Micro-Urban Heatmapping: A Multi-Modal and Multi-Temporal Data Collection Framework

Abstract

Monitoring microclimate variables within cities with high resolution and accuracy is crucial for enhancing urban resilience to climate change. Assessing intra-urban characteristics is essential for ensuring satisfactory living standards. This paper presents a comprehensive methodology for studying urban heat islands (UHIs) on a university campus, emphasizing the importance of multi-modal and multi-temporal data collection. The methodology integrates mobile surveys, stationary sensor networks, and drone-based thermal imaging, providing a detailed analysis of temperature variations within urban microenvironments. The preliminary findings confirm the presence of a UHI on the campus and identify several hotspots. This comprehensive approach enhances the accuracy and reliability of UHI assessments, offering a cost-effective, fine-resolution approach that facilitates more effective urban planning and heat mitigation strategies.

1. Background

Urban heat island (UHI) intensity typically varies between 0.4 °C and 11 °C, with a more pronounced effect at night, exposing residents to higher thermal stress during the summer [1]. This heat exposure can adversely impact human health, increasing mortality and morbidity, particularly among vulnerable populations such as the elderly [2,3]. The impacts are exacerbated during heatwaves, leading to significant health risks. From 2000 to 2019, North America reported an average of six heat-related deaths per 100,000 residents annually [4]. In the United States, rising extreme temperatures are anticipated to result in more heat-related deaths and illnesses [5]. Heatwaves, which interact non-linearly with UHIs, amplify urban heat stress, a trend expected to continue with increasing frequency and intensity [6,7,8,9]. Global warming further intensifies UHIs, endangering urban livability, particularly during heatwaves [10,11]. Urban development and policy circles prioritize strategies to control microclimatic variables influencing UHIs, given the dynamic factors and diverse microclimates created by complex urban topography. The variability of microclimates within the same city necessitates detailed investigation into the spatial–temporal variations arising from local architecture, human activity, urban planning, and fluctuating weather conditions [12,13,14].

The urgency of mitigating the urban heat island (UHI) phenomenon is paramount, as it poses a significant threat to urban livability amidst global climate shifts. Academic, urban development and policy circles are prioritizing strategies to control the microclimatic variables influencing UHIs [15,16]. However, the task is non-trivial, given the dynamic factors at play and the diverse microclimates created by complex urban topography.

Historically, localized meteorological stations have been pivotal for measuring parameters such as air temperature and humidity. Notable research in cities within Greece [17], China Field [18], and France Field [19] have employed networks of weather stations to evaluate UHI intensity. For example, a study in Dijon during a heatwave utilized an extensive network of fixed sensors to analyze the city-wide microclimatic conditions [19,20]. The outcomes of these studies are invaluable; they measure the impact of local climate phenomena and illuminate the mechanics of urban climate, thereby aiding the development of effective mitigation strategies. Nonetheless, these weather stations have inherent limitations due to their sparse distribution, thereby offering snapshots rather than continuous spatial representations. This limitation underscores the need for more comprehensive data collection methods to capture the full scope and patterns of urban microclimates [21].

In response to these challenges, this project aims to utilize the Notre Dame Campus as a living laboratory. This study aims to map the heat island effect by proposing and validating multi-modal and multi-temporal data collection frameworks. These frameworks are designed to identify and monitor the UHI effect at a granular level with an integrated variety of sensor technologies. This initiative is a collaborative effort involving the School of Architecture, College of Engineering, Office of Sustainability, and Office of Facility Management.

2. Types of UHIs and Related Technologies

Urban heat islands (UHIs) manifest in three primary forms, differentiated by the atmospheric layer they impact: surface urban heat islands (SUHIs), canopy layer urban heat islands (CLUHIs), and boundary layer urban heat islands (BLUHIs) [22,23,24]. SUHIs are characterized by elevated temperatures of constructed surfaces such as roads and buildings, detectable via thermal imaging from remote sensing. CLUHIs affect the air layer just above ground level, influenced by both the built environment and human activity, typically measured through ground-based meteorological stations. BLUHIs encompass both the canopy layer and the atmosphere above, requiring advanced techniques like satellite and aircraft observations for study. Each UHI type presents unique thermal characteristics and demands specific measurement strategies to fully understand its impact.

Traditional technologies in CLUHI research, such as premanufactured sensors and data loggers, have been fundamental for data collection. These devices provide reliable and consistent measurements, essential for establishing baseline UHI metrics and understanding temporal patterns. However, these methods have notable limitations, such as failing to capture transient or localized climatic events due to fixed-position sensors or loggers. Moving sensors frequently can lead to issues due to delays in data registration. For instance, many data loggers use metal temperature sensors that require several seconds to adjust to environmental changes, leading to significant inaccuracies when used on fast-moving vehicles like cars or bikes [25].

Localized meteorological stations have been essential for measuring parameters such as air temperature and humidity. Research studies in cities like those in Greece [17,18], and France [19] have used weather station networks to evaluate UHI intensity. For example, a study in Dijon during a heatwave used an extensive network of fixed sensors to analyze city-wide microclimatic conditions.

Studies in sustainable design and urban planning have recently focused on enhancing local thermal comfort and thermal regulation in urban environments. For instance, research conducted in Shanghai investigated the local thermal comfort of students on a university campus, highlighting the importance of considering microclimatic conditions and human perceptions in urban planning to improve thermal comfort for occupants [26]. Another study in Qinhuangdao, China, utilized the ENVI-Met model to analyze the thermal regulation of coastal urban forests, emphasizing the role of green spaces in mitigating UHI effects and improving microclimatic conditions [27].

These studies underscore the significance of integrating nature-based solutions and sustainable design principles in urban planning to create resilient and sustainable urban environments. However, these weather stations offer only snapshots rather than continuous spatial representations, highlighting the need for more comprehensive data collection methods to capture the full scope of urban microclimates.

Urban heat island (UHI) studies have been a focal point in understanding the impact of urbanization on local climates. Various research works have utilized data collection methods involving different modes of transportation and remote sensing technologies to analyze and map UHI effects. Imhoff et al. (2010) conducted remote sensing observations across biomes in the continental USA, revealing that the UHI amplitude increases with city size and exhibits seasonal asymmetry [28]. Kaplan et al. (2018) focused on Skopje, Macedonia, using Landsat 8 satellite data to analyze UHIs, emphasizing the temperature alterations in urban areas compared to rural surroundings due to urbanization [29]. Anurogo et al. (2022) employed Landsat 8 imagery to estimate land surface temperature and assess UHI in Batam Municipality, highlighting the capabilities of remote sensing data in mapping UHI [30].

Lee et al. (2019) analyzed the trend of UHI intensity in Asian megacities concerning urban area changes, highlighting the global occurrence of UHI phenomen [31]. Peng et al. (2011) studied surface urban heat islands across 419 global big cities using satellite-derived data, focusing on environmental monitoring and urban heat island effects [32]. Keeratikasikorn and Bonafoni (2018) utilized Landsat 8 imagery to analyze UHIs over the land use zone plan of Bangkok, emphasizing the importance of surface urban heat island maps in sustainable urban planning [33].

Furthermore, Bohnenstengel et al. (2013) investigated the impact of anthropogenic heat emissions on London’s temperatures, emphasizing how human activities contribute to UHI effects [34]. Mohamed (2024) conducted a comparative study on urbanization and heat island effects in Egypt, stressing the significance of urban planning strategies to mitigate UHI effects and calling for further research in this area [35]. Zhou et al. (2015) focused on the footprint of UHI effects in China, highlighting UHIs as a major anthropogenic modification to the Earth’s climate system [36].

While existing studies have made significant contributions to understanding UHI effects using various data collection methods and technologies, there remains a research gap in integrating data collection by cars, bikes, remote-controlled cars, pedestrians, and thermal cameras for comprehensive UHI analysis. Future research could focus on combining data from multiple modes of transportation and remote sensing technologies to provide a holistic view of UHI patterns and dynamics in urban areas. Additionally, exploring the synergies between different data collection methods could enhance the accuracy and spatial resolution of UHI mapping, leading to more effective urban planning and mitigation strategies to address the challenges posed by urban heat islands.

In response to these challenges, this project aims to use the Notre Dame Campus as a living laboratory. The study proposes and validates multi-modal and multi-temporal data collection frameworks to map the heat island effect. These frameworks are designed to monitor the UHI effect at a granular level using an integrated variety of sensor technologies. This initiative is a collaborative effort involving the School of Architecture, College of Engineering, Office of Sustainability, and Office of Facility Management. This study focuses on an integrated approach to studying SUHIs and CLUHIs, with the goal of delineating the benefits and limitations inherent in this holistic method of data collection and analysis. Acknowledging these limitations is critical not only for the prudent interpretation of research findings but also for directing the evolution of methodologies in UHI research.

3. Proposed Framework

Figure 1 outlines the proposed framework for data collection and analysis to study UHIs. For CLUHIs, the methodology incorporates two modes of traverse data collection: pedestrian surveys with sensors, which involve taking measurements at predetermined locations, and the deployment of fixed sensors on rooftops to record temperature data. For SUHIs, a remote sensing strategy is adopted, utilizing overhead 2D thermal imagery captured by drones to analyze the rooftops. There are two distinct phases for data collection: one for traverse data and another for fixed sensor and thermal image data, with the specific collection dates delineated in Figure 1. Both methods are designed to identify and scrutinize ‘hot spots’ within the urban area—specifically, the campus—where temperatures are significantly higher compared to the surrounding environment. This comprehensive framework is intended to enhance our understanding of temperature disparities in urban settings and to aid in devising strategies to mitigate the impacts of UHIs.

4. CLUHI Data Collection

4.1. Traverse Data: First Round

4.1.1. Data Collection

On 20 August 2023, the campus of the University of Notre Dame in Notre Dame, Indiana, United States, was selected for the first round of comprehensive data collection. The campus comprises asphalted pavements and vegetated areas and it is covered by mature trees, as illustrated in Figure 2. The latitude is 41.676 °N and in climate zone 5. Climate Zone 5 features a humid continental climate with cold winters and warm to hot summers, common in parts of the northern United States, southern Canada, and similar regions in Europe and Asia. It experiences moderate to high precipitation, with a mix of rain and snow. Urban heat island (UHI) effects are increasingly concerning in this zone due to urbanization, which leads to the expansion of heat-retaining surfaces like roads and buildings, resulting in higher urban temperatures compared to rural areas. This exacerbates health risks, increases the energy demand for cooling, and stresses the infrastructure. In winter, retained heat reduces the snow cover, affecting ecosystems and water cycles. Elevated urban temperatures in this climate cause more heat-related illnesses and higher mortality rates, particularly during heatwaves, impacting vulnerable populations such as the elderly and children. Increased cooling demands lead to higher greenhouse gas emissions, worsening the climate change. This specific date was chosen based on favorable meteorological conditions conducive to research, characterized by clear skies and abundant sunshine. Data pertaining to temperature were recorded at 1 min intervals throughout the day, while GPS trackers were used to record the location at 15 s intervals. The data collectors were asked to walk through the entire campus and asked to walk through all major academic buildings on the map (refer to Figure 2a). The total campus area is around 1250 acres, comprising around 170 buildings. The covered campus for data collection is around 480 acres and covers all academic buildings.

The timeslots are categorized into morning, afternoon, and evening sessions. The minimum temperature recorded was 19 °C (66.2 °F) at 5:05 a.m., while the maximum reached 32 °C (89.6 °F) at 5:10 p.m. The mean temperature and relative humidity were documented as 77 °F and 74.85%, respectively. A total of six undergraduate research assistants were involved in the data collection, forming three pairs. They collected data on foot by walking through the campus. Notably, one pair was solely dedicated to the afternoon session, with the remaining pairs active during all designated periods. The morning session, spanning from 7:00 a.m. to 10:30 a.m., was conducted by two pairs of research assistants. Each pair was equipped with a Tracki GPS device for accurate geolocation tracking (refer to Figure 2b). Moreover, they each carried an Elitech RCW-360 Plus WIFI Digital Data Logger (illustrated in Figure 2b), which included dual temperature sensor probes. The employment of two sensors was instrumental in augmenting the precision of the temperature readings by enabling the calculation of average temperatures. These devices demonstrated a temperature accuracy of ±0.5 °C (±0.9 °F). To ensure a thorough survey, participants were provided with digital and hardcopy maps, which delineated the specific zones for temperature data collection.

The afternoon session, which lasted from 12:30 p.m. to 5:30 p.m., was critical for examining urban heat islands. During this interval, the team expanded to include two additional pairs. A novel data collection approach was adopted; two pairs were tasked with recording human body perceived temperature data through sensors (the real-feel temperature), while the other two pairs focused on environmental temperature readings. The temperature sensors functioned by measuring the temperature of two metals within, analogous to the physiological process of human thermal perception. These metals require a temporal duration to equilibrate with the ambient temperature during both cooling and heating phases. Accordingly, two pairs remained stationary in predetermined locations for 3–4 min to obtain precise environmental temperatures, while the other pairs moved through these zones to capture real-feel temperature data. The objective was to compare these datasets to derive insights, such as the potential impact—or lack thereof—of heat islands on transient user experiences across the campus. The same devices from the morning session were utilized. The requirement of remaining stationary in the location added additional time, which led to the long timespan of data collection in the afternoon time slot.

The evening session commenced at 5:45 p.m. and concluded at 8:00 p.m., involving the same pairs as the morning session. Upon completion, all devices were collected, and the temperature and GPS data were integrated seamlessly. It is noteworthy that the intervals between the sessions were strategically used for device maintenance and recharging, which is paramount for maintaining the fidelity of the data collected.

4.1.2. Data Processing and Visualization

In this research, ArcGIS, developed by Esri, is used for data visualization due to its exceptional capabilities and user-centric design. The selection of ArcGIS for spatial data analysis was informed by three key factors. Firstly, its versatile data compatibility is notable; ArcGIS is adept at managing a variety of data types, including tables, LiDAR imagery, and contour feature objects. It excels at integrating and displaying multiple datasets, each with unique locational information, on a single map, which is essential for a thorough spatial analysis. Secondly, the software offers robust functionalities that enhance analysis. These include grid sampling for detailed spatial examination, table joins and relates for merging different datasets, and an advanced rasterization tool that is especially useful for generating smooth heat maps. Lastly, ArcGIS provides collaborative and interactive mapping capabilities that are indispensable for teamwork within research communities and for broader public engagement. Users can create and share both static and dynamic web maps, and benefit from the extensive shared resources available within the ArcGIS user community, such as detailed campus maps and varied topographical representations, which streamline the process of constructing comprehensive maps. Hence, despite the expense of ArcGIS software licensing (ESRI 2016), it is the best option for this study since it aligns with our goals of achieving high-quality, comprehensive, and collaborative spatial data analysis and visualization.

Before importing the collected data into ArcGIS for visualization, it is essential for them to first undergo a preprocessing stage. To illustrate, let us consider an example from our first round of pedestrian data collection. The data were sourced from two different types of devices: a temperature sensor and a GPS tracker. Our goal in preprocessing is to accurately merge these datasets, linking each temperature reading with its corresponding location. The preprocessing begins by filtering out undesirable entries in the data from temperature sensors, particularly those displaying abnormal values, referring to the difference between the temperature of two sensor probes recorded at the same timestamp. Typically, a larger temperature difference indicates lower accuracy and confidence in the readings.

A threshold is established to retain entries falling within a specific range of temperature difference, and this threshold can be adjusted as needed to retain more entries. For this study, the threshold is set at 1 degree Celsius. During the data processing from the initial walk, 1484 rows of entries are refined into 931 rows of more accurate and reliable data points. Following this, the timestamps between the temperature readings and the GPS tracker data are aligned, ensuring that each temperature sample is accurately associated with a precise geographical location.

Once the data are properly processed and aligned, they become ready for visualization in ArcGIS. The data are imported using the ‘Add XY Point Data’ feature, transforming the dataset into mappable points. The next step involves customizing the map’s symbology, where varying colors are assigned to points based on their temperature values. Moreover, ArcGIS’s filter function allows the display of data within specific time frames. For instance, to analyze temperature data collected from 7:00 a.m. to 10:30 a.m., the corresponding filter is applied, resulting in a detailed and informative map. This map can then be integrated into a layout in ArcGIS and exported as a static map image, as shown in Figure 3a.

To compute the temperature variation for each data point within the first group, the closely located data points are identified from the second group, considering a specified range of 0.0005 degrees for both longitude and latitude. Among these candidates in the second group, the one that demonstrates the most significant temperature difference is highlighted when compared to the data point in the first group (referred to as point A), designating it as point B. The temperature variation is defined as the temperature difference between point A and its corresponding point B. The point-matching process is depicted in Figure 3d where circles represent data points recorded in the morning, and dots signify those recorded in the afternoon. A circle encompassing dots indicates a successful point match, which is then utilized to calculate the temperature variation, as presented in Figure 3d.

4.1.3. Lessons Learned

To begin with, the local weather station (close to the airport) recorded the highest temperature as 31.7 °C (89 °F) and the lowest as 20 °C (68 °F). The lowest temperature is slightly higher than that recorded on campus, 19 °C (66.2 °F), and the highest temperature is slightly lower than that on campus, 32 °C (89.6 °F), which can be interpreted as that the campus has more considerable temporal temperature variation than the location around the airport, that is around 8.37 km away from the campus.

Two critical lessons are learned from the first round of data collection. The first lesson pertains to the duration over which data were gathered; it is apparent that the diurnal temperature variations are predominantly attributable to the temporal extent of the data acquisition process, particularly within the afternoon intervals. This observation complicates the task of distinguishing the inherent daily temporal temperature oscillations from the urban heat island effect during the summer months. In response, the data collection period is confined to one hour in the second round of data collection, thereby aiming to attenuate the confounding effects of natural diurnal temperature changes.

The second lesson revolves around the sensitivity and veracity of the data procured. The utilized instrument, an Elitech RCW-360 Digital Data Logger (refer to Figure 2b), is equipped with dual probes for temperature recording. Notably, concurrent temperature measurements at identical locations exhibited discrepancies up to 6.7 °C between probes, casting doubts on the precision and reliability of the readings. Upon instituting a criterion that accepts data within a predetermined temperature differential of 1 °C, a mere 63% of the recordings (931 out of 1484 datasets) were deemed suitable for analytical purposes. Consequently, this threshold for data inclusion necessitated the initiation of a subsequent data collection phase to enhance the robustness of the dataset; also, another device with higher accuracy was chosen to improve the data accuracy and reliability.

4.2. Traverse Data: Second Round

4.2.1. Data Collection

On the 29th of October, 2023, the second series of data collection was executed. The selection of this particular date was predicated on its diurnal temperature constancy, thus providing an opportunity for the investigators to authenticate the urban heat island effect. To elucidate, detecting any data variances on a day characterized by stable temperatures would incontrovertibly indicate the influence of UHI factors other than diurnal thermal fluctuations. The lowest temperature recorded on this date was 6 °C (42.8 °F), while the highest was 9 °C (48.2 °F), aligning with the local meteorological station’s reports. The mean temperature was determined to be 7.09 °C (44.77 °F). Humidity levels were elevated throughout the day, averaging 92.48%, attributed to intermittent rainfall, as illustrated in Figure 4b).

Informed by the insights from the initial data collection effort, this session involved the concerted efforts of five individuals, each tasked with surveying a specific sector of the campus. A total of 14 routes were predefined on Google Maps, with the campus territory distributed amongst the data collectors (see Figure 4a). The objective was to procure data for each time segment within a duration not exceeding one hour to preclude the impact of daily temperature oscillations. Furthermore, the personnel were directed to avoid indoor environments to eliminate potential data inaccuracies. The imperative of adhering to the designated completion time for each route was underscored. All paths were rigorously verified on Google Maps to ascertain the feasibility of timely completion. Moreover, team members were mandated to document their commencement and conclusion times for each time slot precisely.

Data were accumulated across three designated temporal segments: morning, afternoon, and evening. Commencing at 9 a.m. and extending until 10 a.m., two groups handled the morning session, each provisioned with a Tracki GPS device to facilitate pinpoint locational accuracy (as depicted in Figure 4c). Concurrently, an Elitech GSP-6 PTE Ultra Low Digital Data Logger (illustrated in Figure 4c), equipped with a single temperature sensor probe and a humidity probe, was carried by each team. This arrangement permitted the simultaneous acquisition of both temperature and humidity metrics. The aforementioned devices are characterized by a temperature precision of ±0.2 °F/±0.1 °C and a humidity precision of ±3% RH. To ensure exhaustive data collection, each group was furnished with both digital and physical cartographic materials, with areas designated for temperature data collection clearly annotated. The midday slot, spanning from 12 p.m. to 1 p.m., was particularly pivotal, with a focus on the documentation of heat islands. The final collection interval occurred between 5 p.m. and 6 p.m., engaging the identical cohorts as the morning session. After this session, all instrumentation was aggregated, facilitating the integration of temperature and geospatial data. It is worth noting that the intervals between these time slots were strategically utilized for device maintenance and recharging, ensuring the accuracy of all collected data.

4.2.2. Data Processing and Analysis

Upon completion of data acquisition, each dataset was cataloged with an identifier corresponding to the respective device. The collection transpired over the course of the entire day, necessitating the exclusion of any data extraneous to the pre-defined collection intervals. A GPS tracker was synchronized with each temperature and humidity data logger, recording positional data at 15 s intervals. Thus, the gathered coordinates were aligned with the relevant temperature/humidity device data based on their temporal stamps. Subsequently, all datasets were amalgamated and scrutinized for potential anomalies.

Post data cleansing, Python was employed for data visualization. The analysis of heat islands was predicated on the temperature variance among different locales. The underlying premise was that any substantial temperature variation within a particular area, relative to others, likely emanated from UHI factors of that specific zone since the temperature remained fairly constant throughout the day. Visualization was achieved using Python libraries, including Folium, GeoPandas, and Pandas. The campus terrain was subdivided into a grid system to transmute and depict the geospatial data as a matrix over the map. Each grid segment documented the minimal and maximal temperatures, the instances of their documentation, and the extent of the temperature disparity. A chromatic scale from green, symbolizing the minimal temperature deviation, to red, indicating the maximal, was implemented to present these data graphically. As can be seen in Figure 5a, certain areas on the campus have higher temperature differences compared to other ones. The green areas’ temperature change throughout the day is about 1.5 degrees Celsius lower than that of the red areas. The same method was used to process data collected in the first round, as shown in Figure 5b; there are differences in the identified ‘hot spot’ grid. However, the overall trend is similar; that is, the west side of the campus has a higher temperature variation throughout the day compared to the east side. In addition, there is consistency in the cluster with the highest temperature oscillation, that is, grid 31 through 35.

4.2.3. Lessons Learned

The local weather station recorded the highest temperature as 8.9 °C (48 °F) and the lowest as 5.6 °C (42 °F). The lowest and highest temperatures are both slightly lower than those recorded on campus, 9.3 °C (48.7 °F) and 6 °C (42.8 °F), which is different from the summertime. The strict one-hour data collection window was a very effective method change compared to the first round; more than 1555 datasets were collected, and all of them were deemed suitable for further analysis, which is a 100% success rate. Therefore, the identified hot spots are more reliable in the second round of collections.

Table 1. Comparison between two rounds of data collection.

	Round One	Round Two
Data logger
Data logger interval	1 min	20 s
Device accuracy	±0.9 °F/±0.5 °C	±0.2 °F/±0.1 °C
Groups	2 groups (2 person per group)	5 individuals
Data collection time slots	7:00 a.m.–10:30 a.m. 12:30 p.m.–5:30 p.m. 5:45 p.m.–8:00 p.m.	9:00 a.m.–10:00 a.m. 12:00 p.m.–1:00 p.m. 5:00 p.m.–6:00 p.m.
Route planned	No	Yes

summarizes the major differences in method between round one and round two of data collection.

4.3. Fixed Sensor: Rooftop

Sensors were installed on the rooftops of ten buildings (refer to Figure 6a for building location) on August 22nd, utilizing the same devices from the initial round of traverse data collection to ensure data quality consistency. Due to the device’s battery life, they required recharging before the second installation on October 29th. Two factors determined the selection of installation sites. First, the buildings situated in hot spots identified during the first traverse data collection phase were selected. Second, the primary goal of installing fixed sensors was to enhance our understanding of how different roof materials mitigate UHIs, specifically temperature fluctuations; thus, buildings with varied roofing materials were targeted.

The campus features five types of roof materials: KEE roofs, green roofs, white roofs, EPDM roofs, and slate roofs. Figure 6b demonstrates the variety of roofs on campus. Table 1 details the Solar Reflectance Index (SRI) of the different roofing materials. The SRI quantifies a roof’s ability to remain cool in sunlight by reflecting solar radiation and emitting thermal radiation. It is calculated by comparing the surface temperature under standard conditions to that of a standard black (SR = 0.05, emissivity = 0.90; SRI = 0) and standard white (SR = 0.80, emissivity = 0.90; SRI = 100) surface.

KEE, or ketone ethylene ester, is a high-molecular-weight polymer with high flexibility, commonly used in commercial roofing. Known for its durability and ability to maintain flexibility and strength, it is available in multiple colors, with white being the most common due to its reflective properties that can reduce cooling costs and improve a building’s energy efficiency. Sensors were installed on O’Neill Hall and Morris Inn.

Notre Dame is home to Indiana’s largest extensive green roof, measuring 79,096 ft² on the Joyce Center. Other green spaces include the Morris Inn, Corbett Family Hall, Duncan Student Center, and O’Neill Hall. Collectively, at over 122,000 square feet, the University’s green roofs comprise the largest vegetative roofscape in Indiana. O’Neill Hall and Morris Inn were selected to install two fixed sensors.

A white roof, also referred to as a ‘cool roof’, is designed to reflect a substantial percentage of sunlight, thus decreasing heat absorption by the building. This can lead to cooler interior temperatures, lower energy costs for cooling, and mitigation of the urban heat island effect. White roofs usually feature a coating of solar reflective material or other highly reflective components. Eck Law School and Hessert Aerospace Research Center were chosen for sensor installation on their white roofs.

EPDM, or ethylene propylene diene monomer, is a synthetic rubber roofing membrane known for its durability and widespread use in low-slope buildings. It is typically black and available in various widths and thicknesses. EPDM can be installed through different methods and is recognized for its ease of installation and cost-effectiveness. Sensors were deployed on the Multidisciplinary Engineering Research Building and the Fitzpatrick Hall of Engineering, which have these materials.

Slate roofing, comprising mainly natural slate tiles, is lauded for its durability, fire resistance, and aesthetic appeal. Used on a variety of buildings, from historic structures to residential homes, slate typically has a darker color and lower reflectivity, leading to greater solar radiation absorption and potentially higher cooling needs. Sensors were installed on Alumni Hall to examine this aspect.

4.4. KMeans Classifier and Image Processing

UHIs are usually caused by environmental factors in the built environment. Two of these factors include vegetation and the density of the buildings and their height in the built environment. However, extracting these factors to predict UHI behavior is not an easy task in heat mapping projects. Therefore, one of the objectives of this research was to extract these factors and connect them to the UHI maps.

Machine learning methods are among the methods that have started to being used recently for predicting the UHIs in the United States in related studies. As demonstrated in Figure 5, the campus area was divided into grids for better analysis of the UHIs and converting the outcome into a mathematically known matrix form to apply statistical methods. As a noteworthy innovation and lesson learned, this study uses these methods not only to predict the UHIs but also for data collection purposes of vegetation data and building density for these grids.

As the first step of this approach, the terrain view of the bounding box was downloaded from Google Maps images. The default image from Google Maps includes many different colors which makes the data extraction task difficult. Therefore, KMeans as an unsupervised clustering machine learning method was applied to the image to flatten different vegetation colors into one. This method also flattened the buildings’ colors enabling the building density calculation. Figure 7 demonstrates the results of the Google Image processing after applying the KMeans clustering algorithm. KMeans is calculated through Equations (1)–(3) where $\left|\right|{\mathrm{x}}_{\mathrm{i}}-{\mathsf{\mu }}_{\mathrm{k}}\left|\right|$ is the Euclidean distance between data point $x_i$ and the cluster centroid ${\mu }_k$. Equation (2), where N_k is the number of data points in the cluster, recalculates the centroids of each cluster by taking the mean of all data points assigned to that cluster. Afterward, the algorithms repeat the assignment and update steps until the centroids no longer change significantly, or a predefined number of iterations is reached. Equation (3) indicates the objective function of KMeans clustering, which the algorithm aims to minimize and is given by the within-cluster sum of squares (WCSS) [37].

$$\mathrm{Assign} {\mathrm{x}}_{\mathrm{i}} \mathrm{to} \mathrm{cluster} \mathrm{k}:\underset{\mathrm{k}}{\min }|{\mathrm{x}}_{\mathrm{i}}-{\mathsf{\mu }}_{\mathrm{k}}{|}^2$$

(1)

$${\mathsf{\mu }}_{\mathrm{k}}={\displaystyle \frac{1}{{\mathrm{N}}_{\mathrm{k}}}}\sum _{\mathrm{i}=1}^{{\mathrm{N}}_{\mathrm{k}}}{\mathrm{x}}_{\mathrm{i}}$$

(2)

$$\mathrm{J}=\sum _{\mathrm{k}=1}^{\mathrm{K}}\sum _{\mathrm{i}=1}^{{\mathrm{N}}_{\mathrm{k}}}|{\mathrm{x}}_{\mathrm{i}}-{\mathsf{\mu }}_{\mathrm{k}}{|}^2$$

(3)

After implementing the KMeans clustering on the Google Maps image, the same grids as for Figure 5 were applied to the image, which resulted in Figure 7. In the demonstrated image, the green pixels demonstrate the vegetated area and the brown ones are the buildings and constructed area. Therefore, the area of the vegetation was calculated through dividing the number of these pixels to total pixels of the grids. Equations (4) and (5) show the formulae used to calculate these properties.

$$Vegetation Coverage={\displaystyle \frac{Number of Pixels with RGB of Green in {Cell}_i}{Total Number of Pixels in {Cell}_i}}$$

(4)

$$Construction Density Coverage={\displaystyle \frac{Number of Pixels with RGB of Brown in {Cell}_i}{Total Number of Pixels in {Cell}_i}}$$

(5)

The same approach, using the combination of USGS Lidar Heightmaps, resulted in the buildings’ density (volume) ratio in each cell. As demonstrated in Figure 8, the map is a black and white map where the whiter values demonstrate higher buildings. Therefore, the brightest white values in the image were considered as coefficient 1 and the other heights were calculated based on a proportion of their brightness compared to this value, resulting in a representation of height ratio. Equation (6) indicates the equation to solve this problem.

$$The Building Volume Density={\displaystyle \frac{Height Coefficient\times Bright Pixels Area}{Total Cell Area}}$$

(6)

The outcome of this approach led to a matrix-based representation of the geographical data enabling the application of statistical methods in a GIS-based framework. Figure 9 demonstrates a sample of methods being enabled using this approach. The darker colors in this heatmap demonstrate the higher density of the analyzed subject.

5. Surface Urban Heat Islands (SUHIs)

The technique used for measuring SUHIs is mainly thermal remote sensing, which employs non-contact instruments that detect longwave or thermal infrared radiation to estimate the surface temperature. This method provides a spatial view of the urban surface; however, it is important to note that not all surfaces may be seen, and the measurements are relative, often requiring corrections for atmospheric and surface effects.

5.1. Data Collection Using a Thermal Drone Camera

On 17 October 2023, the facility management team at the University of Notre Dame employed a drone equipped with a thermal camera to capture thermal images of ten buildings that had fixed sensors installed on their roofs. The drone used was a DJI Inspire I Pro Black, paired with a Zenmuse XT thermal camera that boasts a resolution of 640 × 512. The local weather station at the airport reported temperatures ranging from 5.6 to 17.2 degrees Celsius on that day.

While the ideal perspective for capturing a thermal image of a building’s roof is from directly overhead, practical constraints often necessitate alternative approaches. To circumvent this issue, the multiple non-ideal thermal images (as shown in Figure 10, upper row) were synthesized using advanced stitching and distortion correction techniques. This process resulted in a composite image that closely replicates the desired top-down view (as shown in Figure 10, lower row). An accessible method to accomplish this synthesis is by employing the photomerge tool in Adobe Photoshop. The RGB images were transformed into grayscale using the ‘iron’ color palette for reference. In the grayscale image, areas of higher thermal intensity are depicted as white; thermal intensity represents the roof surface temperature at a particular spot in relation to the average roof temperature. To ensure precision in the resultant distribution and descriptive statistics, a mask was applied over the roof area to exclude non-roof regions. Ultimately, matplotlib and Pandas within Python were utilized to generate box plots that visually represent the distribution of thermal intensity across pixels, accompanied by pertinent descriptive statistics.

5.2. Roof Temperature Analysis

Upon analyzing the thermal data from various rooftops, distinct temperature patterns were identified. As illustrated in Figure 11, notably, the O’Neill and DeBartolo buildings maintained a more consistent temperature range, in contrast to the Fitzpatrick Hall and Engineering North, which demonstrated greater temperature fluctuations. The Facilities Building stood out with the highest temperature readings, attributed to its function as a power plant, whereas the Morris Inn, with its green roof, recorded the lowest temperatures.

Thermal intensity variations and average temperatures were summarized with detailed statistics to enhance accuracy.

Table 2. Descriptive statistics of relative thermal intensity distribution for various rooftop areas.

Building Name	Roofing Type	Mean	std	min	25%	50%	75%	max
O’Neill	White/green roof	0.38	0.20	0.01	0.23	0.38	0.50	1.00
morris inn	Green roof	0.37	0.23	0.02	0.19	0.30	0.52	0.97
hessert	White roof	0.41	0.24	0.02	0.18	0.42	0.60	0.97
fizpatrick	KEE roof	0.44	0.27	0.01	0.20	0.42	0.66	1.00
facilities building	KEE roof	0.51	0.25	0.02	0.26	0.55	0.71	0.96
engineering north	EPDM	0.39	0.29	0.02	0.13	0.29	0.64	0.97
eck law	Slate roof	0.40	0.25	0.02	0.18	0.38	0.56	1.00
DeBartolo	Stone ballast	0.40	0.20	0.01	0.23	0.43	0.52	1.00
Alumni	Slate roof	0.38	0.26	0.01	0.16	0.28	0.55	0.97

shows that Engineering North’s rooftop exhibited the most significant temperature variability, while DeBartolo’s and O’Neill’s lower variability may reflect the thermal properties of different roofing materials. These temperature discrepancies may suggest that KEE and EPDM roofs exhibit less uniform heat distribution compared to white, green, and stone ballast roofs. An ANOVA test was conducted to understand whether a green roof (i.e., white, green roof) performed different than a non-green roof (i.e., KEE roof, EPDM, slate, stone roof). The results showed there is no statistically significant difference. Additional factors such as roof size, equipment layout, and the influence of surrounding vegetation are proposed to further understand how roofing materials affect temperature.

6. Discussion

6.1. Contribution of the Proposed Framework

The proposed framework’s multi-modal and multi-temporal measurement sets distinguish it from previous single-day and single-method data collection efforts [4,34]. The proposed framework integrates both CLUHI and SUHI measurements, providing a more holistic and robust analysis of UHIs than using either method in isolation. The framework captures a multi-dimensional view of UHIs. CLUHI measurements, which involve traverse data collection and fixed sensors, typically assess the air temperature at the pedestrian level or within the urban canopy layer, directly affecting human comfort and urban energy demand. This layer is where human activities and living spaces are most affected by UHIs. On the other hand, SUHI measurements focus on the thermal properties of surfaces using remote sensing technology. These measurements are crucial for understanding how different materials and colors of urban infrastructure absorb and re-emit solar radiation, which contributes to the overall heat profile of an urban area.

First, this integrated approach facilitates a thorough assessment of the urban thermal environment by capturing not only the ambient air temperature, which affects human comfort and urban energy dynamics, but also the radiative temperature of surfaces, that is, rooftop temperature, which plays a crucial role in the overall urban heat profile. Another significant advantage is the capacity for correlative analysis. By combining CLUHI and SUHI data, researchers can discern correlations between surface and air temperatures, providing deeper insights into how the heat emitted by various urban surfaces influences the broader UHI effect experienced within the urban canopy layer. It sheds light on the contribution of different roofing materials and urban designs to UHI intensity. These insights are invaluable for developing strategic interventions, such as selecting roofing materials with higher reflectivity or implementing green roofs to mitigate the heat island effect.

The second contribution of the proposed work is its multi-temporal nature. The multiple traverse data collection proves advantageous. By adjusting the collection procedure and taking the lessons learned from the first round into account, the accuracy, reliability, and efficiency of the second round of collection were greatly improved. Comparing the data collected on a hot summer day with a large diurnal temperature swing with those collected on a cool fall day with constant temperature, the identified overlapping hot spots have high fidelity. This allows for an examination of how UHIs fluctuate over time, offering an understanding of diurnal and seasonal variations, as well as the influence of varying weather conditions on urban heat dynamics.

Lastly, the framework serves as a potent tool for policy and planning. It equips urban planners and policymakers with a detailed dataset that is instrumental in guiding decisions related to urban design and greening initiatives. The ultimate goal is to leverage this information to mitigate the effects of UHIs, thereby reducing energy consumption and enhancing overall urban sustainability.

6.2. Comparison to Other Studies

This study distinguishes itself from other similar research in several key ways. While many studies on urban heat islands (UHIs) have employed either stationary meteorological stations or satellite remote sensing to measure temperature variations [38,39], our study integrates a multi-modal and multi-temporal data collection framework. This comprehensive approach combines mobile surveys, stationary sensor networks, and drone-based thermal imaging to provide a more granular and detailed analysis of urban microclimates. Unlike traditional methods that offer snapshots of temperature distributions, our study’s methodology captures continuous spatial representations and temporal variations. For instance, previous studies in cities like Dijon, France, and various urban areas in Greece and China have relied on extensive networks of fixed sensors to evaluate UHI intensity [4,38]. These methodologies, while invaluable, often miss the finer spatial and temporal nuances due to their static nature.

In contrast, our research leverages the dynamic capabilities of mobile data collection, allowing us to map detailed temperature variations across different times of the day and under varying weather conditions. This integration of different data sources enhances the accuracy and reliability of our findings. Moreover, the cost-effective nature of our approach makes it accessible for a wider range of urban areas, particularly those with limited resources. Furthermore, while satellite thermal imaging provides extensive coverage for SUHIs, it often lacks the fine resolution needed for street-level analysis and is affected by obstructions like cloud cover [40]. By incorporating drone-based thermal imaging, our study overcomes these limitations, providing detailed thermal maps that are critical for understanding localized heat patterns.

Overall, this study adds additional knowledge to the field by offering a holistic and detailed methodology for UHI assessment, facilitating more effective urban planning and heat mitigation strategies.

6.3. Limitations

The proposed methodology and technologies are not without limitations. As for the first limitations of the proposed workflow, the accuracy limitations of the technological devices should be mentioned. The devices used in this study had high accuracy (Elitech RCW-360 with ±0.5 °C (±0.9 °F) and Elitech GSP-6 PTE Ultra Low Digital Data Logger ±0.2 °F/±0.1 °C); nevertheless, even this amount of error can cause bias if the temperature difference between the UHI and non-UHI regions is not significantly high. In other words, this bias might put some regions inside or outside of the UHI threshold. The lesson learned from this limitation led these researchers to conduct multiple data collection rounds to verify the results. Moreover, the double thermocouple was used to minimize this bias.

The second limitation caused by the data loggers is caused by the materials used in the data loggers. Similar to the feeling of body temperature, the environmental temperature does not hit the data loggers instantly, and it is measured by the change in the temperature in the data loggers’ metal alloy. This feature is specifically noticeable when using mobile sensors that connect the coordinates and record temperature in a timely manner. The thermocouple’s response time is significantly higher than the thermometer and this study used these data logger types to reduce the temporal errors. However, a certain amount of temporal errors is inevitable while using mobile data loggers. Therefore, researchers should consider the movement pace and their sensors’ response time in deploying these devices as mobile ones. Moreover, the GPS influence might be impacted by weather conditions, such as cloud coverage, and the difference between the satellite data transmission and temperature record should be considered. This research estimated 5 s as an experimental number to cover this difference. However, there is no certain way to calculate the exact amount of these temporal differences, and a certain amount of error cannot be avoided. Considering all these factors, the on-site data collection temporal biases are severely lower than those of satellite imaging methods.

In addition to the temporal differences, the spatial errors in translating the collected geographical data to numeric datasets should be considered. This research used grids to convert the coordinate-based collected temperatures to the data matrix for further analysis. This method involves a certain amount of spatial error in allocating the data points to each grid for the data close to the border of each grid cell. This means if the high-temperature data points are located alongside the cell’s border on one side, one of the cells might be identified as the UHI region while the adjacent cell is identified as a non-UHI region. However, an adequately high number of data points (low interval times) reduces the possibility of such an error. Moreover, currently, there is no better way available to eliminate this type of error.

7. Conclusions

The study conducted at the University of Notre Dame offers pivotal insights into urban heat islands (UHIs), utilizing a novel combination of traverse data, fixed sensors, and thermal imaging to understand UHIs on a micro-urban level. This multifaceted approach deepens our understanding of temperature variations in urban areas and underscores the significance of strategic urban planning in mitigating heat effects. The research lays the groundwork for future studies in sustainable urban heat management and highlights the importance of innovative methodologies in addressing environmental challenges. This initiative paves the way for further research aimed at developing practical strategies for urban heat mitigation and bolstering climate resilience in urban environments. The next steps are twofold. First, we plan to install additional fixed ground and rooftop sensors on buildings in identified hotspots, leveraging Wi-Fi connections for continuous temperature monitoring. Second, we aim to discern the most impactful contributors to UHIs by examining three categories identified from prior research: building morphology, vegetation cover, and the materials of buildings and roads.

Future work will focus on adapting the methodology to various urban environments by developing specific strategies for different climatic conditions, urban densities, and vegetation types. Conducting case studies in diverse settings will help validate and refine the approach, ensuring broader applicability. Additionally, comprehensive assessments of urban thermal comfort will be integrated, evaluating physiological and psychological responses to various thermal environments. This will provide a holistic understanding of how urban green spaces can enhance human well-being.

Recognizing the critical role of nature-based solutions (NBS) in mitigating urban heat island (UHI) effects, future research will emphasize these strategies. Comparative analyses of different NBS, such as urban forests, green roofs, and green walls, will be conducted to identify the most effective approaches. Furthermore, a global-scale analysis of UHI mitigation through urban green spaces will be pursued by collaborating with international research teams. This will enable cross-comparative studies and the development of universally applicable guidelines, contributing to resilient and sustainable urban environments and providing actionable insights for urban planners and policymakers.