Application of remotely sensed data for spatial approximation of urban heat island in the city of Wrocław, Poland

The study addresses the issue of potential usefulness of remotely sensed data and their derivatives for urban heat island (UHI) modeling. The methodology is illustrated with examples of selected UHI cases in Wrocław, a mid-sized city in SW Poland. Three cases of UHI (early summer, autumn and winter) are analyzed with equivalent remotely sensed data. Measurements of air temperature in each case were done by mobile meteorological stations, and available from 206 sites. Corresponding Landsat ETM+ and LIDAR-originated data were prepared and cover: albedo, selected vegetation indices (NDVI, SAVI, NDMI), emissivity, land surface temperature, roughness length, porosity, sky view factor and sums of daily solar irradiance. All these spatially continuous parameters were filtered using focal mean to simulate the role of source area around measurement site. Circular matrices, with radii varying from 25 to 1000 m, were applied in filtering procedure. Next, correlation analysis was used to determine the most influencing variables for each UHI case. The best correlations were achieved while considering the area of 550-600 m from a given measurement site. Regardless the seasons, the most influential factors for air temperature are: albedo, roughness length, sky view factor and sums of daily irradiance. Some parameters are significant only seasonally, e.g. vegetation indices in summer. Because spatial variables are in most cases multicollinear, step-wise regression supported with the analysis of variance inflation factor was used to determine final multiple linear models. Statistically significant models explain from 71% to 85% of the air temperature variance.