2.1. Thermal adaptation theory

Pedestrian activity is significantly influenced by the surrounding thermal environment, a phenomenon explained by the thermal adaptation theory [8,17]. This theory posits that individuals adapt to their thermal environments through both internal and external mechanisms, with behavioral adjustment being a key aspect of this adaptation [9]. Behavioral adjustment refers to spontaneous, physical reactions to thermal stimuli, which profoundly impact people’s decisions regarding outdoor walking [8,18]. Behavioral adaptation unfolds across three interconnected decision levels—strategic, tactical, and operational [10,11]. At the strategic level, individuals make higher-order plans such as when to depart, where to go, and how to organize activities, typically guided by broader weather forecasts or expected conditions. At the tactical level, people adjust their routes or schedules in response to evolving on-site cues like shade availability or wind direction. Finally, operational decisions occur in real time, involving immediate physical responses such as seeking shelter, slowing pace, or temporarily stopping movement due to discomfort. Ultimately, these adaptive behaviors establish acceptable thresholds of thermal conditions for pedestrians, demonstrating their capacity for tolerating thermal stress [8].

2.2. Key factors of urban microclimate influencing pedestrian activity

Urban microclimate conditions, particularly Ta and SLR, significantly influence pedestrian activity [19,11]. While moderate air temperatures (e.g., approximately 20–25 °C) generally correlate with increased pedestrian volume, this relationship reverses during hotter summer periods, leading to a reduction in walking counts [20]. Research consistently identifies a parabolic relationship between Ta and pedestrian activity, with an optimal thermal threshold typically observed between 25°C and 30°C [12,19]. Beyond this range, walking counts sharply decline. A recent city-scale, IoT-based empirical study further supports these findings, demonstrating a nonlinear decline in walking counts as ambient temperatures exceed approximately 30°C, indicating a clear drop in comfort and outdoor activity [13].

Direct SLR intensity, encompassing factors like cloud cover and shade, also plays a crucial role, though its impacts can appear inconsistent. For instance, studies from European and North American cities often report a positive correlation between less cloud cover, higher SLR, and increased walking counts and longer walking durations [20,21]. This trend may be influenced by cultural contexts, where direct sunlight is sometimes perceived more positively or is associated with outdoor activities.

Conversely, the shade provided by built features or trees, which is associated with lower SLR, consistently tends to increase walking counts, a phenomenon particularly pronounced on extremely hot summer days [19]. This preference for shade is observed globally, highlighting a universal human response to mitigate heat stress. An observational study in Tel Aviv-Yafo found that approximately 60% of all observed pedestrians walked in available shade [15]. Crucially, the proportion of pedestrians in the shade was directly proportional to the amount of shaded areas: sites with larger shaded areas attracted more people. In locations where shaded and unshaded spaces were roughly equal in extent and function, over 70% of pedestrians chose to walk in the shade. Similar findings emerged from a controlled behavioral experiment in Singapore, which revealed that pedestrians are willing to detour significantly to stay in the shade [16]. Likewise, route-choice experiments in Tokyo indicated that about 28% of pedestrian decisions explicitly prioritized shaded pathways to mitigate health risks like heat stroke, further underscoring behavioral adaptations in response to urban heat stress [14]. Similarly, an observation study in Boston, USA found that providing shade, which mitigates heat stress, would counteract this reduction, effectively increasing willingness to walk and pedestrian accessibility, and thereby concentrating or potentially increasing walking counts in those more thermally comfortable areas [5].

2.3. Thermal adaptation, behavioral adjustment, and tolerance by walking types

People engage in walking for diverse reasons and purposes, which can broadly be categorized into recreational walking and utilitarian walking [22]. Recreational walking is a leisure activity undertaken without a specific purpose or destination, such as strolling through a neighborhood or park [22]. In contrast, utilitarian walking involves trips made to fulfill routine purposes and tasks [22]. Utilitarian walking could further be classified into walking associated with mandatory activities, including commuting to work or school, and non-mandatory activities, including going to shops [23]. Walking to work and school typically accounts for the largest share of daily trips in urban areas, while shopping is also a very common travel purpose, often encouraged by the presence of amenities like grocery stores, supermarkets, restaurants, banks, and fitness centers [23].

The fundamental difference between these two types of walking lies in their inherent flexibility and the degree to which they are constrained by external factors [24]. Mandatory utilitarian walking is often constrained by specific routes and schedules [25,26]. This inherent rigidity leaves less room for spontaneous behavioral adaptations in response to discomfort [25,26]. Non-mandatory walking has more room for thermal adaptations, such as altering the time of walking to avoid peak heat, compared to mandatory walking [25,26]. For instance, studies show that people tend to walk less to shops than to work due to the burden of carrying purchases, and trips for education and shopping/personal business are often shorter than those for work.

In stark contrast, recreational walkers have the greatest flexibility. They can freely choose longer shadier routes or adjust their walking times to enhance comfort and optimize their experience [25,26]. This differential weighting of factors means that thermal discomfort profoundly affects both decision-making and overall experience differently for each walking type [25,26]. Recreational pedestrians, valuing comfort as a primary objective, are more likely to seek out thermally comfortable environments.

2.4. Land use as a contextual moderator of walking

The distinct reasons and purposes for walking are closely linked to land use. For example, commercial areas with shops and amenities are strongly associated with utilitarian walking, while parks and waterfronts are typically associated with leisure walking [27,28]. Different land uses, therefore, inherently influence the types of walking activities occurring within them and, consequently, the thermal comfort expectations and adaptive behaviors of pedestrians in those areas. This connection between walking type, thermal tolerance, and land use highlights the importance of considering land use in studies of pedestrian activity, as it can significantly moderate the relationship between microclimatic conditions and pedestrian behavior [27,28].

2.5. Research gaps and objectives of this study

While previous research has linked urban microclimate to pedestrian activity, critical gaps remain regarding scale, heterogeneity, thresholds, and determinants. First, most studies rely on short-term, localized observations, lacking the city-wide, longitudinal perspective essential for adaptive planning against climate change. Second, the heterogeneity of pedestrian response across different land uses and trip purposes remains largely unquantified, despite its importance for place-specific mitigation strategies. Third, precise thermal and radiative thresholds (e.g., tipping points for Ta and SLR) where walking activity declines are largely unidentified, hindering the creation of actionable design guidelines. Finally, limited research has isolated the specific built-environment attributes—such as building density or openness—that most strongly condition the magnitude of shade’s cooling effect on street vitality.

To address these research gaps, this study develops a city-scale, multi-year analytical framework to quantify the nonlinear interactions between shade and street vitality. The primary goal is to specify the thermal and spatial conditions under which shade effectively sustains pedestrian activity. Specifically, this research pursues the following three objectives: (1) Objective 1: To examine how the relationship between shade and street vitality differs across various land-use contexts (e.g., commercial, residential, and green areas), accounting for varying walking purposes and heat tolerances. (2) Objective 2: To identify the specific air temperature (Ta) and solar radiation (slr) thresholds at which the buffering capacity of shade is maximized, or conversely, where its effectiveness in sustaining vitality diminishes. (3) Objective 3: To determine the built-environment and environmental attributes—such as building density, openness, and vegetation—that condition the magnitude and stability of shade’s impact on pedestrian behavior. By fulfilling these objectives, this study provides an empirical evidence base for climate-sensitive urban design, enabling planners to prioritize heat mitigation interventions that are contextually and thermally appropriate.

3.1. Study area

Incheon lies on the mid-western coast of the Korean Peninsula (37.28° N, 126.37° E), about 28 km west of Seoul. Facing the Yellow Sea, the city’s coastal setting strongly influences its climate. Although classified as humid continental, Incheon’s summers are moderated by maritime influence from Yellow Sea, with daily mean temperatures around 28.3 °C, maximums reaching 34.7 °C, and relative humidity (Rh) near 80 percent [29].

Within Incheon, this study examines Jung-gu, one of the city’s oldest and most densely developed districts (Fig 1). Jung-gu covers 140.4 km²—approximately 13.2 percent of Incheon’s total area—and with a population of about 1.36 million in 2020. The district contains a compact yet heterogeneous mix of residential, commercial, industrial, and green spaces, supporting continuous pedestrian flows along workplaces and retail corridors. Jung-gu was selected because its diverse land-use composition and morphological variation provide ideal conditions for comparing shade effects across contrasting urban contexts and identifying nonlinear thermal-behavior thresholds. Moreover, the district offers fine-resolution mobile-phone-based walking data and detailed environmental layers, enabling a direct link between walking count, modeled shadow ratio, and local weather. These attributes make Jung-gu a robust empirical setting for investigating how shade contributes to street vitality in hot-humid seasons across various urban environments.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Study area: Jung-gu District, Incheon, South Korea.

https://doi.org/10.1371/journal.pclm.0000841.g001

3.2. Methodological steps

This study integrates high-resolution spatio-temporal datasets to assess how shade, together with thermal stress, shapes street vitality in Jung-gu, Incheon. Street vitality is measured as hourly walking counts derived from anonymized LTE mobile records and aggregated to 50 × 50 m grids. The analysis targets August afternoons (12:00–17:00) on precipitation-free days over five summers (2018–2022), when thermal stress and shade variability are most pronounced.

As summarized in Fig 2, three data streams were aligned and time-matched at the grid-hour level: (i) shade/thermal conditions, (ii) street vitality, and (iii) built-environment attributes. The shadow ratio—the fraction of walkable ground in shade within each grid—was modeled using a Digital Surface Model (DSM) augmented with Normalized Vegetation Index (NDVI)-based canopy, and paired with observed meteorological variables from the Automated Synoptic Observing Stations (ASOS): Ta and SLR. Interaction terms among shade, Ta, and SLR were specified to capture joint thermal effects. Built-environment controls included the Building Coverage Ratio (BCR), Floor Area Ratio (FAR), Sky View Factor (SVF), NDVI, transit accessibility (bus and subway), elevation, and distances to parks and waterfronts.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Methodological steps.

https://doi.org/10.1371/journal.pclm.0000841.g002

Modeling proceeds in two complementary steps aligned with the study objectives. First, a random-effects panel model relates hourly walking counts to shade, Ta, SLR, and their interactions, controlling for hour- and year-specific fixed effects as well as built-environment attributes to account for unobserved temporal variation and morphological context. This step addresses Objectives 1 and 2 by testing whether shade effects on street vitality differ across land-use types and by identifying the Ta–SLR thresholds at which these effects reverse. Second, XGBoost is applied to capture nonlinearities and higher-order interactions among climatic and urban-form variables. This corresponds to Objective 3, which explores how built-environment and environmental attributes condition the magnitude and stability of shade’s influence on street vitality.

3.3. Data acquisition and processing

3.3.1. Walking count variable.

The walking count data, used as a measure of urban street vitality, were derived from anonymized LTE mobile records collected by SK Telecom, the largest wireless telecommunications provider in South Korea. SKT’s p-Cell system estimates device location at a fine spatiotemporal resolution (50 × 50 m grids, hourly) by analyzing network signals and handover patterns, which enables trajectory inference even in shadow areas such as building interiors. Duplicate records from the same user within the same cell and day were removed to ensure unique counts [30]. The City of Incheon provided the processed dataset in a map-based format, classifying users into three categories—resident (nighttime stays), working (daytime stays), and visitor—based on movement speed and dwell-time patterns inferred from network handovers, and distinguishing outdoor movement from indoor stay.

For the analysis, we used the visitor category to capture variations in street vitality as reflected in walking counts. This category was intended to represent discretionary street-level pedestrian activity, by reducing signals associated with pass-through users (cars, buses, or subways) based on short signal duration recorded at specific base stations (Statistics Data Center [31], data.mods.go.kr). Importantly, however, these classifications are inference-based rather than direct observation; therefore, walking counts should be interpreted as a proxy for street-level presence/activity and may still include some non-walking or indoor-associated signals. Walking counts were aggregated hourly for each 50 × 50 m grids. As our focus is the relationship between street vitality and heat, we confined the sample to precipitation-free afternoons (12:00–17:00), when thermal stress is highest and shade availability most meaningfully affects pedestrian behavior. This dataset provides a fine-grained spatiotemporal measure of street vitality and serves as the empirical foundation for our statistical models (Fig 3).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. August (2018–2022) hourly mean walking counts derived from LTE-mobile data.

https://doi.org/10.1371/journal.pclm.0000841.g003

3.3.2. Shade availability and thermal condition variables.

Shade availability was treated as an hour-specific environmental exposure that varies across spatial units and influences street vitality. Capturing such exposure required a spatially detailed representation of the urban environment, so we constructed a high-resolution three-dimensional surface model that combines ground elevation, building heights, and vegetation canopies. Building footprints and tree-covered pixels were masked to isolate walkable ground surfaces, allowing shadow estimation to reflect the conditions pedestrians actually encounter. Each 50 × 50 m grids (2,500 m²) was then divided into building and non-building areas, with walkable ground defined as the latter. Shading was calculated at the pixel level using this classification (S3 and S4 Tables). From these data, four shadow metrics were derived to capture different temporal and spatial dimensions of shading: (1) Instantaneous Shadow Coverage (ISC), (2) Instantaneous Non-building Shadow Coverage (INBSC), (3) Cumulative Shadow Coverage (CSC), and (4) Cumulative Non-building Shadow Coverage (CNBSC). The instantaneous measures quantify shading at each hour, while the cumulative metrics aggregate exposure across multiple hours. These complementary metrics allowed us to compare alternative representations of shade exposure within an hourly behavioral framework.

Among the four metrics, INBSC was selected as the primary shading variable because it most accurately reflects the shading conditions that pedestrians actually experience on walkable ground surfaces. Unlike ISC and CSC, which include shading on building footprints or aggregate exposure across long temporal spans, INBSC isolates only the non-building portion of each grid cell, producing a conceptually coherent and behaviorally meaningful measure of hourly environmental exposure. Across alternative model specifications, INBSC also exhibited the most stable and statistically robust associations with walking counts and thermal conditions, reinforcing its appropriateness for an hourly behavioral framework. For consistency and clarity, INBSC is reported as the main shading variable (Shadow) in all model tables, while the remaining metrics are retained in the Appendix to document the robustness of the results to alternative representations of shading. Descriptive contrasts among the four metrics are provided in S1 Fig.

Hourly shading estimates were generated following established procedures in urban microclimate and shadow modeling. Solar-geometry calculations were first used to determine the sun’s position for each observation hour, and a ray-tracing procedure classified every walkable pixel as shaded or sunlit. Digital Elevation Models (DEM) and DSM were obtained from the National Spatial Data Infrastructure Portal (www.nsdi.go.kr), providing detailed information on ground elevation and building heights [32,33]. Because DSM data do not fully represent vegetation structure, tree canopies were incorporated using Sentinel-2 imagery: pixels with NDVI values greater than 0.25 were classified as tree-covered and assigned a representative canopy height of 12 m, following standard practice in remote-sensing-based urban canopy modeling [34]. The integration of these layers produced a composite three-dimensional urban surface capable of generating spatially accurate shadow patterns at the grid scale [35,36]. Shading was simulated for all precipitation-free August days from 2018 to 2022 during 12:00–17:00, resulting in an hourly panel aligned with observations of street vitality. Ambient thermal conditions were incorporated using hourly Ta and SLR from the Incheon ASOS station of the Korea Meteorological Administration, which provided background heat load and solar radiation intensity matched to the same temporal structure (Fig 4).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Ray-tracing simulated shadow duration and hourly shadow maps (August 15, 12:00–17:00).

https://doi.org/10.1371/journal.pclm.0000841.g004

3.3.3. Built environment attribute variables.

Beyond shade and thermal exposure, street vitality was influenced by the broader physical context of the city. To capture this heterogeneity, we incorporated environmental features representing urban form, vegetation, topography, accessibility, and proximity to major green and blue spaces (Fig 5). These variables served as controls in the statistical models, ensuring that the estimated effects of shade and heat were not confounded by background morphological or locational conditions.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Spatial distribution of urban form features (BCR, FAR, SVF, and NDVI).

https://doi.org/10.1371/journal.pclm.0000841.g005

Urban form was characterized by the BCR, FAR, and SVF. BCR was calculated as the percentage of building footprint area within each grid, while FAR was derived by multiplying the floor area of each building by the number of stories and aggregating these values across the grid. Both metrics were obtained from building-footprint and floor-area records provided by the Environmental Geographic Information Service (egis.me.go.kr). SVF was estimated using the Urban Multi-Scale Environmental Predictor in QGIS, with DEM and DSM from the National Spatial Data Infrastructure (www.nsdi.go.kr) and canopy layers derived from Sentinel-2 imagery. Together, these indicators reflected the density and enclosure of the built environment, which shaped radiative exchange and natural ventilation at street level.

Vegetation was represented by the NDVI, derived from Sentinel-2 imagery collected on September 9, 2022, corresponding to peak summer greenness. Although not a direct indicator of urban form, NDVI provided a complementary measure of surface greenness that might have interacted with shade availability. Topographic features included elevation extracted from DEM data. Proximity to green and blue spaces was measured as the shortest straight-line distance from each grid centroid to the nearest large park and waterfront, reflecting access to localized cooling environments. Accessibility features were represented by network distances to bus stops and subway stations, calculated from national public data (data.go.kr). To capture nonlinear effects, these continuous distances were recoded into categorical bands (e.g., 0–100 m, 100–200 m, > 500 m), and grids without direct network access, such as port waterfronts, were assigned to the most distant category. By controlling for these environmental features—urban form, vegetation, topography, accessibility, and green/blue-space proximity—the models isolated the contribution of shade and thermal conditions to street vitality, minimizing the risk of conflating microclimatic effects with broader urban or locational characteristics.

3.4. Random-effects panel model

Panel data combines repeated temporal observations with cross-sectional variation, allowing us to track hourly street activity across 50 × 50 m grids while simultaneously accounting for heterogeneity in urban form and thermal exposure. A key methodological issue is how to treat unobserved, unit-specific heterogeneity. Fixed-effects estimators address this by absorbing grid-level intercepts, but in doing so they eliminate time-invariant covariates such as urban-form characteristics. Random-effects estimators, by contrast, retain these covariates under the assumption that unobserved grid effects are uncorrelated with the regressors; when this assumption holds, they yield more efficient estimates [37,38].

In this study, the random-effects specification was adopted because the models explicitly included time-invariant features—such as urban form, land-use type, and elevation—that were central to explaining variation in walking counts. A Hausman test confirmed the validity of the random-effects assumption, supporting both consistency and efficiency. The model was specified as follows:

where denotes the walking count for the i-th grid at hour t; denotes the shadow ratio and thermal variables, whose values vary by grid and across time; represent urban forms, environmental features, and proximity to public transportation, all of which are defined at the grid level; denotes binary indicator variables, including year and hour fixed effects as well as land-use type dummies; , , , and are the estimated parameters; denotes the coefficients for the temporal dummy variables; and is the composite error term which is the sum of individual-specific error term (), time-specific error term , and within- and between-individual error term ().

3.5. Nonlinear analysis with gradient boosting model

XGBoost was employed to capture nonlinear patterns and higher-order interactions [ 39,40,41]. This regularized gradient-boosted decision-tree algorithm iteratively fits additive trees to minimize a specified loss function, enabling efficient learning of complex response surfaces. The dependent variable was the hourly walking count, modeled using the count: poisson objective, which implements a Poisson log-likelihood with a log link. Under this specification, the model predicts the log of the expected count, and the final expected walking counts are obtained by exponentiating the model output. Predictors matched the panel specification and included the shadow ratio, Ta, SLR, land-use indicators (one-hot encoded), and built-environment variables such as BCR, FAR, SVF, NDVI, transit accessibility, elevation, and proximity to parks and the waterfront. Tree-based boosting is well suited to skewed, zero-inflated count outcomes and is robust to overdispersion because it fits residuals stage-wise rather than assuming Poisson mean–variance equality. Regularization was tuned through cross-validation and early stopping across key hyperparameters—maximum depth, minimum child weight, subsample, column sampling ratio, and learning rate—to prevent overfitting. Model performance was monitored on a held-out validation fold.

A set of additional nonlinear models was fitted to confirm the stability of the boosting-based approach. Random Forest and Extra Trees were trained using identical predictors and the same time-based train–test split to ensure that the comparative assessment reflected genuine model behavior rather than differences in data partitioning. Across these auxiliary specifications, XGBoost consistently maintained strong predictive performance while exhibiting more stable behavior across heterogeneous spatial contexts. These checks reinforced the appropriateness of XGBoost as the main framework for the nonlinear analysis.

The selection of XGBoost was also supported by methodological considerations beyond predictive performance. First, it complements the random-effects panel by learning data-driven interaction structures and nonlinear response surfaces without a priori specification. Second, it handles multicollinearity and mixed-scale predictors while retaining time-invariant morphological features (e.g., BCR, SVF) that fixed-effects estimators would otherwise absorb. Third, it performs reliably on heterogeneous spatial datasets and overdispersed counts, enabling the detection of joint temperature–radiation thresholds and form-dependent amplification. Interpretability was supported by gain-based variable importance and scenario-based marginal-effect analysis [42].

4.1. Descriptive statistics on thermal and urban conditions

August afternoons in Jung-gu, Incheon, were marked by elevated Ta, intermediate Rh, and intense SLR, creating conditions under which street shade exerted a decisive influence on outdoor activity. Across 528 hourly observations from precipitation-free August days between 2018 and 2022, mean Ta was 29.3 °C (22.4–35.7 °C), Rh averaged 60.5% (28–88%), and SLR averaged 611.2 Wh/m² (66.7–947.2 Wh/m²). These metrics captured both the prevailing thermal stress of late-summer afternoons and the intermittent periods of relief that shaped pedestrian responses. Weather data were drawn from the Incheon ASOS station (ID: 112, Korea Meteorological Administration; https://data.kma.go.kr). Table 1 summarizes these hourly thermal conditions, which defined the climatic baseline for analyzing shade– street vitality dynamics. Additional descriptive patterns of hourly shadow levels by land-use type are reported in S2 Fig, providing complementary baseline information on diurnal variation in shade availability.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive statistics on thermal conditions in Jung-gu, Incheon.

https://doi.org/10.1371/journal.pclm.0000841.t001

The study area displayed a highly heterogeneous built environmental context. The BCR averaged 0.29 (0.00–1.00), the FAR averaged 0.77 (Std.dev. = 0.91; Max = 7.56), and the SVF averaged 0.55 (0.01–1.00), reflecting diverse levels of spatial openness. Average ground-level shade coverage across grids was 29%, ranging from no shade to complete shade. The NDVI averaged 0.10 (Std.dev. = 0.08; Max = 0.46), while elevation spanned 0.28m to 100.2 m (Mean = 12.2 m). Access to environmental amenities varied: mean distances were 536.5 m to major parks and 485.9 m to the nearest waterfronts. Transit connectivity also diverged; bus stops were typically close (Mean = 168.6 m, Range: 0.05–499.2 m), whereas subway stations were far less evenly distributed (Mean = 1,386.3 m, with a maximum of nearly 5 km). Table 2 reports these descriptive statistics, which established the urban–environmental baseline against which shade effects on street vitality were assessed.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Descriptive statistics of environmental features.

https://doi.org/10.1371/journal.pclm.0000841.t002

4.2. Street shade effects on street vitality

4.2.1. Panel-model estimates across land-use types.

Shade exerted clear but highly context-contingent effects on street vitality, with the strongest positive associations observed in commercial areas, and comparatively modest or negative associations elsewhere. Table 3 presents random-effects panel estimates relating hourly walking counts to the proportion of shaded pedestrian ground, Ta (°C), SLR (Wh/m²), and their two- and three-way interactions. Models were estimated separately for commercial, residential, industrial, and green grids, with a pooled (“All”) specification for comparison. Full specifications with extended environmental and accessibility controls are reported in S2 Table, which confirms the robustness of the core results. A COVID-19 risk index was explored but excluded due to strong collinearity with year effects, yielding a more parsimonious specification.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Random-effects panel estimates of shade–thermal interactions on walking counts.

https://doi.org/10.1371/journal.pclm.0000841.t003

Across land uses, the baseline association between shade and walking counts was strongly positive only in commercial areas. Meanwhile, in other land uses it was relatively small or negative, with interaction terms—especially Shadow × SLR—accounting for most of the context-dependent gains. Commercial areas experienced the largest and most consistent benefits, with a one-percentage-point increase in shaded ground associated with approximately eight additional pedestrians per grid per hour, indicating a substantial effect magnitude relative to other land uses. By contrast, industrial areas exhibited negative baseline associations with shade, while residential and green areas showed small or statistically weak effects. These contrasts indicate that the contribution of shade is highly contingent on functional context—substantial in pedestrian-oriented commercial corridors, but weak or adverse in utilitarian or low-activity settings.

Interaction terms clarify these patterns. The Shadow × SLR interaction is positive across all land uses and strongest in commercial areas, implying that shade becomes disproportionately valuable under intense insolation. In turn, Shadow × Ta is negative in commercial areas but modestly positive in industrial areas, suggesting that rising Ta constrains the marginal benefit of shade where discretionary pedestrian flows are high, yet partially offsets penalties in work-related environments. The three-way interaction (Shadow × SLR × Ta) is uniformly negative, indicating that concurrent high radiation and temperature compress the marginal returns to shade, even where shade is otherwise advantageous.

Consistent with this interpretation, SLR independently suppresses walking counts in the commercial, residential, and pooled models, whereas Ta alone is significant primarily in industrial areas. Taken together, the results show that street shade reliably offsets insolation, but its marginal benefits diminish under elevated thermal loads—particularly when high temperature and strong radiation coincide. This motivates the subsequent analysis of temperature–radiation thresholds at which the net effect of shade reverses.

4.2.2. Temperature–radiation thresholds of shade impacts.

The influence of shade varied sharply across thermal conditions, strengthening under moderate heat but collapsing once temperature and radiation exceeded critical thresholds. Fig 6 shows the simulated marginal effect of a 1-percentage-point increase in shade over the temperature–radiation plane (Ta: 25–35 °C; SLR: 0–900 Wh m⁻²). And Table 4 summarizes the zones where the effect changes sign and statistical significance. Simulations are based on the full interaction specification of the panel model (shade, Ta, SLR, all two- and three-way interactions) with built-environment controls and temporal fixed effects, revealing pronounced nonlinear responses and thermal–radiative tipping points. Negative marginal values do not imply that shade actively suppresses walking. Rather, they indicate that the marginal benefit of additional shade becomes negligible or exhausted under extreme heat and radiation, when baseline walking activity is already very low. Physically, shade primarily reduces shortwave radiation and mean radiant temperature (mRT); however, under extreme conditions it may be insufficient to offset elevated air temperature and other heat loads, which could limit the marginal benefit of additional shade.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Thresholds for significant shade effects by Ta and SLR.

https://doi.org/10.1371/journal.pclm.0000841.t004

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Thermal thresholds of shade effect on walking counts by land use type.

* Note: Threshold lines indicate the Ta–SLR combinations at which the marginal effect of shade transitions from positive (enhancing walking) to negative (diminished marginal benefit due to low baseline activity).

https://doi.org/10.1371/journal.pclm.0000841.g006

The shade benefits were conditional, not universal. As thermal loads rose, the marginal effect of shade transitioned from positive to neutral to negative at land-use–specific Ta–SLR thresholds (Fig 6; Table 4): ~ 30 °C in commercial streets (with earlier crossings under higher SLR), ~ 31 °C in residential areas, ~ 29 °C in green spaces, and > 31 °C in industrial areas where the sign reversed to positive under high radiation. These thresholds confirmed that shade performance depended on functional context rather than exhibiting a universal benefit.

Across land uses, commercial corridors showed the clearest thresholds. Shade enhanced walking at approximately 25–27 °C, particularly under strong radiation, but the marginal benefit attenuated and crossed zero near 30 °C, after which suppression intensified. As radiation increased, the threshold shifted toward lower temperatures. In scale, fluctuations exceeded ±4 pedestrians per grid per hour in commercial settings, whereas residential, industrial, and green areas were largely within ±2. This contrast reflected the concentration of discretionary pedestrian flows in retail districts, where small shifts around thermal comfort thresholds translated into disproportionately large changes in walking counts.

Thermal threshold trajectories differed across other land uses. In residential areas, benefits persisted up to ~28 °C and turned suppressive beyond ~31 °C, with radiation amplifying both phases. In industrial districts, shade was penalizing under cooler conditions but became strongly positive above ~31 °C, and benefits strengthened with higher radiation—consistent with utilitarian, mandatory movement responding to physiological rather than comfort thresholds. Green spaces were least robust: weak positives at lower temperatures vanished quickly, suppression emerged by ~29 °C, and the negative effect deepened as radiation rose, aligning with the elasticity of recreational walking under discomfort.

These findings showed that shade’s performance was non-monotonic and jointly governed by Ta, SLR, and functional context. Commercial streets emerged as the most radiatively sensitive—combining strong gains under mild warmth with sharp reversals at extremes—while other land uses followed discretionary, utilitarian, or recreational demand logics. These threshold dynamics argued against universal prescriptions and supported context-specific strategies, with priority to high-density commercial corridors under concurrent high temperature and strong radiation.

4.3. Urban-form moderation of shade’s marginal effect in commercial streets

The marginal effect of shade on walking counts in commercial areas was strongly conditioned by built-environment and environmental characteristics (Fig 7). Here, the marginal effect of shade denoted the change in walking counts per grid per hour associated with a 1-percentage-point increase in shaded ground. Among the examined attributes, BCR and SVF exerted the most consistent and interpretable influence, whereas the FAR and the NDVI, a complementary vegetation indicator, showed comparatively weak and uncertain effects. These patterns indicated that the geometry and openness of the built environment, rather than surface greenness, determined how effectively shade sustained walking counts under thermal stress. Fig 7 visualized these relationships across varying thermal conditions (Ta × SLR), highlighting how the marginal effect sizes differed systematically by key morphological attributes.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Marginal effects of shade on walking counts across urban form and vegetation attributes under varying thermal conditions (Ta × SLR).

* Note: Derived from XGBoost residual modeling with 95% bootstrap CIs; higher values indicate greater marginal effect of a 1%p shade increase (pedestrians per grid per hour). Panels vary Ta (24, 29, 34 °C) with SLR lines (300/600/900 Wh/m²).

https://doi.org/10.1371/journal.pclm.0000841.g007

Two urban form attributes, BCR and SVF, emerged as the key determinants of shade’s marginal effect increases. The interaction with BCR followed a robust U-shaped pattern: shade yielded the strongest benefits in both sparse and highly compact street fabrics but lost its effect at intermediate densities (BCR ≈ 0.4–0.6). Under mild heat, shade remained advantageous across most contexts, yet as temperature and radiation rose, the mid-density range persistently suppressed walking counts, while low- and high-BCR settings retained relative resilience. SVF, in contrast, revealed a near-monotonic openness advantage. In highly enclosed spaces (SVF < 0.4), the marginal benefit of additional shade was limited, and in some cases showed negligible association with walking activity. This pattern may plausibly be attributed to restricted ventilation and radiative trapping constraining overall thermal comfort. At high SVF values (> 0.8), shade effects were uniformly positive, with narrow confidence bands confirming their stability even under extreme heat. Together, these two measures indicated that shade effect varied systematically with street morphology, with openness and coverage each associated with distinct spatial configurations through which shade influenced walking counts under thermal stress.

Other attributes of FAR and NDVI contributed far less to explaining shade variability. A suppressive zone appeared at moderate FAR (≈ 1.5–3), in which shade effectiveness consistently declined, while modest positive responses at very low FAR were relatively stable, similar responses at very high FAR remained highly uncertain due to wide confidence intervals. NDVI did not significantly mediate shade–walking count relationships across temperature–radiation conditions. This limited influence of greenness was consistent with the sparse vegetation typical of dense retail corridors, suggesting that urban morphology, rather than greenery, played a primary role in shaping local microclimatic conditions.

Overall, these findings underscored that shading in commercial streets was not universally beneficial but conditional on urban form. Density and openness interacted to define the boundaries of shade performance: shade faltered in mid-density, semi-enclosed environments but strengthened at the extremes of compactness and exposure. Vegetation and floor area intensity provided little additional explanatory power. Design strategies aiming to protect street vitality under rising heat should therefore focus on optimizing urban form—balancing density and openness—to enhance the microclimatic benefits of shade where street vitality was most concentrated.

5.1. Shade effects across land-use types

Panel estimates showed that the influence of shade on walking counts varied systematically with Ta and SLR. This pattern aligned with prior evidence that thermal and radiative conditions shaped pedestrian mobility [12,20,19], establishing the climatic basis on which shade exerted its influence. Building on that foundation, the land-use analysis indicated that shade’s contribution to street vitality was strongest in commercial areas but small or negative in residential, industrial, and green zones. These contrasts were plausibly explained by trip-purpose differences: commercial areas concentrated discretionary walking with higher thermal sensitivity, whereas residential and industrial areas included more commuting- and task-oriented walking, where adaptation to heat was constrained [27,23,28]. Complementing these behavioral differences, commercial corridors often have continuous shopfronts, canopies, and narrow street canyons that generate block-level contiguous shade along primary desire lines; this morphology couples shade with storefront access, making incremental shade more likely to translate into higher walking counts. In sum, shade effects differed by land use and thermal conditions, indicating where shade most effectively sustained walking and street vitality under thermal stress.

5.2. Shade effects across thermal thresholds and street vitality

The temperature–radiation analysis revealed distinct thermal responses across land uses. In commercial areas, shade supported higher walking counts under moderate heat but lost effectiveness as Ta and SLR intensified, turning negative beyond ~30 °C. This reversal was consistent with earlier studies showing that excessive heat and strong radiation suppressed outdoor walking and shifted activity toward shaded or indoor environments [12,14,21]. In physical terms, shade mainly reduced shortwave radiation and mRT but did not substantially lower ambient Ta or longwave/convective loads; therefore, when high Ta coincided with strong SLR, the marginal benefit of additional shade was limited. By contrast, industrial districts exhibited the opposite tendency: shade became increasingly beneficial at higher temperatures, suggesting that in work-related environments dominated by mandatory walking, shade helped maintain walking counts under thermal stress—albeit with smaller magnitudes than in commercial areas. This pattern also carried occupational safety and welfare implications, underscoring the value of shaded rest areas and cooling refuges for heat-exposed workers. Overall, these opposing responses showed that shade’s impact was nonlinear and context-dependent, reinforcing that temperature, radiation, and functional land use jointly governed when shade sustained—or failed to sustain—street vitality.

5.3. Urban form modulation of shade effects

Urban form appeared to moderate how shade affected walking counts in commercial streets. Benefits were strongest in very open or highly compact street fabrics and in areas with greater sky openness, whereas effects linked to floor-area intensity or surface greenness were comparatively weak. These configurations may have moderated local thermal loads in complementary ways: open forms could enhance ventilation and heat dissipation, while compact forms could increase self-shading, thereby reducing direct solar exposure at pedestrian height. Such morphological moderation helped sustain walking activity as heat and radiation intensified, consistent with outdoor-thermal-comfort research [9,18]. The mechanism also accorded with thermal-adaptation theory, whereby pedestrians adjusted at strategic, tactical, and operational levels; continuous shaded segments facilitated pacing and route choices that helped maintain street vitality [8,10]. Collectively, these results indicated that, in commercial environments, urban form did not merely provide shade but determined when and how shading configurations translated into sustained walking—demonstrating that thermal structure, behavioral adaptation, and spatial geometry jointly governed shade’s contribution to street vitality.

5.4. Policy and practical implications

These findings highlight that the impact of shade is contingent upon urban morphological and environmental conditions. Heat-resilient street design therefore required context-specific strategies rather than uniform shading guidelines. In commercial corridors—where walking counts and heat sensitivity were both high—the analysis identified a thermal vulnerability threshold around 30 °C, where street vitality declined most sharply. These segments should be prioritized in heat-resilient planning through shading configurations that balance coverage with adequate airflow while maintaining pedestrian visibility and accessibility. Evidence from Jung-gu pointed to concrete design levers: shade performed best in very open or highly compact street fabrics, indicating that mid-density, semi-enclosed environments—where ventilation and self-shading were both limited—were most vulnerable. In such areas, design should preserve sky openness, incorporate ventilation corridors, and combine shading with cooling infrastructure (e.g., water features, reflective surfaces) or time-based activity management to sustain walking during extreme heat. In industrial areas, where shade remained functionally effective at high temperatures, targeted interventions—shaded rest zones, canopies, and cooling facilities—were warranted for workers with limited behavioral flexibility. In residential and green areas, effects were weaker and inconsistent, suggesting that additional evidence is needed before strong design prescriptions are made. Overall, the results supported a locally adaptive urban-design framework that integrates shading, ventilation, and urban-form management to sustain street vitality under intensifying heat.

5.5. Limitations and future studies

First, the reliance on a single ASOS station and the absence of explicit wind variables limit the capture of intra-urban climatic variability, particularly given the coastal setting. While the study treated temperature as a common background condition and used urban morphological indicators (SVF, FAR) to indirectly proxy ventilation potential, these measures cannot fully replicate complex fluid dynamics such as channeling or sea-breeze effects. Although this approach effectively isolated the relative impact of shading, it may have simplified absolute thermal exposure. Future research should integrate high-resolution meteorological networks and Computational Fluid Dynamics (CFD) simulations to explicitly model pedestrian-level airflow and its interaction with shading.

Second, street width and functional hierarchy were not explicitly modeled, potentially introducing residual confounding. Wider arterial streets often have structurally higher pedestrian flows due to land-use intensity, connectivity, and the presence of transit nodes, while also tending to exhibit lower shadow coverage because of greater open-sky exposure. Although geometric openness and built enclosure were partially controlled through urban form indicators (e.g., SVF, BCR, FAR), these proxies do not fully capture the functional hierarchy (e.g., arterial vs. local streets) and its non-climatic walking traffic. In addition, while our mobile-based walking measure aims to exclude in-transit users, it relies on inference-based classification and thus some misclassification may remain. Future studies should integrate detailed street-network attributes such as road class, number of lanes, or width proxies from OpenStreetMap or national road classifications, to better disentangle the microclimatic benefits of shade from the functional impacts of street capacity.

Finally, inherent limitations in the activity and environmental data must be acknowledged. While LTE-derived walking counts offered extensive spatial coverage, they represent an inference-based proxy for street-level presence/activity rather than direct observation of walking behavior. Although the data provider’s processing and our use of the visitor category are designed to reduce pass-through and indoor-stay signals, residual misclassification may remain. Also, the data cannot identify specific trip purposes, meaning distinctions between recreational and industrial walking remained inferred. Furthermore, shade modeling relied on simplifying assumptions—specifically fixed vegetation heights and grid-scale NDVI—which may have muted the precision of canopy–building interactions. Combined with the specific study window (August afternoons), these constraints suggest that the observed marginal effects should be interpreted within the context of these modeling boundaries. Future work would benefit from higher-resolution canopy data and extended temporal windows to validate these findings across broader conditions.

Future study should validate these findings across diverse spatio-temporal contexts. Crucially, studies should transition from discrete variables to comprehensive thermal indices (e.g., UTCI, PET) via mRT simulations (e.g., SOLWEIG) to capture holistic heat stress. Integrating these with behavioral data (e.g., surveys, wearables) and quasi-experimental designs will strengthen causal inference, ultimately refining guidelines for climate-sensitive street design.

Nomenclature

Download:

• PNG
  larger image
• TIFF
  original image

https://doi.org/10.1371/journal.pclm.0000841.t005