A cost-effective method for producing spatially continuous high-resolution air temperature information in urban environments

Abstract

Sustainable city planning requires detailed information on spatial temperature variations. Remotely sensed land surface temperature (LST) is known to differ substantially from air temperature (AT) causing misinterpretations of the ambient conditions. We demonstrate a reliable and cost-efficient method for AT modelling in urban environments using open data and few temperature observations. The study area is the city of Turku SW Finland, where we have a dense in situ AT observation network of 64 Onset Hobo temperature loggers as a reference. Landsat 8 thermal data from different seasons were used to extract pixel-based LST by employing MODIS and ASTER emissivity libraries and CORINE land cover classification. The LSTs were analysed against the in situ AT first with the correlation analysis. Except for December, the Pearson's correlation coefficients were statistically significant (0.449–0.654, p ≤ 0.001). Seasonally adjusted linear regression models were applied to predict spatially continuous air temperatures (ATp) based on the extracted LST. Our results demonstrate that it is possible to predict urban ATs reliably - within ca. half-a-degree accuracy (MAE 0.36–0.62 °C). The prediction works best in spring, summer and autumn. It improves the capacity to produce reliable high spatial resolution AT information even if in situ observations are sparse.