CLASSIFICATION AND MODELLING OF URBAN MICRO-CLIMATES USING MULTISENSORAL AND MULTITEMPORAL REMOTE SENSING DATA

Abstract. Remote sensing has widely been used in urban climatology since it has the advantage of a simultaneous synoptic view of the full urban surface. Methods include the analysis of surface temperature patterns, spatial (biophysical) indicators for urban heat island modelling, and flux measurements. Another approach is the automated classification of urban morphologies or structural types. In this study it was tested, whether Local Climate Zones (a new typology of thermally 'rather' homogenous urban morphologies) can be automatically classified from multisensor and multitemporal earth observation data. Therefore, a large number of parameters were derived from different datasets, including multitemporal Landsat data and morphological profiles as well as windowed multiband signatures from an airborne IFSAR-DHM.

The results for Hamburg, Germany, show that different datasets have high potential for the differentiation of urban morphologies. Multitemporal thermal data performed very well with up to 96.3 % overall classification accuracy with a neuronal network classifier. The multispectral data reached 95.1 % and the morphological profiles 83.2 %.The multisensor feature sets reached up to 97.4 % with 100 selected features, but also small multisensoral feature sets reached good results. This shows that microclimatic meaningful urban structures can be classified from different remote sensing datasets.

Further, the potential of the parameters for spatiotemporal modelling of the mean urban heat island was tested. Therefore, a comprehensive mobile measurement campaign with GPS loggers and temperature sensors on public buses was conducted in order to gain in situ data in high spatial and temporal resolution.