Drone-based microclimate analytics reveal vegetation–shade interaction as a key environmental drivers of Aedes-prone microhabitat suitability

Abstract

Urban microclimate interactions between vegetation and shading play a critical role in shaping environmental conditions that influence mosquito habitat suitability in tropical cities. This study presents a drone-based analytical framework to map fine-scale Aedes-prone microhabitats across contrasting urban morphologies in Shah Alam, Malaysia. High-resolution UAV imagery was used to quantify vegetation and shading patterns and integrate them into a composite ecological risk model to identify micro-environmental conditions associated with Aedes habitat suitability. Terrace housing exhibited substantially higher ecological risk compared with high-rise forms, reflecting differences in vegetation density and spatial configuration. Statistical modelling suggested that the interaction between vegetation and shading significantly improved predictive performance, highlighting their combined influence on habitat suitability. Spatial analyses further identified persistent shaded zones associated with building orientation and vegetation adjacency, indicating localized micro-environments conducive to mosquito persistence. These findings highlight a critical trade-off in urban design, where greening strategies may inadvertently support mosquito survival if not properly managed. The proposed framework provides a scalable and data-driven approach for identifying high-risk microhabitats and supports its integration into a promising complementary tool for precision surveillance, urban planning, and climate-sensitive public health strategies.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-52556-x.

Introduction

Dengue fever remains one of the most pervasive vector-borne diseases worldwide, infecting an estimated 390 million people annually, with Southeast Asia accounting for nearly 70% of the global burden^1,2. In Malaysia, dengue incidence continues to escalate despite decades of vector-control programs, with Selangor consistently recording the highest case numbers³. This persistent transmission reflects an underlying limitation in how mosquito habitats are monitored and understood within rapidly urbanizing environments. Conventional surveillance methods; larval surveys, ovitrap indices, and field inspections are labour-intensive, spatially biased, and reactive. They capture breeding foci only after outbreaks emerge, offering little predictive value⁴. Yet the ecological drivers of Aedes proliferation, including vegetation density, shade persistence, and humidity, vary at fine spatial scales that traditional tools cannot capture^5,6. To close this gap, computational geospatial approaches are increasingly employed to integrate environmental, morphological, and climatic variables into predictive surveillance systems⁷.

Recent advances in drone-based remote sensing now allow centimetre-scale observation of urban environments, revealing relationships between built form, vegetation, and microclimate that govern Aedes habitat suitability^8–10. Previous studies have suggested the application of remote sensing and aerial survey technologies for mosquito surveillance across tropical and subtropical regions¹¹. Satellite-based analyses in Southeast Asia and Latin America have been used to map environmental suitability for Aedes mosquitoes by integrating land surface temperature, vegetation indices, and urban land-use patterns¹². More recently, UAV-based approaches have enabled finer-scale detection of potential breeding habitats, including water-holding containers, vegetation clusters, and shaded micro-environments within residential areas¹³. In tropical urban settings with monsoon-driven climatic variability, seasonal fluctuations in rainfall, temperature, and vegetation growth have been shown to influence mosquito population dynamics and habitat persistence¹⁴. Remote sensing platforms, particularly UAVs, offer the capability to capture these temporal and spatial variations, supporting dynamic surveillance frameworks. However, most existing studies rely on coarse-resolution imagery or focus on macro-scale environmental predictors, with limited attention to microclimatic interactions such as vegetation–shade coupling at the neighbourhood scale^15,16.

However, most remote-sensing applications in dengue ecology still rely on coarse-resolution satellite imagery (e.g., MODIS, Sentinel-2) or city-level land-cover classes, which cannot resolve household-scale breeding environments such as drains, courtyards, and backyard vegetation^17,18. In Malaysia, this limitation is particularly evident in mixed morphologies where terrace and high-rise housing coexist, generating diverse microclimatic and shading conditions that influence mosquito survival differently¹⁹. Paradoxically, urban greening initiatives intended to enhance heat-mitigation and climate resilience can also intensify dengue risk by expanding shaded, humid refugia near dwellings^20,21. Addressing this paradox requires an analytical framework that explicitly links vegetation structure, shading intensity, and urban form within a computational modeling environment. Such integration aligns with the CEUS research agenda, which emphasizes data-driven urban analytics and environmental decision-support systems.

This study introduces a drone-based computational framework for mapping Aedes-prone microhabitats in Seksyen 24, Shah Alam, Selangor. High-resolution RGB orthomosaics were processed to derive vegetation (Excess Green Index, ExG) and shading (brightness) indices, combined into a Composite Ecological Risk Index (CERI) to quantify habitat suitability across contrasting morphologies^22–24. The analysis integrates shade–vegetation interaction modeling, kernel-density hotspot detection, and directional shade analysis to capture micro-ecological variations that conventional surveys overlook²⁵.

Materials and methods

Study area

The study was conducted in Seksyen 24, Shah Alam, Selangor, Malaysia (3.02° N, 101.31° E) a dengue-endemic urban sector situated within the Klang River Basin (Fig. 1). The area experiences a tropical humid climate, with a mean annual temperature of approximately 27 °C and monthly rainfall exceeding 200 mm²⁶. The study area was selected to capture a representative range of urban residential morphologies commonly found in Malaysian cities, including high-rise, medium-rise, and terrace housing types. Seksyen 24, Shah Alam, provides a heterogeneous urban landscape within a compact spatial extent, allowing controlled comparison of vegetation density, shading patterns, and built-form configurations across contrasting morphologies. This diversity makes the area a suitable testbed for examining micro-scale environmental drivers of Aedes habitat suitability. Four representative residential morphologies were selected to capture urban heterogeneity in built form, vegetation cover, and shading intensity. Site A (Flat B) represents a compact high-rise complex (8–12 floors) with low vegetation coverage (approx. 38%) and a high proportion of impervious surfaces (> 70%). Site B (Flat H) comprises medium-rise residential blocks (4–6 floors) with moderate vegetation coverage (approx. 38%) and semi-enclosed shaded courtyards. Site C (Teres B) consists of dense terrace housing (1–2 floors) characterized by relatively low vegetation coverage (approx. 31%), high building density (approx. 60%), and narrow back-lane configurations. In contrast, Site D (Teres D) represents low-density terrace housing (1–2 floors) with higher vegetation coverage (approx. 50%), lower building density (approx. 45%), and more open spatial layouts with mature trees and grassed areas. Detailed quantitative characteristics of each morphology type are provided in Table 1. These contrasting morphologies reflect typical Malaysian urban typologies where differential shading and vegetation configurations influence Aedes habitat suitability. The four residential morphology types were classified based on a combination of building structure, spatial configuration, and vegetation characteristics derived from UAV orthomosaics and supported by cadastral data from the Shah Alam City Council (MBSA). Classification criteria included building height (high-rise vs. low-rise), housing density, spatial arrangement (clustered vs. linear), and the extent of surrounding vegetation. These features were visually and spatially distinguishable within the study area, allowing clear separation of morphology types with minimal overlap. The selected categories represent common residential typologies in Malaysian urban environments, facilitating comparative analysis of micro-environmental conditions across contrasting built forms. Site boundaries and coordinates were georeferenced to Shah Alam City Council cadastral data and verified against UAV orthomosaics for spatial accuracy.

Fig. 1.

Location of the study area in Seksyen 24, Shah Alam, Selangor, Malaysia. The study boundary is delineated on a UAV-derived orthophoto. All spatial data and maps were generated by the authors using UAV imagery processed in Agisoft Metashape Professional version 2.0 (https://www.agisoft.com/) and visualized using ArcGIS Pro version 3.1 (https://www.esri.com/) and QGIS version 3.34 (https://qgis.org/). No third-party satellite imagery was used.

Table 1.

Quantitative characteristics of residential morphology types.

Parameter	Flat B (High-rise)	Flat H (Medium-rise)	Teres B (Dense terrace)	Teres D (Low-density terrace)
Building height (floors)	8–12	4–6	1–2	1–2
Building coverage (%)	~ 75	~ 65	~ 60	~ 45
Vegetation coverage (%)	~ 38	~ 38	~ 31	~ 50
Shaded vegetation (%)	12.6	9.4	6.7	11.9
Mean CERI	2.43 ± 0.38	2.08 ± 0.35	3.24 ± 0.51	3.87 ± 0.42
Spatial configuration	Compact cluster	Semi-clustered	Linear dense	Open + vegetated

Note: (~) approx.

UAV image acquisition and pre-processing

High-resolution imagery was acquired using a DJI Phantom 4 RTK UAV equipped with a 20-megapixel RGB camera (1-inch CMOS sensor, 8.8 mm focal length). A total of four UAV survey flights were conducted, with one flight per residential morphology site, to ensure consistent spatial coverage and minimize variability in environmental conditions. All flights were performed between 0900 h and 1100 h during clear-sky conditions to reduce illumination variability and shadow distortion. All UAV operations were conducted over public spaces, and no personal or identifiable data were collected. Flights were performed during daylight under clear weather conditions, maintaining safe altitude and operational distances in accordance with the Manual of Aerodrome Standards (DCA Malaysia, 2022).

Flight missions followed a boustrophedonic (grid-based) pattern with 80% forward and 70% side overlap at an altitude of 80 m, resulting in a ground sampling distance (GSD) of approximately 2.5 cm per pixel. The total area covered by the orthomosaics across all sites was approximately 2.1 hectares, encompassing representative sections of each residential morphology. Images were processed in Agisoft Metashape Professional 2.0 (Agisoft LLC, St. Petersburg, Russia) to generate orthorectified mosaics and Digital Surface Models (DSMs), with radiometric normalization and geometric correction applied to ensure consistency and spatial accuracy. The final orthomosaics were exported as GeoTIFFs (EPSG: 4326 – WGS 84) for spatial analysis in ArcGIS Pro 3.1 (Esri, Redlands, CA, USA) and QGIS 3.34 (QGIS Development Team, Open Source Geospatial Foundation, USA). Georeferencing accuracy was evaluated based on the expected performance of the RTK-enabled UAV system and standard photogrammetric processing workflows. Previous studies report Root Mean Square (RMS) errors typically within the range of 0.03–0.10 m for RTK-based UAV mapping, which is considered sufficient for fine-scale spatial analysis in urban environments.

Computational workflow and spatial analysis

Orthomosaics were processed through a structured multi-step geospatial workflow integrating vegetation, shading, urban morphology, and ecological risk modelling (Table 2). All spatial analyses were conducted at the pixel level based on the UAV-derived ground sampling distance (GSD) of approximately 2.5 cm per pixel. Vegetation density was quantified using the Excess Green Index (ExG = 2G − R − B), a widely established RGB-based index for discriminating green vegetation in high-resolution imagery²². Vegetation density classes (dense, moderate, and low) were derived using a data-driven thresholding approach based on the distribution of ExG values. Histogram analysis was performed to identify natural breakpoints within the ExG range, and thresholds were subsequently defined to categorize vegetation into three classes. This approach ensures that classification reflects the underlying variability of vegetation intensity within the study area.

Table 2.

Summary of computational workflow for Aedes habitat-risk modeling.

Step	Objective/Parameter	Method/Formula	Software/Tool	Output/Application
Vegetation Index (ExG)	Quantify vegetation density and green cover	ExG = 2G − R − B (normalized RGB reflectance)	ArcGIS Pro 3.1/QGIS 3.34	Classified vegetation map (dense > 30; moderate 10–30; low < 10); % vegetation per morphology
Shade Detection	Identify shaded zones from brightness layer	Histogram segmentation, threshold < 100 (mean brightness)	ArcGIS Pro 3.1/Raster Calculator	Binary shade raster showing canopy/building shadows
Shade–Vegetation Coupling	Map overlap between vegetation and shaded pixels	Raster overlay (ExG > 30 ∩ Brightness < 100)	ArcGIS Pro 3.1	Shaded-vegetation coupling zones (m², %) per site
Composite Ecological Risk Index (CERI)	Integrate vegetation and shading into a continuous risk model	Weighted overlay (equal weight; 0–5 scale via min–max normalization)	ArcGIS Pro 3.1	Composite risk map (0–5 scale) classified into three categories: low (0–2), moderate (2–3), and high (≥ 3)
Morphological Correlation	Relate built-form density to vegetation and risk	Digitized building footprint (%), Pearson correlation (p < 0.05)	QGIS 3.34/IBM SPSS 29	Correlation matrix (building coverage vs. vegetation & risk)
Spatial Hotspot Detection (KDE)	Identify spatial clusters of ecological risk	Kernel Density Estimation; 15 m search radius, quartic kernel	ArcGIS Pro 3.1	Hotspot heatmap using a yellow–red gradient to represent increasing risk levels (yellow = low risk; red = high risk)
Directional Shade Gradient Analysis	Examine orientation-driven shade persistence	Solar azimuth & façade direction via DSM; 8 cardinal classes	QGIS 3.34 Solar Radiation Plugin	Rose diagrams of shade persistence (%) and vegetation adjacency (≤ 5 m)

Shaded areas were delineated using brightness thresholding derived from RGB intensity values (threshold < 100), a commonly applied method for shadow detection in high-resolution remote sensing analyses²³. The brightness threshold (< 100) was determined using histogram-based segmentation of RGB intensity values (0–255 scale), where a clear separation between shaded and non-shaded pixels was observed. This data-driven threshold is consistent with established approaches for shadow detection in high-resolution RGB imagery. Shaded zones were identified using brightness values derived from RGB imagery on an 8-bit scale ranging from 0 to 255, where lower values represent darker (shaded) areas. A threshold of < 100 was applied to delineate shaded regions based on histogram distribution.

Subsequent overlay analysis was performed to identify zones of shade–vegetation coupling, representing micro-environmental conditions conducive to Aedes habitat persistence. These coupled layers were normalized to a common scale and integrated using a weighted overlay approach to generate the Composite Ecological Risk Index (CERI). To ensure methodological transparency and reproducibility, all environmental variables were standardized prior to integration into the Composite Ecological Risk Index (CERI) using a min–max normalization approach. Continuous raster layers for vegetation (ExG) and shading (brightness) were rescaled to a common range (0–1) using the following transformation:

where X represents the original pixel value, and X_min ​ and X_max ​correspond to the minimum and maximum values of the respective raster layer.

The shading component was derived from RGB brightness values (0–255 scale), where lower values represent darker (shaded) areas. Initially, a threshold of < 100 was used to delineate shaded zones for spatial overlay analysis. For integration into the CERI model, brightness values were retained as a continuous variable and subsequently normalized using the min–max transformation. To enhance interpretability, the normalized shading layer was inverted such that higher values represent greater shading intensity:

The Composite Ecological Risk Index (CERI) was then computed using an equal-weighted linear combination of normalized vegetation and shading layers:

The resulting CERI values were rescaled to a 0–5 range for interpretability and subsequently classified into three ecological risk categories: low (0–2), moderate (2–3), and high (≥ 3). The threshold of CERI ≥ 3 was used to define high-risk areas for subsequent statistical and spatial analyses. The CERI was developed in this study as an integrative spatial metric that combines vegetation density and shading intensity, following established principles of spatial multi-criteria risk modelling²⁴. This approach enables the transformation of discrete environmental variables into a unified ecological risk surface, facilitating fine-scale identification of Aedes-prone microhabitats within heterogeneous urban landscapes. Spatial correlation between building density and vegetation–shade variables was analyzed using Pearson coefficients (p < 0.05). Kernel Density Estimation (KDE) identified hotspot clusters of ecological risk, and Directional Shade Gradient Analysis derived from DSM-based solar azimuth quantified orientation-dependent shade persistence. Kernel Density Estimation (KDE) was applied to the spatial distribution of CERI values to identify clusters of ecological risk. KDE was performed using a quartic kernel with a fixed bandwidth (search radius) of 15 m. This bandwidth was selected based on the spatial scale of residential features and typical dispersal distances of Aedes mosquitoes. Sensitivity testing using alternative bandwidths (10–20 m) yielded consistent hotspot patterns, supporting the robustness of the selected parameter. Specifically, KDE was computed using pixels classified as moderate-to-high risk (CERI ≥ 3), enabling detection of spatial Aedes-prone microhabitats. The complete stepwise analytical workflow from UAV-derived orthomosaics to final CERI classification is summarized in Fig. 2.

Fig. 2.

Stepwise computational workflow for deriving the Composite Ecological Risk Index (CERI) from UAV-derived RGB orthomosaics. The pipeline illustrates sequential processing from image acquisition and feature extraction (vegetation and shading indices) to normalization, model integration, ecological risk classification, and spatial analysis.

Data analysis

The analytical workflow transformed UAV-derived imagery into quantifiable ecological indicators of Aedes habitat suitability. Orthomosaics were processed to extract vegetation (ExG) and brightness indices, which were then combined to delineate shaded-vegetation overlaps microhabitats conducive to mosquito survival. These parameters were normalized into CERI risk scores and classified into low, moderate, and high categories. Spatial clustering (KDE) and orientation analysis quantified micro-scale patterns of habitat persistence. Statistical validation tested the association between built-form density, vegetation coverage, and ecological risk, enabling fine-scale ecological mapping for a promising complementary tool for precision surveillance. To assess differences in vegetation density, shaded–vegetation coupling, and CERI across residential morphology types, statistical comparisons were performed using the Kruskal–Wallis test due to non-normal distribution of the data. Where significant differences were detected, post hoc pairwise comparisons were conducted using Dunn’s test with Bonferroni correction. Statistical significance was set at p < 0.001. Building coverage was calculated as the proportion of built-up surface area derived from UAV orthomosaics, representing the extent of impervious surfaces within each study site. This metric is conceptually comparable to sealing layer indicators used in urban land-use datasets.

To evaluate whether vegetation–shade coupling exerted a synergistic effect, two hierarchical models were developed using pixel-level CERI values as the dependent variable. The beta regression model was specified as: logit(CERI) = β₀ + β₁(Vegetation) + β₂(Shade) + β₃(Vegetation × Shade) The logistic regression model for high-risk classification (CERI ≥ 3) was defined as: logit(P(high-risk)) = β₀ + β₁(Vegetation) + β₂(Shade) + β₃(Vegetation × Shade). The additive model (M₀) included vegetation and shading as independent predictors, whereas the interaction model (M₁) incorporated their multiplicative term (Vegetation × Shade) to capture potential synergistic effects. Model performance was assessed using both beta regression for continuous CERI values and logistic regression for high-risk classification (CERI ≥ 3). Predictor variables were standardized prior to analysis, and multicollinearity was evaluated using the variance inflation factor (VIF < 3). Comparative model performance was evaluated using Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), pseudo-R², and 10-fold cross-validation root mean square error (RMSE). Likelihood Ratio Tests (LRT) were used to assess the significance of the interaction term, with ΔAIC > 10 and p < 0.05 indicating substantial model improvement. Residual spatial autocorrelation was examined using Moran’s I to ensure independence of model residuals following spatial smoothing. All statistical analyses were conducted using IBM SPSS Statistics 29 (IBM Corp., Armonk, NY, USA).

Results

Vegetation density and potential Aedes resting habitats

Analysis of the UAV-derived RGB orthomosaics using the Excess Green (ExG) index revealed distinct spatial heterogeneity in vegetation density across the four residential morphologies in Seksyen 24, Shah Alam (Fig. 3). Terrace housing (Teres D) recorded the highest dense vegetation coverage (approx. 50%), followed by Flat H and Flat B (approx. 38%), while Teres B exhibited the lowest (approx.31%). Vegetation cover was higher in terrace areas compared to high-rise morphologies, and these differences were statistically significant (Kruskal–Wallis test, p < 0.001). Quantitative differences between morphology types are summarized in Table 1, highlighting clear variation in building density, vegetation coverage, and ecological risk. Dense vegetative structures were identified as areas with higher vegetation index values, corresponding to increased spatial coverage of vegetated zones across the study sites.

Fig. 3.

Spatial distribution of vegetation density derived from the UAV-derived RGB orthomosaics. Vegetation was classified using the Excess Green Index (ExG) into three categories: dense (> 30), moderate (10–30), and low (< 10). The bar chart illustrates the proportion of vegetation classes across the four residential morphologies.

Shade–vegetation coupling and microhabitat suitability

Overlay analysis of shaded areas (brightness < 100) and densely vegetated areas (ExG > 30) identified distinct shade–vegetation coupling zones indicative of high Aedes microhabitat potential (Fig. 4A–B). The extent of shaded–vegetation coupling varied across residential morphologies: Flat B exhibited the largest coverage (421.3 m²; 12.6%), followed by Teres D (398.7 m²; 11.9%), Flat H (315.2 m²; 9.4%), and Teres B (218.4 m²; 6.7%). The extent of shaded–vegetation coupling differed significantly across residential morphologies (Kruskal–Wallis H = 18.72, df = 3, p < 0.001), suggesting heterogeneity in the spatial distribution of coupled vegetation–shade zones. Overall, higher shaded–vegetation proportions observed in Flat B and Teres D suggested areas of greater overlap between vegetation and shading within the study sites.

Fig. 4.

Spatial distribution and quantification of shade–vegetation coupling across residential morphologies. (A) Overlay analysis derived from the integration of vegetation (Excess Green Index, ExG) and brightness-based shading layers (brightness < 100), where green-highlighted areas represent zones of shade–vegetation coupling. (B) Bar chart showing shaded vegetation area (m²) and corresponding percentage of site coverage across the four residential morphologies. All spatial layers were derived from UAV orthomosaics processed in Agisoft Metashape Professional version 2.0 and analyzed using ArcGIS Pro version 3.1. Figures were prepared by the authors without use of external satellite imagery.

Composite ecological risk and habitat suitability

The Composite Ecological Risk Index (CERI), integrating ExG and brightness layers, produced continuous risk surfaces (0–5 scale) illustrating spatial differentiation in Aedes habitat suitability (Fig. 5A–B). The mean CERI scores ranked as follows: Teres D = 3.87 ± 0.42; Teres B = 3.24 ± 0.51; Flat B = 2.43 ± 0.38; Flat H = 2.08 ± 0.35. High-risk zones (≥ 3) accounted for 46.5% of Teres D, 41.2% of Teres B, 13.4% of Flat B, and 11.5% of Flat H. The CERI values differed significantly across residential morphologies (Kruskal–Wallis H = 21.36, df = 3, p < 0.001), indicating variation in ecological risk levels among study sites. Spatially, terrace morphologies displayed risk concentrations along backyard corridors, vegetated compounds, and canopy-overlap areas features absent in compact high-rise typologies. Composite risk values above 3.0 suggested humid, shaded microhabitats highly conducive to Aedes persistence, while scores below 2.0 represent thermally exposed, low-vegetation environments.

Fig. 5.

(A) Spatial distribution of composite risk (0–5) integrating vegetation and shading layers. (B) Comparison of morphology types showing terrace environments exhibiting approx. 4× greater high-risk coverage than high-rise complexes. Spatial analysis and visualization were conducted using ArcGIS Pro version 3.1 and QGIS version 3.34. All imagery was generated from UAV data processed by the authors.

To ensure appropriate ecological interpretation, high CERI values should be understood as indicators of environmental suitability for Aedes persistence, rather than direct evidence of mosquito presence or potential breeding habitats. Specifically, areas with elevated CERI scores reflect micro-environmental conditions characterized by dense vegetation and persistent shading, which are known to support adult mosquito resting behaviour, reduce desiccation risk, and create favourable conditions for oviposition and larval survival. Thus, the identified high-risk zones represent Aedes-prone microhabitats encompassing both potential resting environments and suitable ecological conditions for breeding, rather than discrete or verified larval habitats.

Across all residential morphologies, descriptive statistics suggested substantial variability in environmental indices within the study area. The Composite Ecological Risk Index (CERI) had an overall mean of 2.91 ± 0.78, with values ranging from 1.12 to 4.76. Vegetation (ExG) values exhibited moderate variability (mean = 28.4 ± 9.6, range = 10.2–52.7), while brightness-based shading values showed a wider distribution (mean = 112.6 ± 24.3, range = 68.5–168.9). The proportion of shaded–vegetation coupling areas averaged 10.2 ± 3.1%, ranging from 4.8% to 16.7% across all sites. These results reflect the heterogeneous micro-environmental conditions across residential morphologies. Detailed summary statistics are provided in Table 3.

Table 3.

Descriptive statistics of ecological indices across all residential morphologies.

Variable	Mean ± SD	Median (IQR)	Min	Max
ExG (Vegetation Index)	28.4 ± 9.6	27.9 (21.3–34.8)	10.2	52.7
Brightness (Shade Index)	112.6 ± 24.3	109.5 (95.2–128.7)	68.5	168.9
CERI	2.91 ± 0.78	2.85 (2.21–3.54)	1.12	4.76
Shaded–Vegetation (%)	10.2 ± 3.1	9.8 (7.4–12.9)	4.8	16.7

To statistically validate these observed differences, non-parametric comparisons were performed across residential morphology types. The Kruskal–Wallis test revealed significant differences in vegetation density, shaded–vegetation coupling, and CERI values across all sites (p < 0.001). Post hoc Dunn’s tests with Bonferroni correction suggesting that terrace morphologies (Teres D and Teres B) exhibited significantly higher ecological risk compared to high-rise forms (Flat B and Flat H). These findings provide statistical support for the spatial patterns observed in the composite risk maps (Table 4).

Table 4.

Statistical comparison of ecological indicators across residential morphology types.

Variable	Statistic (H)	p-value	Post hoc (Dunn–Bonferroni)	Interpretation
Vegetation Density (ExG)	18.72	< 0.001	Teres D > Flat H, Flat B; Teres B > Flat B	Significant differences across morphologies
Shaded–Vegetation Area (%)	15.43	< 0.001	Flat B > Teres B; Teres D > Flat H	Significant variation in coupling zones
CERI (0–5)	22.91	< 0.001	Teres D > Flat H, Flat B; Teres B > Flat H	Terrace areas significantly higher risk
High-Risk Proportion (CERI ≥ 3)	20.36	< 0.001	Teres D > all; Teres B > Flat B, Flat H	Strong morphology-driven risk pattern

Note: Statistical comparison of ecological indicators across residential morphology types using Kruskal–Wallis test and Dunn’s post hoc analysis with Bonferroni correction. Significant differences were observed across all variables (p < 0.001).

Morphological determinants of ecological risk

Correlation analysis (Fig. 6) suggesting that built-form density strongly regulates ecological vulnerability. Building coverage was negatively correlated with vegetation cover (r = − 0.97, p = 0.03). However, it was not significantly correlated with high-risk areas (r = − 0.83, p = 0.17). Thus, compact morphologies suppress vegetation availability, while open layouts such as terrace housing increase exposure to ecological risk. These results suggested that higher building coverage is associated with reduced vegetation availability. However, its relationship with high-risk ecological areas was not statistically significant, suggesting that additional environmental factors may influence spatial risk patterns. These findings support the hypothesis that urban porosity and green connectivity amplify vector-suitable microhabitats.

Fig. 6.

Correlations between building density, vegetation cover, and ecological risk. (A) Strong negative correlation (r = − 0.97, p = 0.03) between building coverage and vegetation. (B) Moderate negative relationship (r = − 0.83, p = 0.17) between building coverage and high-risk area.

Statistical evidence for vegetation–shade interaction

Inclusion of the Vegetation × Shade interaction term substantially improved model performance relative to additive forms (Table 5). For the continuous CERI model, pseudo-R² increased from 0.41 to 0.56 (ΔR² = +0.15), while AIC decreased by 35.2 units (612.8 → 577.6). The observed increase in pseudo-R² (+ 0.15) and reduction in AIC (> 35 units) represent substantial improvements in model explanatory power, indicating a strong effect size associated with vegetation–shade interaction. Likelihood Ratio Testing confirmed the superiority of the interaction model (χ² = 28.9, df = 1, p < 0.001). Cross-validation error declined by 14%, confirming predictive robustness. The logistic model for high-risk pixels (CERI ≥ 3) yielded McFadden R² = 0.39 (vs. 0.27 additive), with AIC/BIC reductions of 31.4 and 25.6. The interaction coefficient (β = 0.63 ± 0.11, p < 0.001) suggested that risk increased disproportionately when both vegetation and shading were high, with predicted probabilities rising from 32% to 68%. The interaction coefficient (β = 0.63 ± 0.11, p < 0.001; 95% CI: 0.41–0.85) suggests a positive synergistic effect between vegetation and shading on ecological risk. Predicted probabilities of high-risk conditions increased from 32% (95% CI: 28–36%) to 68% (95% CI: 62–73%) under combined high vegetation and shading conditions. These metrics empirically validate that vegetation–shade synergy explains Aedes microhabitats more effectively than independent predictors.

Table 5.

Comparison of additive (M₀) and interaction (M₁) models demonstrating improved fit with the Vegetation × Shade term.

Model Type	Metric	Additive (M₀)	Interaction (M₁)	Improvement
Beta (CERI)	Pseudo-R²	0.41	0.56	+ 0.15
	AIC	612.8	577.6	–35.2
	CV-RMSE	0.284	0.244	–14%
Logistic (CERI ≥ 3)	McFadden R²	0.27	0.39	+ 0.12
	AIC/BIC	484.3/492.8	452.9/467.2	–31.4/− 25.6

Note: All predictors standardized; VIF < 3; Moran’s I ns after spatial smooth.

Spatial hotspot detection of Aedes-prone micro-environments

Kernel Density Estimation (KDE) of composite risk values revealed statistically significant clusters of ecological risk across all morphologies (Fig. 7A–D). The most intense Aedes-prone microhabitats occurred along interfaces between shaded vegetation and built-up structures zones typically associated with container accumulation and roof drains. Teres D and Flat B exhibited the highest hotspot intensities among the studied morphologies.

Fig. 7.

Spatial distribution of Aedes-prone hotspot intensity derived from KDE analysis across residential morphologies. Panels (A–D) represent different study sites. Color gradients from dark-blue–cyan suggest low values for Aedes hotspot intensity, transitioning to warmer colors (yellow–red) indicating higher hotspot intensity. Spatial analyses were performed using ArcGIS Pro version 3.1. The base spatial layers were derived from UAV orthomosaics generated by the authors.

Directional shade gradient and orientation-driven persistence

Directional Shade Gradient Analysis (Fig. 8) showed that microclimatic conditions conducive to Aedes persistence are orientation-dependent. North- and east-facing façades exhibited the highest shade persistence and vegetation adjacency (ExG > 30 within 5 m), receiving limited solar exposure during peak hours. Terrace morphologies displayed the strongest directional effects, indicating that façade orientation and geometry shape micro-scale thermal refugia that extend mosquito survival and larval development.

Fig. 8.

Directional Shade Gradient Analysis showing orientation-driven shade–vegetation interactions across morphologies. Bar length = percentage of shaded façade; color = vegetation adjacency within 5 m (ExG > 30). North- and east-facing façades maintain prolonged shading.

Morphological pattern classification and ground validation

Ground validation through field inspection and visual comparison with UAV-derived orthomosaics confirmed four principal morphological typologies that consistently aligned with model-identified high-risk zones (Supplementary Table S6). The Building Perimeter Patterns, encompassing north- and east-facing façades and corridor edges, exhibited persistent shading throughout the day, providing ideal sheltered resting environments for adult Aedes mosquitoes. Architectural Interface Patterns, typically located at building corners and junctions, created semi-enclosed microclimates with restricted air circulation and elevated surface moisture, further enhancing mosquito refuge potential. Organized Vegetation Patterns, represented by linear rows of shade trees in parking areas, produced continuous canopy cover and maintained cooler, more humid soil surfaces conducive to larval and adult survival. Meanwhile, Private Realm Patterns, including rear-yard vegetation clusters and boundary hedges, exhibited the longest shade duration and strongest vegetation–shade coupling, often in poorly maintained or secluded domestic spaces. Collectively, these morphological configurations substantiate the spatial models, demonstrating that micro-scale architectural and vegetative arrangements directly influence the persistence of shade and humidity. The strong correspondence between drone-based analytical outputs and ground observations reinforces the conclusion that fine-grained urban design particularly façade orientation, vegetation organization, and enclosure geometry plays a critical role in governing Aedes habitat suitability, persistence, and ecological resilience within urban residential landscapes.

Discussion

This study provides new spatial and computational evidence that vegetation density, shading intensity, and built-form geometry interact to shape the micro-ecological framework of Aedes mosquito habitats in tropical urban landscapes. The framework is designed as a proxy-based environmental risk assessment, capturing micro-environmental conditions associated with Aedes habitat suitability rather than directly measuring mosquito presence. These findings suggest that areas characterized by high vegetation density and persistent shading may create favorable microclimatic conditions, including reduced temperature and increased humidity, which have been associated with Aedes resting and oviposition behaviour. Through high-resolution UAV orthomosaics, we suggested that terrace morphologies with greater vegetation density and lower building coverage exhibited markedly higher ecological risk compared with compact high-rise complexes. These findings reveal a persistent paradox in tropical urban design: while urban greening mitigates heat stress, it simultaneously generates shaded and humid micro-habitats that favour Aedes persistence. Dense vegetation and sustained shading enhance adult resting sites, prevent larval desiccation, and maintain high relative humidity microclimatic conditions repeatedly associated with dengue transmission across Southeast Asia^27–30.

Importantly, the CERI framework should be interpreted as a proxy-based environmental suitability model, rather than a direct measure of Aedes population density or breeding site confirmation. The index integrates vegetation density and shading intensity to identify micro-environmental conditions that are ecologically favourable for mosquito persistence. High CERI values therefore represent general environmental suitability, encompassing both potential adult resting habitats and conditions conducive to oviposition and larval development, rather than explicitly delineating active potential breeding habitats. This distinction is particularly important in the absence of direct entomological validation and underscores the role of CERI as a complementary spatial decision-support tool for identifying priority areas for targeted surveillance and intervention.

The identification of shaded-vegetation overlap zones (brightness < 100; ExG > 30) as high-risk habitats aligns with spatial entomology research showing that vegetation continuity and shade persistence explain fine-scale variability in Aedes aegypti abundance^31,32. By quantifying shaded vegetation area from 218 m² in Teres B to 421 m² in Flat B and linking these directly to composite ecological risk, this study advances current understanding by translating micro-environmental conditions into measurable spatial indicators. The coupling of vegetation and shade forms localized thermal refugia that moderate surface temperature and extend larval viability factors that can now be detected, quantified, and mapped automatically using UAV analytics. This represents a methodological leap from descriptive ecology to computational environmental modeling, enabling systematic surveillance at scales previously inaccessible through field observation³³. The Composite Ecological Risk Index (CERI) revealed that terrace housing contained almost four times more high-risk area (score ≥ 3) than high-rise morphologies, suggesting that open layouts and backyard vegetation intensify habitat suitability. These findings suggest that horizontal density and vegetation persistence may play a more influential role than building height in shaping ecological vulnerability across urban residential morphologies. These patterns mirror findings from Singapore, Bangkok, and Jakarta, where suburban terrace developments remain dengue-prone despite lower population densities^34–36. By deriving these risk surfaces from aerial imagery, this study suggests how drone-based sensing has the potential to augment, rather than replace, conventional surveillance methods with automated, predictive mapping of Aedes-prone microhabitats. Kernel-density mapping further localized statistically significant clusters along built–vegetation interfaces precisely the micro-zones where gutters, drains, and discarded containers accumulate. Such evidence supports a transition from blanket fogging toward a promising complementary tool for precision surveillance, where interventions target ecologically validated risk clusters instead of administrative boundaries. A particularly novel contribution of this work is the detection of orientation-dependent shade persistence. Directional analysis showed that north- and east-facing façades sustained the longest shade durations and greatest vegetation adjacency (ExG > 30 within 5 m). These façades, receiving less direct afternoon sunlight, maintain cooler, more humid conditions conducive to Aedes oviposition. The integration of solar geometry with ecological metrics extends drone analytics into the third dimension linking façade orientation and vertical form to habitat persistence. Such data enable 3D risk visualization, allowing local authorities to identify not just surface Aedes-prone microhabitats but also vulnerable building orientations. When integrated into municipal dashboards, this capability could inform inspection priorities, community clean-ups, and façade-specific environmental design guidelines.

These observed differences likely reflect variation in canopy maturity, vegetation density, and building configuration across residential morphologies. High-rise environments such as Flat B, despite limited ground-level vegetation, can generate extensive shaded areas through overlapping building shadows on courtyard greenery. In contrast, compact terrace forms such as Teres B may limit both sunlight interception and vegetation development, reducing the extent of shade–vegetation coupling. Areas characterized by dense vegetation (ExG > 30) and persistent shading (brightness < 100) may create localized micro-environmental conditions, including reduced temperature and increased humidity, which have been associated with Aedes resting and oviposition behaviour. The higher shaded–vegetation overlap observed in Flat B and Teres D therefore suggests the presence of more persistent microclimatic conditions that may support mosquito survival.

Formal statistical comparison between additive and interaction models suggesting that vegetation–shade coupling is the one of the key environmental drivers of Aedes-prone microhabitat suitability. Including the Vegetation × Shade term improved pseudo-R² from 0.41 to 0.56 and reduced AIC by > 35 units, indicating a substantially better model fit. Sensitivity analyses, including variation in KDE bandwidth (10–20 m) and alternative model specifications, produced consistent spatial patterns and model performance, supporting the robustness of the findings. This synergy underscores that the coexistence of dense vegetation and persistent shading amplifies ecological risk far beyond additive effects a finding consistent with the “synergistic habitat” hypothesis proposed in tropical entomological modelling²⁷. The strong performance of the interaction model validates the conceptual framework underlying the CERI, highlighting the advantage of integrated spatial indices over single-factor predictors for urban health analytics.

Collectively, these results establish drone-based ecological surveillance as a transformative decision-support tool for vector control. Conventional surveillance manual larval surveys and ovitrap inspections is reactive, costly, and spatially constrained^37–39. In contrast, UAV orthomosaics can capture large residential sectors, extract vegetation and shading indices, and generate composite risk layers within hours. This supports the creation of Dynamic Aedes Risk Maps (DARM) automated, continuously updated layers that health authorities can integrate with epidemiological databases^40,41. Such integration aligns with the World Health Organization’s Integrated Vector Management (IVM) principles, which prioritize environmental modification and data-driven decision-making over routine chemical control⁴². Furthermore, UAV surveillance reduces occupational exposure, lowers inspection costs, and ensures coverage of otherwise inaccessible zones such as high-density flats, riverbanks, or informal settlements⁴³.

From a policy perspective, institutionalizing UAV monitoring could enable local councils to implement predictive environmental management rather than reactive outbreak response⁴⁴. Regular drone flights can track vegetation growth, detect new shade accumulation, and assess compliance with sanitation bylaws⁴⁵. When complemented by community engagement, visual evidence from drone imagery can enhance public understanding of how household vegetation and waste practices influence dengue risk⁴⁶. Such participatory visualization fosters accountability and behavioral change, transforming drones from surveillance instruments into communication tools that support co-produced urban health governance⁴⁷. These findings should be interpreted within the context of model-based inference, where uncertainty and variability are inherent. This study is based on environmental and spatial modelling of Aedes-prone microhabitats and does not include direct entomological validation such as larval indices, ovitrap data, or adult mosquito counts. Therefore, the results should be interpreted as indicators of ecological suitability rather than suggesting vector presence. Future studies should integrate UAV-derived environmental metrics with field-based entomological data to validate and enhance predictive accuracy.

Despite these advances, several limitations must be acknowledged. While the study area encompasses diverse residential typologies, it represents a single urban locality and may not fully capture the variability present across different cities or climatic regions. Therefore, the findings should be interpreted as context-specific. Future studies should extend this framework to multiple urban settings to evaluate its generalizability and robustness under varying environmental and morphological conditions. First, this study inferred ecological risk solely from environmental proxies without concurrent entomological validation (e.g., larval or adult density). Future work should pair UAV-derived indices with field sampling to calibrate risk thresholds; for instance, correlating CERI ≥ 3 with an ovitrap index > 10%. Second, the analysis was cross-sectional, capturing a single temporal snapshot; seasonal drone surveys could reveal how vegetation phenology and solar trajectories alter habitat persistence across wet and dry seasons. Third, the study focused on one urban section of Shah Alam; applying the framework to multiple districts would strengthen generalizability and facilitate regional risk benchmarking. Lastly, although RGB imagery is cost-efficient, it cannot measure surface temperature or moisture directly.

Conclusion

In conclusion, this study suggested the potential of drone-enabled spatial analytics to enhance how urban environments are monitored, modelled, and managed in dengue-endemic settings. Terrace neighborhoods characterized by dense and persistent shaded vegetation were identified as the most ecologically vulnerable, whereas high-rise complexes exhibited more localized risk concentrated at built–vegetation interfaces. These findings highlight the critical role of vegetation–shade interactions and urban morphology in shaping fine-scale Aedes habitat suitability. By integrating high-resolution orthomosaic imagery with computational indices of vegetation and shading, the proposed framework provides a replicable and data-rich approach for identifying micro-scale ecological risk. This enables more precise, spatially targeted interventions compared to conventional surveillance methods. The ability to quantify and map these interactions further supports the development of Dynamic Aedes Risk Mapping (DARM), translating environmental data into actionable intelligence for vector control.From a practical perspective, incorporating UAV-based analytics into dengue prevention strategies could support early risk detection, prioritization of high-risk zones, and more informed urban planning decisions. Ultimately, this approach offers a pathway toward climate-sensitive, data-driven, and vector-smart urban systems that balance green infrastructure development with public health resilience.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Author contributions

N.C.D. : Conceptualization, Methodology, Data curation, Formal analysis, Writing – original draft, Writing – review & editing, Supervision. A. N. : Conceptualization, Writing – review & editing, Validation. Z.M.: Conceptualization, Methodology, Writing – original draft, Writing – review & editing. S.A.S.: Data extraction, Visualization, Formal analysis. Pradeep Isawasan: Writing – original draft, Resources. M. A. O.: Writing – review & editing, Project administration.

Funding

This research did not receive any external funding and was fully supported by the internal resources of the Faculty of Health Sciences, Universiti Teknologi MARA (UiTM).

Data availability

The datasets generated and analyzed in this study are not publicly available due to privacy and geolocation restrictions associated with drone-captured imagery in residential areas. Processed orthomosaic layers, vegetation (ExG) indices, and composite ecological risk maps can, however, be shared by the corresponding author upon reasonable request for research or educational purposes, subject to approval by the Universiti Teknologi MARA Research Ethics Committee (REC/FSK/2025/001). Derived analytical codes and model scripts used for vegetation–shade coupling, kernel density estimation, and composite ecological risk index (CERI) computation were developed in ArcGIS Pro 3.1, QGIS 3.34, and IBM SPSS v29, and are available from the authors on request to facilitate replication and further methodological development. All data were stored within UiTM’s secure institutional repository and managed under the Research Data Management Policy (UiTM-RMC, 2024), ensuring compliance with national data-protection and research-integrity standards.

Declarations

Competing interests

The authors declare no competing interests.

Ethical and regulatory compliance

All operations complied with Universiti Teknologi MARA (REC/FSK/2025/001) ethics requirements for environmental and UAV research. Drone missions followed the Civil Aviation Directive CAD 6011 Part XVI (Unmanned Aircraft Systems – Malaysia, 2023) and were approved by both the Civil Aviation Authority of Malaysia (CAAM) and Shah Alam City Council (MBSA).