Geospatial data analyst at 3DOM research unit, FBK Trento, Italy
Heat waves are more and more heavily affecting population and this is even more enhanced in cities rather than in the countryside where the urban heat island (UHI) effect worsen their duration and intensity. Current research on the estimation of the UHI effect adopts 3 main approaches: i) observational studies describing its driving processes, spatial patterns and/or magnitude; ii) Earth observation (EO) studies focusing on the surface urban heat island (SUHI) determined from land surface temperature (LST) retrieved from satellite thermal sensors (e.g. Landsat-TIRS, Sentinel-SLSTR); iii) modeling via mesoscale meteorological models like the Weather Research and Forecasting (WRF) or microscale models (i.e. ENVI-met) that require large computational effort and/or fine tuning of parameters. Municipalities need actionable data to support their decisions. It is thus crucial to develop intermediate approaches for the estimation of the UHI intensity and spatial extension without the need of advanced expertise to tune parameters or run complex meteorological models but, at the same time, able to provide reliable insight into the urban air temperature. The work presented in this contribution is performed in the framework of the USAGE project activities and focuses on providing a pipeline for the development of UHI maps in urban areas utilizing open data like EO, IoT ground sensor data, surface properties and a hybrid model based on machine learning and geostatistics. We present a pipeline that can be deployed with minor adaptation (i.e STAC end points) within GIS software environments. The ground sensor data are accessed via OGC SensorThings API and fed into the analysis. Pre-loaded 'semi-static' layers, like DTM, DSM, LU/LC, vegetation fraction, urban building morphology and shade maps are accessed via OGC Feature API and utilized to spatialize the air temperature at each time stamp received from the IoT sensors. Based on the revisit time of EO thermal data and its cloud-coverage level, LST observations are integrated to help the spatialization of the ground sensor's temperature, performed using a hybrid model combining machine learning and geostatistics. This allows for faster computation compared to classical geostatistics but, at the same time, to explicitly handle spatial correlation of data and errors. The aforementioned pipeline is suitable to derive UHI maps from given IoT ground sensor data. On the other end, to forecast the UHI effect up to 48 h, the pipeline ingest 2-m air temperature, relative humidity as well as wind speed and direction from the meteorological models (WRF or AROME) the open-meteo API.
The proposed pipeline is applied in the Alpine valley and city of Trento (Italy) and is then validated against high-resolution simulations with the WRF model, offline coupled with an urban parameterization scheme to reach a resolution of 100 m. The proposed pipeline can be used not only as a forecasting tool, but also as a UHI mitigation and planning tool by changing the ‘semi-static’ layers that involve the study area. This allows municipalities to predict the effects of their decisions.
Urban land use and surface properties play a major role in determining the quality of life for citizens and for urban planning. They have a strong impact alongside extreme events and phenomena, such as flash flooding due to heavy rains on a highly impermeabilized city, Urban Heat Island (UHI) intensification during heat waves, biodiversity reduction in green areas, etc.
To allow the study of such phenomena, high-resolution aerial and terrestrial data acquired with different sensors can be merged and processed with Machine Learning (ML) approaches in order to describe and predict the state of urban landscapes and create data-driven actionable insights.
Within the USAGE - Urban Data Space for Green Deal - project [https://www.usage-project.eu/], this work investigates the integration of multi-source data in urban areas for environmental analyses. Two pilot cities are considered, Graz (Austria) and Ferrara (Italy), where multispectral, thermal, hyperspectral and LiDAR data were acquired from aerial flights.
Firstly, all multi-modal data were processed and co-registered in order to align them. Then, the proposed workflow uses aerial hyperspectral images to classify the surface material with ML algorithms (16-18 classes, normally), thanks to the availability of spectral information in the VNIR and SWIR ranges. The material properties are used to support the calculation of land surface temperatures (LST) from the aerial thermal images acquired in the LWIR range. The operation is critical in case of special materials, like metals, that originate false temperature values in the thermal images, given their low emissivity. As ground truth for the LST estimation, a series of ground measurements are performed during the thermal flights. The comparison of the temperatures measured on the ground and from the thermal camera underlines the influence of the atmosphere, and therefore the need of rigorous modeling for the correction of atmospheric absorption and scattering. The LST values derived from aerial thermal images are compared to those retrieved from Landsat TIRS images, in order to characterize the representativeness of the Landsat pixel over the urban landscape. Finally, the LiDAR point clouds can be enriched with the outcomes of the thermal and hyperspectral analyses for a more realistic and exploitable visualization of the territory.
The proposed workflow has been tested first on the data acquired on the city of Graz in 2021, then replicated and validated on the data acquired on the city of Ferrara in 2022. The proposed methodology could be replicated also in other similar cities to gain more insight from LST retrieved from Earth Observation data that are less resolute in space (~ 70 m) but guarantee higher revisit time, thus allowing for monitoring the evolution of the land cover within the urban environment.