spatial evidence rather than categorical zoning alone.

While future work should incorporate three-dimensional microclimate modeling, species-level canopy characteristics, extreme-event diagnostics, and socioeconomic vulnerability metrics, the present framework establishes a robust and replicable spatial foundation for evidence-informed vegetation-based adaptation. The approach is transferable to other tropical megacities where high-resolution thermal diagnostics, land-cover indicators, and exposure data can be harmonized within a comparable regression-informed prioritization structure. By integrating spatial analytics, robustness testing, and governance translation, the study establishes a scalable model for bridging urban climate science and implementable adaptation planning under intensifying thermal stress.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

PER developed study concept, methods, draft and revised manuscript. RF and RAF collected and processed the spatial and statistical data. HSA initiated the study. DYDI prepared and revised manuscript formats and graphical illustrations. YS advised geospatial analysis and remote sensing. PAR, LS, and AF administered the study and data collection, and MS supported data compilation and policy context. All authors reviewed and approved the final manuscript.

Funding

This research was funded by the Regional Research and Innovation Center (PRID) of the Jakarta Provincial Development Planning Agency (Bappeda) (#39551516).

Data availability

All datasets are open access or available upon reasonable request from the corresponding author. The representation of the datasets is already provided in the Supplementary Annexes.

Declarations

Competing interests

The authors declare no competing interests.