Automated detection of urban heat islands based on satellite imagery, digital surface models, and a low-cost sensor network

Today, more than half of the world’s population lives in urban areas and the proportion is projected to increase further in the near future. The increased number of heatwaves worldwide caused by the anthropogenic climate change may lead to heat stress and significant economic and ecological damages. Therefore, the growth of urban areas in combination with climate change can increase future mortality rates in cities, given that cities are more vulnerable to heatwaves due to the greater heat storage capacity of artificial surfaces towards higher longwave radiation fluxes.

To detect urban heat islands and resolve the micro-scale air temperature field in an urban environment, a low-cost air temperature network, including 450 sensors, was installed in the Swiss cities of Zurich and Basel in 2019 and 2020. These air temperature data, complemented with further official measurement stations, force a statistical air temperature downscaling model for urban environments, which is used operationally to calculate hourly micro-scale air temperatures in 10 m horizontal resolution. In addition to air temperature measurements from the low-cost sensor network, the model is further forced by albedo, NDVI, and NDBI values generated from the polar-orbiting satellite Sentinel-2, land surface temperatures estimated from Landsat-8, and high-resolution digital surface and elevation models.

Urban heat islands (UHI) are processed averaging hourly air temperatures over an entire year for each grid point, and comparing this average to the overall average in rural areas. UHI effects can then be correlated to high-resolution local climate zone maps and other local factors.

Between 60-80 % of the urban area is modeled with an accuracy below 1 K for an hourly time step indicating that the approach may work well in different cities. However, the outcome may depend on the complexity of the cities. The model error decreases rapidly by increasing the number of spatially distributed sensor data used to train the model, from 0 to 70 sensors, and then plateaus with further increases. An accuracy below 1 K can be expected for more than 50 air temperature measurements within the investigated cities and the surrounding rural areas. 

A strong statistical air temperature model coupled with atmospheric boundary layer models (e.g. PALM-4U, MUKLIMO, FITNAH) will aid to generate highly resolved urban heat island prediction maps that help decision-makers to identify local heat islands easier. This will ensure that financial resources will be invested as efficiently as possible in mitigation actions.