GIS-based techniques for urban heat island spatialization

This paper addresses the issue of usefulness of selected spatialization techniques for the characterization of an urban heat island (UHI). Five interpolation methods (including both deterministic and stochastic methods or their combination) namely: inverse distance weighting (IDW), regularized spline with tension (RST), ordinary kriging (OK), multiple linear regression (MLR) and residual kriging (RK)-were evaluated for their ability to estimate air temperature in Wroclaw, Poland, during 7 cases of the UHI. Spatial interpolation was performed based on time-adjusted air temperature data gathered by mobile measurements. Additional explanatory variables for multidimensional spatialization methods (MLR and RK) were developed based mainly on the land-use map and Landsat thematic mapper (TM) images. Statistically significant predictors were selected using a stepwise regression procedure. Parameters for optimal interpolation were chosen by cross-validation (CV) of results. The CV technique was also used to compare results obtained with the different algorithms together with evaluation of errors (e.g. root mean square error, RMSE; mean absolute error, MAE) and visual examination of the final maps. The least plausible maps, both in terms of error statistics and visually, were obtained with the IDW method. Inside the convex hull of sample points, the OK and RST techniques were characterized by simplified but acceptable air temperature surfaces. The MLR method expressed the land-use background of the UHI, even outside the convex hull, but distorted results when the process tended towards non-stationarity, e.g. clue to wind influence. The most accurate results of the UHI spatialization were obtained with the RK technique.