Urban Spatial Heat Resilience Indicator Based on Running Activity Z-Score

Abstract

The assessment of urban heat resilience has become crucial due to increasing extreme weather events. This study introduces the Running Activity Z-score (RAZ) index based on running activity trajectory data to evaluate heat resilience. Through a case study of an August 2022 heatwave in Beijing, we examined the index’s sensitivity to extreme heat and explored its spatial relationships with key built environment factors, including plot ratio, green coverage, population density, and blue space proximity. Our results reveal two key findings: (1) the RAZ index serves as an effective real-time, high-precision indicator of urban heatwave impacts, as evidenced by extremely low RAZ values consistently coinciding with heatwave periods, and (2) the RAZ index offers valuable insights for identifying potential low heat resilience areas and supporting planning decisions, as demonstrated by its significant correlations with built environment factors that align with previous studies while uncovering more detailed spatial relationships. Although RAZ effectively complements traditional measurement methods, its application requires careful consideration of external factors such as social dynamics and climate variability.

1. Introduction

Extreme heat will threaten the health of 50–75% of the global population by 2100, according to the Intergovernmental Panel on Climate Change (IPCC) [1]. Due to the urban heat island effect, cities, which are projected to house approximately two-thirds of the population by 2050 [2], will experience heat stress that is twice as intense as surrounding rural areas. Extremely high temperatures significantly reduce work efficiency and increase heat-related deaths [3], making it crucial to measure heat resilience across urban spaces. Therefore, precise spatial measurements serve as a critical foundation for policymakers to identify vulnerable areas and design more effective and targeted interventions that enhance urban heat resilience.

Traditional heatwave impact assessments rely on two main approaches: urban climatology factors (e.g., temperature, humidity, and solar radiation) and health outcomes (e.g., medical records, clinical data, and public health surveys). Climatology factors usually measure heat exposure in urban spaces, while health outcomes reflect how citizens respond to heat stress. Citizens’ direct responses to high temperatures offer valuable insights into their perception of urban thermal environments. However, current research emphasizes medical data, which often lack real-time accessibility and spatial precision and fail to capture how heatwaves affect daily urban life. The recent rise of digital platforms provides researchers with dynamic, high-resolution data sources, particularly movement trajectories. Potentially, these new data streams can complement traditional sources, enabling deeper analysis of urban spaces through the lens of human behavior.

This study proposes a novel real-time index derived from citizen-generated outdoor running activity data to assess urban thermal environments. The index serves two purposes: (1) providing a new approach to observe the spatial impact of extreme heat events on outdoor physical activity during heatwaves and (2) identifying urban spaces that are vulnerable for outdoor physical activities during heatwaves to assist in evidence-based urban planning and heat mitigation strategies. The rest of the paper proceeds as follows: First, we introduce citizen activity data in the context of social sensing and explore its connection to urban heat resilience. The Section 3 details our data collection and data cleaning process, followed by a novel Running Activity Z-score (RAZ) index, which is derived from running trajectory data within the Sixth Ring Road of Beijing in July and August 2022. Using the early August heatwave as a case study, we examine RAZ’s sensitivity to heatwaves and its correlations with urban factors. The Section 4 further demonstrates RAZ’s potential applications for planning guidance, highlights the index’s strengths for monitoring spatial heat resilience, and addresses challenges including data stability and sampling bias.

2. Background

2.1. Heat Resilience Indicators

Urban heatwave refers to an extended period of abnormally hot weather exceeding a relative temperature threshold [1]. Extremely-high-temperature weather poses significant threats to human health by reducing motor–cognitive performances and work efficiency while damaging physical and mental health [4,5]. According to IPCC, spatial heat resilience refers to the capability to withstand extreme heat, maintain essential function, and keep the ability of adaptation. As a system characteristic, it contains three key factors: exposure, vulnerability, and adaptability [1]. In practice, the evaluation of heat resilience is based on the assessment of the above three factors [6].

Exposure indicates the intensity of heat exposure, which is usually assessed through meteorological data, including air temperature, relative humidity, and land surface temperature. Air temperature and humidity, which can be acquired through weather stations and sensors, directly correlate with human thermal perception and form common heat stress indicators like heat indicator ($E_T$) [7]. Previous studies demonstrate strong relationships between temperature, humidity, and the built environment (BE) features (i.e., density, plot ratio, green coverage) [8,9]. Land surface temperature (LST), obtained via remote sensing, shows a positive correlation with air temperature and serves as a key measure in urban heat studies [10]. Research findings confirm that higher proportions of impermeable surfaces and lower green space coverage lead to elevated LST and physiological equivalent temperature (PET) [11,12,13,14]. While high-precision meteorological data effectively reveal urban heat exposure and its relationships with the BE, supporting efforts to enhance urban heat resilience, it captures only heat exposure without showing how heatwaves impact urban spaces. Compared with heat exposure, researchers prioritize understanding a city’s ability to withstand extra heat stress and maintain normal daily activities.

Adaptability is the public’s ability to adjust to manage high temperatures, typically measured through socioeconomic factors, including income, air-conditioning usage, and electricity consumption [15]. Moreover, since the built environment factors can affect the warming of spaces under thermal exposure, they can also be used to measure the thermal adaptability of spaces [16]. Vulnerability is the propensity to be negatively affected and is always evaluated through loss of life. Current Heat Vulnerability Indexes (HVIs) primarily emphasize epidemiological data alongside factors such as exposure, demographic characteristic, socioeconomic conditions, and built environment [17]. The evaluation of vulnerability is especially important since it represents the outcome of exposure and adaptability. By analyzing the relationships among exposure, adaptability, and vulnerability, researchers can provide valuable recommendations for enhancing spatial heat resilience. Researchers have attempted to analyze the correlation between built environment, socioeconomic factors, and vulnerability indicators, including mortality and power outages [15,18]. Clinical data are widely used to assess high-temperature weather effects, including emergency department presentations [19], all-case mortality [20,21], and heat-related mortality rates [22]. Related findings proved that clinical data exhibit significant correlations with various urban factors, including green coverage [23], normalized difference vegetation index, and percentage of impermeable area [18]. Demographic characteristics also strongly influence these health outcomes, such as urbanization, education, ethnicity, income, and insurance. These correlations offer valuable insights for improving spatial heat resilience. Unfortunately, since clinical data often only reach the accuracy of communities and mostly focus on death rates and ambulance service records, their precision is limited and biased towards older age and low-income communities, leading to challenges in planning suggestions for general urban environments.

2.2. Social Media and Human Activity Analytics

Social media data reveal how extreme heat may affect citizens, serving as a valuable resource for quantifying heatwaves’ impact. For instance, a study in five Chinese megacities demonstrates strong correlations between high-temperature-related social media content and actual high-temperature weather events, with heat exposure levels influencing these correlations [24]. Another study on twitter across the UK, US, and Australia shows exponentially increased heatwave-related posts, with distinct national concerns emerging during extreme heat events [25]. Social media data offer distinct advantages over traditional data sources by enabling swift, real-time urban analysis. However, social media data typically only offer city-level geographic precision and are subject to the influence of trending topics. This may compromise the accuracy and reliability of the data, making it challenging to provide specific planning references for improving local heat resilience [26].

Outdoor physical activity (PA) offers a critical complementary measure for spatial heat resilience (SHR) by revealing heat exposure and public adaptive behaviors. Research indicates that PA correlates with various factors. Specifically, climate conditions influence participation in outdoor activities among citizens, and the extent of such influence is related to the season, weekday, and individual physical indicators [27,28]. Concurrently, BE factors affect the public’s propensity for physical activities (e.g., jogging, walking, and cycling) [29,30,31]. Regarding the impact of high temperatures, studies have revealed a significant decline in PA once the temperature exceeds a neutral heat threshold [32,33,34], particularly in terms of social activities and optional activities. Additionally, consistent with the findings in the heat exposure section, research indicates that built environments, including green spaces, are correlated with changes in running activities during extreme heat [35].

In summary, citizens’ outdoor physical activity (PA) effectively reflects changes in thermal environments and provides valuable opportunities for widespread data collection in residential areas, aiding in the monitoring of urban heat resilience. However, current studies primarily rely on manual data collection methods, which lack robust validation and pose challenges for large-scale, continuous monitoring and early warning systems.

2.3. Fitness App Data Analytics

Fitness apps have emerged as a popular data source for human activity analytics due to three distinct benefits. First, these apps show high adoption rates across populations. In France, 28% to 43% of people across age groups use digital tools to track their exercise routines [36]. Similarly, fitness apps have gained widespread popularity, reaching 41.3% adoption in China and 51.8% in America [37]. Second, fitness apps enable a cost-effective approach for collecting near real-time, high-precision information on citizen activities. Users track their exercise through smartphones and wearable devices, recording information including exercise duration, trajectories, heart rate, and speed. Such detailed data provide researchers with trajectory data on activities, supporting the study of the relationship between PA and BE.

Running trajectory data from fitness apps reveal significant relationships between running behavior and BE. Studies demonstrate positive correlations between running activity and urban features, including greenery ratio, Green View Index (GVI), Sky View Index (SVI), and housing price. Conversely, running activity decreases in areas with higher population density and more frequent street intersections. These findings illuminate the profound influence of BE on running patterns and provide valuable guidance for urban planning decisions. Despite the rich insights spatial-temporal trajectory data offers, analyzing these relationships presents four critical challenges. First, running activity and BE exhibit complex, nonlinear connections, with factors like temperature and subway station proximity with clear threshold effects [32,38,39]. This nonlinearity limits the effectiveness of traditional parametric models such as Ordinary Least Squares (OLS) and Geographically Weighted Regression (GWR). Second, the relationship between running activities and BE indicators shows substantial spatial heterogeneity across different urban areas. To enhance result credibility, researchers often manually select research units or employ weighted versions of traditional models, such as GWR, Geographical and Temporal Weighted Regression (GTWR), and Geographically Weighted Random Forests (GWRFs), to estimate stable local correlations. Third, running activity data suffer from distribution bias, characterized by uneven distribution patterns and heteroscedasticity, which compromises model reliability. Fourth, running activities respond to various external influences and often exhibit temporal lag effects, necessitating careful identification of special disturbances and exclusion of irregular data in research.

Previous research confirms correlations between PA and BE. Since high temperature influences PA patterns, researchers can investigate how BE shapes these temperature effects. Therefore, it is natural to utilize running data to monitor the phenomenon of heatwave events. There are various studies that employ time-series data to detect sudden events. Traditional approaches analyze such data through conventional time-series models to estimate predicted mortality, and then detect outliers using Z-scores [40,41]. Some researchers have advanced these methods by applying first-order difference calculations and machine learning algorithms for anomaly detection. Building upon existing research, we propose a spatial heat resilience detection indicator based on running trajectory data and explore its potential in supporting planning decisions.

3. Data and Methods

3.1. Data Collection

This study collected data on running activities and climate conditions within the city of Beijing. To begin, we obtained city street network data from OpenStreetMap (OSM) and divided our study area into 9251 grids of 500 m-by-500 m resolution based on the boundaries of the Sixth Ring Road. These grids serve as the spatial statistical units in our study. To quantify the activity intensity, running trajectory data were acquired from Keep, the largest digital fitness application in China [37], for the period of 17 July to 13 August 2022. We collected running records (n = 8,365,660) and extracted each trajectory’s start and end points with temporal attributes and downsampled the data into 9251 grids.

One-hour precision weather data were acquired from Weather Underground, a commercial weather information service. The weather observations are based in the Beijing Capital International Airport, which is located within the Sixth Ring Road. The data contain information on temperature, relative humidity, wind speed, precipitation, and special weather conditions, which allow us to estimate the overall thermal environment.

To acquire environmental and social indicators, we estimated the green coverage rate, plot ratio, and population density using multiple data sources. The plot ratio within the Sixth Ring Road of Beijing is estimated using the 10-meter precision building height data for China [42]. An approximate distribution of plot ratio is computed based on the average building height within each grid and divided by the average floor height (estimated at 4 m). The green coverage rate is estimated using land-use data from OpenStreetMap (OSM) by calculating the proportion of various green spaces within the grids. Similarly, water coverage is estimated by determining the proportion of different water areas within the grid cells. Finally, to estimate the city population distribution, we collected 1 day of mobile signaling data from Baidu Maps. This dataset was derived from location-based services (LBSs) that provide commuting origin and destination points at a 1 km-by-1 km resolution. Based on these data, we were able to estimate the population size of residents and workers within each grid cell (

Table 1. Data collection.

Name	Description	Unit	Source
Running Activity	Outdoor running trajectories	Polyline with temporal information	Keep
Weather	Temperature, humidity, precipitation	Values with hourly resolution	Weather Underground
Green Space	Boundaries of green space within the sixth ring of Beijing	Polygon	OpenStreetMap
Plot Ratio	Plot ratio approximated by building height	Values within grid cell (4 m)	Census Data
Population	Population of residents and workers	Values within grid cells (500 m)	Baidu

).

3.2. Indicator Quantification

The Running Activity Z-score (RAZ) indicator is based on the running trajectories that reflect urban spatial resilience to extreme heat. This indicator seeks to overcome the limitations of traditional methods in capturing human activity by offering a more sensitive and real-time approach to detecting thermal environmental impacts in urban spaces. Using this framework, areas with significantly high or low resilience can be identified through the local Moran’s Index, enabling further analysis of the correlations between urban factors and these spatial patterns.

The RAZ index is based on the assumption that heatwaves can significantly impact citizens’ willingness to participate in outdoor activities (Figure 1). We sample outdoor running activity data at daily intervals to obtain the running frequency per grid. The running frequency data can be standardized by calculating the average and standard deviation of grid frequencies. The RAZ index is defined in the following Equation (1):

$$RA{Z}_{i,t}={\displaystyle \frac{N_{i,t}-\overline{N_i}}{{\sigma }_i}},$$

(1)

where $N_{i,t}$ represents the running activity frequency on day t of the grid i. This standardization provides a measure of deviation and assists in quantifying the impact of heatwaves on the running activities within grids. To ensure index reliability, grid cells with the lowest 10% of average frequencies can be excluded, and data affected by social and weather factors (weekends and precipitation) can be removed. In addition, the heat index [43] and humidex [44] can be calculated for comparison.

To quantify heat exposure, we employed the heat indicator $E_T$ as proposed by Huang et al., which was developed based on Bosen’s Discomfort Index [7]. The indicator accounts for the combined effects of air temperature and relative humidity on human thermal comfort, with critical thresholds determined from historical meteorological data from 10 major Chinese cities. The definition of $E_T$ is presented in the following Equation (2):

$$E_T=\left\{\begin{array}{cc}1.8T_a-0.55(1.8T_a-26)\times (1-0.6)+32\hfill & , \mathrm{when}RH\le 60\%,\hfill \\ 1.8T_a-0.55(1.8T_a-26)\times (1-RH)+32\hfill & , \mathrm{when}RH>60\%,\hfill \end{array}\right.$$

(2)

where $T_a$ denotes the air temperature of the day and RH denotes relative humidity. We define a day as experiencing extreme heat if its $E_T$ value exceeds 87.3 °C, the critical threshold for Beijing.

To assess the sensitivity of the RAZ index to heatwave events, we computed the Pearson correlation coefficient between mean RAZ values and $E_T$. This analysis enabled the quantification of the relationship between RAZ and heat exposure. To examine the spatiotemporal characteristics of the negative correlation between the RAZ index and heatwave events, we conducted one-sided statistical tests on the daily RAZ index for each grid cell. The empirical distribution for each grid was derived from RAZ values observed during non-heatwave days across its 3 × 3 neighborhood (including the target grid itself). We determined that the RAZ value is significantly low if it falls below the 5th percentile of the corresponding empirical distribution. The dates identified as having significantly low RAZ values are presumed to be caused by extreme heat, excluding other influencing factors. To account for spatiotemporal continuity, we conducted this test on the mean RAZ values over 1, 2, and 3 consecutive days, as well as at spatial resolutions of 500 m by 500 m and 1500 m by 1500 m. Although previous studies used the quantile of $t_{N-1}$ as the test threshold, our approach differs because the running frequency is a count variable whose Z-score does not follow a t-distribution. Instead, we determined the one-sided test threshold ($\alpha =0.05$) referencing the quantile of the empirical distribution $F_{Emp}^k$ of the mean values from the time-consecutive groups of k observations. The definition of $F_{Emp}^k$ is

$$F_{Emp}^k\left(x\right)=\sum _{(i,t)\in A}{\displaystyle \frac{I({\sum }_{t^{\prime }=t-k+1}^{t}RA{Z}_{i,t^{\prime }}/k<x)}{\left|A\right|}}$$

(3)

where

$$A=\left\{(i,t)|RA{Z}_{i,t^{\prime }}\mathrm{is}\mathrm{not}\mathrm{removed},t^{\prime }=t-k+1,\cdots ,t\right\}$$

(4)

$F_{Emp}^k$ can be estimated using a bootstrap method. Then, the relationship between the BE and the index can be analyzed based on the validity of the RAZ index. Urban zones where outdoor physical activities are significantly inhibited or sustained during extreme heat events can be identified according to the index’s spatial distribution. Specifically, the indicator $RA{Z}_{diff}^i$ can be estimated by the following Equation (5):

$$RA{Z}_{diff}^i={\displaystyle \frac{{\sum }_{t}RA{Z}_{i,t}I(E_{T,t}\ge 87.3)}{{\sum }_{t}I(E_{T,t}\ge 87.3)}}-{\displaystyle \frac{{\sum }_{t}RA{Z}_{i,t}}{{\sum }_{t}1}}$$

(5)

where i and t denote the grid index and time, respectively, enabling an assessment of how heatwaves influence citizen behavior within each grid. Based on this, the spatial autocorrelation of the RAZ distribution can be analyzed and verified using a Moran’s I test. The spatial autocorrelation patterns and statistically significant ($p<0.05$) clusters can be analyzed using local Moran’s I statistics. The analysis revealed distinct hot spots and cold spots, representing areas of high heat resilience (HHR) and low heat resilience (LHR), respectively.

The spatial correlations between RAZ and conventional urban factors can be further analyzed, including calculating the Pearson correlation coefficient between RAZ and these factors. Additionally, the distribution of urban indicators—such as plot ratio, green coverage, working population, and residential population—can be compared across different types of identified significant areas using violin plots. Finally, a Kolmogorov–Smirnov (K–S) test can be conducted for each pair of these areas to assess differences in the distribution functions of the factors across different areas.

4. Results

4.1. Sensitivity of RAZ Indicator to Heatwaves

The temporal distribution of the RAZ indicator suggests local sensitivity to heatwaves. The result of the Pearson coefficient between the RAZ indicator and $E_T$ (statistic = −0.55, p-value = 0.04) indicates a strong correlation with heat stress. The results of humidex, heat index, and $E_T$ are consistent with the high-temperature yellow warning issued by the Beijing Meteorological Bureau, indicating that Beijing experienced extremely high temperature weather in early August. Thus, the heatwave period (from August 1st to 7th) is identified based on these results. Figure 2 presents the climate indicators for the study period.

The daily RAZ curves (Figure 3) illustrate a significant successive decrease during the extreme heat period (1–5 August), with an average decrease of 0.26 and a minimum of −1.02 in the overall mean curve. This result aligns with previous research on citizen behavior during hot days in Australia [45]. Meanwhile, the results show that rainfall and holidays have a substantial impact on the RAZ values, demonstrating the need to exclude the effects of unexpected factors.

The decrease in RAZ values during heat waves is related to urban factors, including the green coverage rate, the plot ratio, and the density of the population, as observed in the grouped temporal distribution of RAZ. By comparing the mean RAZ curves for areas with the highest 25% green coverage rates, plot ratios, and population density, we observed that high green coverage can alleviate the decrease in RAZ during extremely high temperature periods. Conversely, high plot ratio and population density intensify the reduction in RAZ values, with population density having a more pronounced impact. These findings are consistent with previous studies [23,31]. A possible explanation is that high plot ratios and population density exacerbate the urban heat island effect [46].

Notably, we observed an apparent mismatch between the highest $E_T$ (5 August) and RAZ values, which can be attributed to temporal dynamics. While 5 August recorded the highest daily $E_T$, running activities predominantly occur during evening and night hours. On this day, a cooling and dehumidification process that began in the afternoon created more favorable temperature conditions during peak running hours. Consequently, despite the high daily $E_T$, the improved evening conditions prevented RAZ from reaching its minimum value.

Figure 4 illustrates the results of significant low RAZ value detection across varying spatiotemporal resolutions. For the 1-day detection, the proportion of significantly low RAZ values during heatwaves and rainy days is found to be significantly higher than the hypothesized 0.05 threshold at both the 500 m and 1500 m geographical resolutions. This indicates that the RAZ index is sensitive to and effective at capturing the impacts of these key meteorological factors.

Regarding the detection at the three different temporal resolutions, the detection rate of extremely low RAZ values during heatwave periods increases as the time span is expanded. This is because heatwaves, as a continuous extreme weather event, lead to a consecutive decline in the number of runners, and the average RAZ calculated over multiple days is better able to capture this characteristic temporal pattern. Moreover, in the detection analysis at the two spatial scales, the detection rate during heatwaves at the 1500 m resolution is significantly higher than that at the 500 m resolution. This is likely because the impacts of heatwaves exhibit spatial continuity, as evidenced by Moran’s I test results presented in the following section. Appropriately increasing the geographical scale of the assessment can therefore better capture this spatial feature of heatwave effects.

These findings demonstrate the sensitivity of the RAZ index to both temporal and spatial dimensions of extreme meteorological events, providing confidence in its performance as an indicator of heat stress impacts. By monitoring the decrease in the RAZ index, researchers can dynamically observe areas that are potentially affected by extreme heat.

4.2. Low and High Heat Resilience Identification

The distribution of $RA{Z}_{diff}$ (Figure 5a) reveals significant spatial autocorrelation and potential correlations with urban factors. Statistical validation through Moran’s I test (p < 0.001, Z-score = 74.809) confirms the significant spatial autocorrelation of $RA{Z}_{diff}$. The distribution also shows that areas with lower $RA{Z}_{diff}$ values (indicating lower SHR) tend to concentrate in densely populated regions, while a notably higher $RA{Z}_{diff}$ value circular region appears in the green spaces around the Fifth Ring Road. An intriguing observation is the absence of substantially higher $RA{Z}_{diff}$ values in the green spaces within Beijing’s Fourth Ring Road. This phenomenon may be attributed to the pronounced urban heat island effect in the city center, which potentially obscures the thermal mitigation capacity of urban green spaces through direct $RA{Z}_{diff}$ value assessment. This finding underscores the complexity of urban thermal dynamics and the challenges in quantifying green infrastructure’s thermal regulatory potential within dense urban environments.

Figure 5b illustrates the identified areas of significantly high and low SHR within the city. The results indicate that identified LHR areas are concentrated in the high population and plot ratio area (i.e., central and northern regions). Conversely, the significant HHR areas are predominantly located along the Fifth Ring Road, particularly in wetland parks. These findings align with the distribution of the urban heat island effect in Beijing [46] and provide evidence from a citizen behavior perspective that wetlands and parks mitigate extremely high temperature weather conditions within the city. However, it is worth noting that while parks have a positive role in mitigating the urban heat island effect, they may not maintain their spatial vitality during heatwaves (i.e., Old Summer Palace Park). Despite abundant green coverage and water bodies, some wetland parks fall into the significantly low SHR area. One possible reason is their distance from residential areas, resulting in weaker accessibility. Moreover, for fee-based and enclosed parks, higher travel costs may discourage people from participating in outdoor activities during heatwaves.

4.3. Correlation Between RAZ and Urban Factors

$RA{Z}_{diff}$ exhibits a certain degree of correlation with the spatial distribution of various indicators, but lacks a significant linear relationship, with the green coverage being the most pronounced. The correlation results indicate that population density (Pearson coefficient = −0.182, p-value < 0.001) and plot ratio (Pearson coefficient = −0.161, p-value < 0.001) show a negative correlation with $RA{Z}_{diff}$, but no significant linear relationship. While there is no clear linear relationship between population and plot ratio and $RA{Z}_{diff}$, they do exhibit a negative correlation trend, which aligns with existing research findings.

The Pearson correlation coefficient between green coverage and $RA{Z}_{diff}$ was 0.035 (p-value = 0.002), indicating a minor positive correlation. This suggests that the influence of green spaces on outdoor activities may be mediated by other factors, rather than a straightforward promoting relationship. Since green spaces can influence the urban microclimate by reducing local temperatures, we characterized this feature using the average greening rate within a 1500 m radius. The smoothed greening rate exhibited a Pearson correlation coefficient of 0.083 (p-value < 0.001) with RAZ, indicating a more significant positive correlation. Furthermore, considering that the popularity of running at the location and the urban heat island effect might impact the relationship between green coverage and RAZ, we used the number of runners on weekdays and the distance from the city center to depict these two factors. The partial Pearson correlation coefficient was 0.095 (p-value < 0.001), implying a more significant positive correlation between green coverage and $RA{Z}_{diff}$. This result reflects the complex and intricate relationship between citizens’ activity and urban factors.

The blue space coverage and RAZ also did not exhibit a significant correlation (Pearson coefficient = −0.020, p-value = 0.062). Similar to the analysis of green coverage, we averaged the distribution within a 2500 m radius to quantify the impact of water bodies on the microclimate. However, the correlation coefficient remained insignificant (Pearson coefficient = 0.009, p-value = 0.436). The possible reason for this is that the influence of water bodies on the microclimate is related to factors such as the morphology and depth of the water bodies [47]. Within the Sixth Ring Road of Beijing, there are various types of water bodies, including rivers, lakes, and wetlands, and the surrounding socioeconomic factors are also diverse. In future research, a more detailed description of the characteristics of blue spaces is required.

The results of area sensing indicate a significant correlation between SHR and urban factors. Compared with the overall data distribution, the extracted high SHR area (HHR) and low SHR area (LHR) exhibit pronounced differences in urban indicators including green coverage. The results of the K–S test indicated significant differences ($p<0.001$) in the distribution of all indicators between the overall research area, the LHR area, and the HHR area, except for blue space coverage. Specifically, the distribution of blue space coverage within the overall research area and the LHR area was not significantly different ($p=0.119$). Figure 6 displays distributions of plot ratio, green coverage, resident population, and working population within LHR, HHR, and all research areas. The results indicate that the distribution of plot ratio and population in the three identified areas is consistent with and more significant than the results of the Pearson coefficient. Furthermore, green coverage also exhibits significant higher values within areas of higher SHR, differing from previous results.

The difference may arise from two key factors. First, the impact of urban factors on SHR may be synergistic and nonlinear. For example, in areas outside the Fifth Ring Road, lower green coverage does not necessarily lead to reduced running activity during heatwaves, as these regions also have smaller populations and fewer runners. In contrast, the downtown’s lower greenery combined with higher density results in a more pronounced decrease in running frequency. Second, the influence of urban factors on SHR may exhibit spatial continuity, where a grid’s SHR is related to the characteristics of surrounding areas. For example, the plot ratio in neighboring grids can impact the local microclimate, affecting residents’ willingness to jog during heatwaves. This spatial dependency means that a grid’s SHR cannot be fully explained by its attributes alone.

The analysis revealed that the SHR metric derived from $RA{Z}_{diff}$ aligns with the findings of prior studies—high green coverage, low population density, and low plot ratios are positively associated with increased SHR. However, our approach differs from those of previous works, which primarily focused on linear relationships between SHR and individual urban factors. By accounting for the spatial continuity of urban influences, we were able to identify synergistic effects among the various factors. This suggests that $RA{Z}_{diff}$ can capture the complex, interconnected relationships between SHR and the built environment, rather than simplistic, isolated correlations.

5. Discussion

5.1. Urban Planning Implementations

The results indicate that the RAZ indicator effectively captures the fluctuations in citizen running behavior within grids from the perspective of outdoor activities, demonstrating distinct advantages in monitoring the potential impact of heatwaves. Compared with traditional detection indicators, RAZ offers three main advantages for urban planning implementations. First, the indicator’s high sensitivity to thermal conditions enables more geographically precise and timely monitoring of heatwaves, as shown by extremely low value detection tests for the RAZ index. This allows planners to track heat impacts in near real time, leading to a more responsive urban management.

Second, the RAZ index also offers a data-driven approach for policymaking based on citizen behavior. By examining areas where RAZ significantly decreases during heatwaves, researchers can directly and specifically understand which parts of the urban environment are more susceptible to high temperatures, instead of indirect inference through built environment indicators. Furthermore, by employing appropriate causal inference methods, such as the double machine learning (DML) approach [48] and geographically weighted generalized propensity score methods [49,50], researchers can unbiasedly analyze the causal effect of a particular indicator on the spatial heat resilience represented by outdoor physical activity, thereby allowing for the estimation of policy outcomes.

Third, RAZ naturally reflects spatial significance through citizen running patterns, indicating both space utilization and heat vulnerability. By focusing on areas with substantial data and distinct local patterns, planners can identify and prioritize heat-stressed urban zones that require detailed investigation. This spatial intelligence enables more targeted and effective urban heat resilience strategies.

5.2. Research Contribution

This study introduces a novel method for real-time monitoring of the urban thermal environment by leveraging human behavioral data, particularly capitalizing on the increasing use of fitness applications. The proposed RAZ index offers urban planners a powerful tool to quantify and categorize spatial heat resilience in cities, while providing researchers with new insights into the complex interactions between citizen behavior and the urban environment. By utilizing data from citizen activities, this approach allows planners to better understand how urban spaces are perceived and utilized during heatwaves, offering valuable empirical evidence to inform human-centric planning and design decisions.

Our analysis confirms the correlation between $RA{Z}_{diff}$ and common planning indicators such as green coverage, plot ratio, and population density, demonstrating significant variation in urban factors across areas of high and low spatial heat resilience (SHR). While these findings align with existing studies, our research uncovered that the relationship between SHR and urban factors is more nuanced than previously understood. We found neither a direct linear correlation between individual urban indicators (e.g., green coverage) and RAZ nor any clear univariate trend. This highlights the complexity of the relationship between SHR and urban factors, suggesting that urban heat resilience is influenced by a more intricate interplay of variables rather than simple, singular associations. Therefore, this study advances current understanding by emphasizing the need for multidimensional analyses in assessing urban resilience to extreme heat events.

5.3. Limitations and Future Work

The RAZ indicator’s reliability as a direct measure of heatwave impact is limited by its sensitivity to both social factors (such as holidays) and weather conditions, particularly precipitation. This sensitivity is clearly visible in the daily RAZ curves, where precipitation-induced decreases in RAZ are comparable in magnitude to those caused by heatwaves. This makes it challenging to isolate heatwave effects from other influences when using the RAZ index. In our study, given the limited data availability, we addressed this challenge by directly excluding both rainfall and holiday-related data from our analysis. Future research could employ a more detailed approach to account for holiday effects. This may involve modeling holiday-related running patterns separately or quantifying their specific impact on running frequency. Additionally, our relatively short study period did not capture potential long-term trends in citizens’ willingness to engage in outdoor running. To examine the RAZ indicator’s properties over an extended time frame, incorporating a temporally weighted algorithm may be necessary.

Another limitation of the RAZ indicator stems from the demographic bias in fitness application users. Keep’s user base predominantly consists of young, highly educated individuals—74% are under the age of 30, and 76% hold at least a bachelor’s degree [37]. This demographic bias results in the RAZ index primarily reflecting the behavioral changes of young, highly educated users. While this demographic tends to be the most physically active in urban areas, their behavior patterns may not represent the broader population, particularly older adults, who are often more vulnerable to heatwaves. Given these demographic limitations, the RAZ index alone may not provide a complete picture of extremely-high-temperature risks in urban areas. Thus, it should be supplemented with additional indicators to ensure comprehensive heatwave impact monitoring.

Moreover, this study’s scope is limited to analyzing running activity frequency, without considering other forms of outdoor physical activities or potential behavioral adaptations, such as modifications in running distance. Consequently, the RAZ index only evaluates the impact of heatwaves on outdoor activities from a specific perspective, and it is hazardous to apply this indicator solely for SHR assessment. While it is impractical to collect all relevant information in one study, we believe that future work would benefit from a more comprehensive methodology that quantifies the full range of changes in outdoor running behavior and integrates indicators from various fields, including epidemiology and sociology. Our research concentrates exclusively on a single city and a relatively short period, including only one heatwave event. Moreover, meteorological data are collected from only one site, and no precise control is exerted over the measurement accuracy. The limited research data necessitate caution in drawing broad conclusions about the general applicability of the RAZ index. To establish the index’s validity and reliability, further comprehensive studies are essential. Future research should consider a wider range of geographical locations and extended time periods to enhance the reliability of the findings.

6. Conclusions

This study utilizes trajectory data from sports apps to propose a city heat resilience index, RAZ, based on the perspective of outdoor physical activities. The index reflects the impact of high temperature on the urban thermal environment by quantifying the degree of change in running frequency. The case study conducted within Beijing’s Sixth Ring Road from 17 July to 13 August 2022 reveals the following findings: (1) In the temporal dimension, abnormally low RAZ values are highly correlated with extremely high temperature events, establishing a reliable basis for monitoring urban high temperatures. (2) In the spatial dimension, the RAZ indicator aligns with existing research on the relationship between the urban built environment and SHR. Furthermore, by extracting spatial low-value areas, we can identify potential low heat resilience areas in the city. This analysis reveals a more nuanced relationship between SHR and the urban environment, offering targeted planning recommendations to enhance urban heat resilience.

The RAZ indicator proposed in this study provides an effective metric for urban thermal environment monitoring and spatial heat resilience analysis, offering a behavioral perspective supplement to the construction of heat resilience indicators. Meanwhile, the RAZ indicator can be also susceptible to various external factors. Future research should consider employing more detailed temporal models and causal analysis methods to enhance the reliability of RAZ results.