Drone remote sensing to define heat exchange between urban surfaces and stormwater runoff

The proliferation of impervious surfaces and reduced vegetation within cities has led to urban heat islands, where the temperature of urban areas is higher than surrounding non-developed land. The primary causes of urban heat islands are the prevalence of low albedo surfaces, low evapotranspiration (ET) rates, and anthropogenic heat production (Akbari et al. 2001; Stone et al. 2010; Jones et al. 2012). The resulting higher temperatures can lead to increased energy costs for air conditioning, air pollution, and heat-related illnesses or deaths (Wong et al. 2013; Li et al. 2019; Piracha & Chaudhary 2022). They also have an environmental impact, such as increased urban water body temperatures from reduced shading, channelization, and increased runoff temperatures (Timm et al. 2021). The increased temperatures of urban water bodies are known as the hydrologic urban heat island, where urban streams have comparatively higher baseflow temperatures and are subject to greater temperature surges from stormwater runoff (Zahn et al. 2021).

The hydrologic urban heat island can have a profound negative effect on water quality and aquatic health. In urban water bodies, temperature can alter reaction rates, causing an imbalance in important regulating nutrients and compounds. Streams subject to warm urban runoff may contain lower concentrations of dissolved oxygen, impacting respiration and algal production (Butcher 1995; Griffith & Gobler 2020), as well as increased contaminant toxicity (Patra et al. 2015). Furthermore, stream temperatures regulate important ecosystem processes, such as fish migration patterns, reproduction, growth, immune responses, and competitive ability (Armour 1991). The importance of stream temperatures to water quality and aquatic health has led to temperature being the top overall impairment in freshwater bodies in several states (Gunawardana & McDonald 2023). Therefore, it is critical to understand the processes that lead to increased stream temperatures from urban stormwater runoff.

It is unclear; however, the degree to which highly heterogenic urban areas contribute to increased stormwater temperatures. Urban areas are spatially complex and contain a variety of land cover types that are subject to different environmental and anthropogenic forces, such as shading from buildings, vehicular traffic, weathering, and pedestrian use. Therefore, estimating the heat exchange in stormwater runoff from urban surfaces is a challenge. Those studies that do exist rely on in situ monitoring that capture the heat exchange that occurs within a constrained surface type, location, and conditions (Thompson et al. 2008; Herb et al. 2009a; Janke et al. 2009; Omidvar et al. 2018), but may not be representative of land surfaces that are subject to various levels of environmental and anthropogenic effects. It is therefore unclear how different land cover types contribute to heat exchange in stormwater and what specific urban watershed characteristics control thermal processes in stormwater.

One way to overcome this challenge is through the use of thermal imagery from small unmanned aerial systems (sUAS) or drones that can capture land surface temperature data in high spatial resolutions. Drones are able to supplement in situ measurements with data in a high spatial resolution and at on-demand time scales that can provide insights into hydrologic processes within an urban watershed (McDonald 2019). For example, in urban stormwater, drones have been applied to quantify plant stress in urban green stormwater infrastructure (Prettyman et al. 2021), measure urban stream velocity (Tauro et al. 2016; Shariar et al. 2023), determine urban pond geometrics and water ponding (Hassan et al. 2023; Zhao et al. 2023), and assess stream temperatures (Fitch et al. 2018; Kuhn et al. 2021). In the context of urban surface temperatures, while in situ monitoring limits data collection to discrete points in space, thermal data from drones can provide a spatially distributed estimate of surface temperature (Naughton & McDonald 2019). This data, combined with in situ monitoring, can provide additional insights into the influence of surface temperatures on stormwater runoff.

The objective of this study is to understand the impact that the composition of catchment surfaces has on heat transfer in urban stormwater runoff. To meet this objective, this study monitors four urban catchments that are dominated by different land surfaces (concrete parking lot, asphalt road, black bitumen roof, and grass). Monitoring includes both in situ monitoring of stormwater runoff at the catchment outlet using water level, acoustic Doppler velocimeter, and temperature sensors, as well as thermal imagery from a sUAS before and after storm events. Integrating this data together, this study presents a novel approach to evaluate the effect that land surface temperatures of specific land cover types have on heat exchange in stormwater runoff. Doing so provides a greater understanding of heat exchange in urban stormwater systems that can be used to target mitigation measures to protect downstream water quality and human health.

In situ equipment included spatially discrete, temporally continuous data loggers that monitored temperature, runoff, and atmospheric conditions (Supplementary Figures SI-1–SI-4). Devices to measure temperature and runoff were placed at the outlet of each catch basin. Temperature loggers were created using EnviroDIY Mayfly Data Logger boards, 64 gigabyte microSD cards, uxcell 6 V 60 mA Mini Solar Cells, 3.7v lithium polymer batteries, and temperature sensors. Runoff flow rates were captured with two types of sensors: Global Water Instruments flow loggers and ISCO 2150 loggers. Global Water WL 400 water level sensors that measure both temperature and flow depth were placed at the catch basin outlet (i.e., curb inlet or grate inlet) and were programmed to collect at 5-min intervals. Two ISCO 2150s that capture water level and velocity were placed within the sewer network below selected catch basins to validate the water balance. The ISCO 2150s use acoustic Doppler velocimetry and were programmed to log data at 5-min intervals. Finally, an Onset HOBO temperature/relative humidity sensor, wind speed and direction sensor, and precipitation bucket were installed within the greenspace catchment.

Land surface temperature was captured using a DJI M100 drone with a Zenmuse XTR thermal camera and flights were performed using Pix4Dcapture software. Flights took approximately 15 min at 300 ft with an overlap of 90%, depending on weather conditions. The Pix4D desktop application was then used to stitch the individual land surface temperature images into a single mosaic and the data were spatially processed and analyzed using ArcGIS Pro and Python. In addition, land surface temperature was validated using spot measurements from a FLIR TG54 Spot infrared sensor. The sUAS captured land surface temperature at a high resolution both before and after storm events and throughout the day to capture diurnal trends in land surface temperature. To capture land surface temperature before storm events, the sUAS was deployed prior to when rainfall would begin based on a variety of weather forecasts. To capture the diurnal trends in land surface temperature, sUAS flights were performed at 09:00, 12:00, 15:00, and 17:00 across three representative summer days: (1) a hot summer day, (2) a cool summer day, and (3) an overcast summer day (Supplementary Figure SI-8).

From the in situ data, three analyses were performed to analyze the characteristics of each runoff event including analysis of (1) the temporal relationship between wet bulb (rain), surface, and runoff temperature, (2) the change in runoff temperature during discrete runoff events, and (3) the event mean temperature (EMT) of each storm.

Next, the behavior of runoff temperature during an event was computed by calculating the number of fluctuations, or dips, in temperature during precipitation events and the overall magnitude of temperature change. This was done because in the beginning of an event, the runoff temperature is usually highest due to the initial small depth of runoff and high rate of heat exchange between the land surface and runoff. Over the course of an event, runoff temperature typically decreases due to an increase in the flow rate or volume of stormwater runoff, and then subsequently increases as flow rates decrease (Herb et al. 2009a). These fluctuations, or dips, can inform both the hydrologic timing of heat exchange, as well as the magnitude of heat exchange that occurs between the surface and runoff. To define these dips in the temperature time series, precipitation events were first segmented and defined using a 1-h antecedent dry period and an initial rainfall volume of 0.01 inches. Overall differences in runoff temperature were calculated by subtracting the initial temperature from the final temperature. Finally, within these events, the number and magnitude of temperature dips in were counted, defined as the presence of a local decrease in runoff temperature concurrent with a local precipitation peak.

Data were collected between July and October 2022, during which 14 storms were captured. During some of the storm events, there was incomplete data due to several factors including power failure, sensor errors, or theft. In particular, the roadway sensors were stolen at the end of August; therefore, the in situ data presented is segmented based on data availability. A complete list of available in situ data for each storm event is provided in Supplementary Table SI-7. This data was used to evaluate the temporal characteristics of runoff temperatures, their inter-event dynamics, and the EMT in the following sections.

Overall, these results indicate that catchments are distinct in their intra-event energy exchange response to rainfall. The roadway catchment typically experiences the greatest energy export dip counts and magnitudes. While the grass has a high change in temperature per dip, it also has the lowest number of dips per storm, possibly due to the high time of concentration (Supplementary Table SI-1); therefore, even though the grass has a similar temperature change per dip as the roadway, it has less frequent fluctuations in temperature. Finally, the roof catchment had the lowest temperature change per dip and a high standard deviation, possibly due to the air-conditioned subsurface, leading to more rapid thermal equilibrium between the roof surface and the runoff temperatures.

The parking lot exhibited higher EMT than the roof catchment during summer months (Figure 6(b)) despite having comparable overall temperature changes (e.g., Figures 3 and 4). This indicates that the parking lot is prone to high land surface temperatures similar to the roadway catchment, but intra-event dynamics that govern thermal equilibrium and thermal load operate differently, perhaps due to catchment characteristics. On the other hand, the roof catchment produced low EMT and low intra-event dynamics. This could be due to the roof subsurface being cooler than a terrestrial subsurface, leading to a quicker thermal equilibrium and therefore lower land surface temperature as rain events proceed.

To further define land surface temperature variability, the standard deviations of both the complete catchments and individual surface types were quantified using the sUAS land surface temperature data (Figure 8(a)). The parking lot exhibited the widest interquartile range of 1.22 °C and highest median of standard deviations at 2.24 °C. The roadway catchment has the lowest interquartile range, but second highest median. The in situ analysis found that the roadway has the highest energy export and EMT values, and these results indicate the standard deviation of the roadway catchment land surface temperature is also the most consistent. Overall, the roof and grass catchments exhibited the lowest standard deviations, which may be because these catchments are more homogeneous than the other catchments; the roadway and parking lot are made up of fragmented medians, concrete, asphalt, canopy cover, etc., compared with the roof and grass catchments which are continuously a roof or grass cover, respectively. Their surface homogeneity leads to more consistent land surface temperatures throughout the catchments. In addition, categorized by specific land cover type (Supplementary Figure SI-6), natural land cover types (e.g., grass and shrub/mulch) catchments had lower variations in temperature than impervious surface types (e.g., asphalt and pavement).

This research presents a novel sUAS and in situ remote monitoring approach that demonstrates distinct thermal variability of urban surfaces during stormwater runoff. The monitoring campaign captured rainfall, runoff, and temperature data of four urban catchments using in situ and sUAS methods to compare trends in temporal thermal response, intra-event energy exchange dynamics, EMT, and spatial variability due to differences in catchment composition. The results show urban catchments have a high degree of spatial variability in land surface temperatures due to their unique catchment compositions, leading to distinct temporal responses.

Over the course of storm events, runoff temperatures decreased but remained higher than rainfall temperature, indicating that runoff was exporting stored energy from the surface of the catchments. These temperature changes happened at different rates throughout each catchment and are likely dependent on catchment characteristics such as time of concentration and catchment size; thermal characteristics such as albedo and subsurface energy fluxes; and storm characteristics such as magnitude and duration. The roadway catchment experienced the most energy loss across storm events, the most frequent energy export events within storm events, and the highest EMT. This is likely due to the catchment characteristics, such as the second smallest drainage area, high imperviousness (95%), and low albedo.

In this study, the temperature of runoff from impervious surfaces responded differently between buildings and those with a ground subsurface. While the roof exhibited one of the highest initial land surface temperatures, it consistently produced the lowest EMTs. These trends are due to the rapid thermal responsiveness exhibited by the surface, which produces steep energy exchanges at the beginning of storm events that plateau at a low thermal equilibrium. These responses are likely due to the presence of a weaker and cooler subsurface energy flux, unlike the other three terrestrial catchments. Therefore, the overall thermal impact of impervious surfaces is not equivalent across impervious surfaces due to the quicker cooling of roof surfaces as compared with ground-based impervious surfaces (e.g., roads, sidewalks, and parking lots).

Catchments and land surfaces exhibited different ranges of land surface temperatures and land surface temperature variability from the sUAS data, even for those land surfaces of similar types. For example, concrete surfaces responded with different land surface temperatures and standard deviations across the roadway, grass, and parking lot catchments. Additionally, similar surfaces exchanged energy at different rates between catchments demonstrated by the distinct standard deviations of temperature differences due to storm events. These results indicate a land surface temperature as an effective parameter (i.e., assuming that it is the same across an entire land cover type) could misrepresent the energy exchange dynamics due to unique surface characteristics such as local shading, age of surface, and traffic loading, among others.

From this study, there were advantages and limitations to using thermal imagery from a sUAS to evaluate heat transfer in the catchments. An advantage was the collection of surface temperature data in high spatial resolutions that can overcome overly generalized assumptions of surface temperatures and properly capture variability of land surface temperatures within a catchment. However, this data was limited by the temporal resolution of the data, which required the ability to fly the sUAS in adequate weather conditions. Temporal analyses of land surface temperature change and diurnal periodicity highlighted how imprecise flight times leave land surface temperatures vulnerable to diurnal periods. However, this vulnerability is not ubiquitous across catchments; the rapid thermal responsiveness of the roof catchment led to faster land surface temperature changes in the absence of rainfall and presence of clear skies. Precise flight times corresponding to storm events would minimize the potential skew introduced by a diurnal period, although the diurnal period of storm days is subdued and less extreme as compared with a hot and clear day.

The results from this study can be used to inform effective implementation of best management practices (BMPs) to mitigate high runoff temperatures. BMPs such as green stormwater infrastructure have been shown to reduce runoff temperatures (Timm et al. 2020; Gunawardana et al. 2023; Bodus et al. 2024), but it may not be clear where the best locations for implementation would be for reducing downstream temperatures. From these results, it is clear that impervious surfaces on the ground (i.e., roadway and parking lot catchments) produce higher thermal impacts than those that represent the roof of buildings (i.e., roof catchment). Therefore, if reducing thermal impacts of stormwater runoff is a primary goal, BMPs might be more effective in treating stormwater runoff from impervious surfaces at the ground level, rather than those that collect runoff from buildings.

This paper presents a unique monitoring study that integrates in situ and remote sensing of temperature to evaluate the heat exchange in stormwater runoff from urban surfaces. Results indicate that runoff temperatures decreased but remained higher than rainfall temperature, indicating runoff was exporting stored energy from the surface of the catchments. Land surface temperature data from drones indicated a wide variability in temperature among common land surfaces (1.34–2.24 °C), with a higher variation in surfaces subject to foot and vehicular traffic. In addition, the temperature of runoff from impervious surfaces responded differently between buildings and those with a ground subsurface, with higher event mean temperatures from concrete (21.4 °C) and asphalt (21.9 °C) ground surfaces as compared with the bitumen roof (19.8 °C), despite higher initial surface temperatures of the roof. Future work could apply the proposed approach to further define the factors that contribute to differences in runoff temperature between impervious surfaces of buildings and those with a ground subsurface. Ultimately, these outcomes help to improve our understanding of heat transfer in stormwater runoff within complex urban systems and can help to guide the implementation of BMPs to protect downstream water bodies.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.