Characterizing Heterogeneous Environments: Hyperspectral versus Geometric Very High Resolution Data for Urban Studies

Surface imperviousness has proven to be a convenient and universal indicator to characterise environmental states and processes in the urban context. Geometric and spectral very high resolution data were hence employed in this study to quantify imperviousness for selected sites in the city of Berlin, Germany. HyMap data from 2003 and Quickbird data from 2002 were aquired for overlapping areas and compared with available information on imperviousness from the Urban Environmental Information System of Berlin. Both datasets were parametrically geocoded, the HyMap data also atmospherically corrected to match reference endmembers from field and laboratory measurements. The HyMap data were enhanced through a stratified feature space optimisation based on minimum noise fraction transformation. The optimised dataset was then analysed by unsupervised clustering and linear spectral unmixing. The resulting classes were coded for their imperviousness. The Quickbird data were pan-sharpened and the resulting 0.7 m resolution multispectral data segmented. A hierarchical and object-based classification scheme led to the classes for imperviousness mapping. A comparison revealed that hyperspectral imperviousness mapping correlates well with information from the information system, while geometric very high resolution data are not suited to properly map low imperviousness levels. Moreover, imperviousness levels over 80% appear to be underestimated in the Quickbird based analysis. It is concluded that a combination of the strengths of both data types may lead to the most useful results in the future. INTRODUCTION From a remote sensing point of view, urban areas are characterised by spatial and spectral heterogeneity alike. This setting finds its expression in equally heterogeneous environmental conditions and processes that are for example driving forces in urban climatology and hydrology or for characterising urban habitats. During the last decade, numerous indicators have been developed to describe environmental conditions or environmental change. While each approach may be valid and useful, it is an open question how a manageable framework may be elaborated to substitute or underpin a wealth of highly specific indicators with a more universal concept of meta-indicators. One universal indicator, or “meta-indicator”, that has proven its usability is surface imperviousness. Imperviousness stands for direct or indirect impact on manifold urban environmental conditions and can hence serve as a target variable – among others – for urban and environmental planning. As such, maps and cadastres have been derived for numerous urban environmental databases. One of the earliest and most elaborated datasets characterising urban imperviousness is part of the Urban Environmental Information System (UEIS) of Berlin, Germany. First investigations were conducted in the 80ies and updated regularly. However, such databases are difficult and expensive to maintain; on one hand, regular revisions are needed and on the other hand, statistical extrapolation from test areas to large urban agglomerations bear diverse uncertainties. Consequently, it would be a substantial step forward to extract reliable measures of imperviousness from remote sensing data in a (semi-) automated way. With the advent of geometric very high © EARSeL and Warsaw University, Warsaw 2005. Proceedings of 4th EARSeL Workshop on Imaging Spectroscopy. New quality in environmental studies. Zagajewski B., Sobczak M., Wrzesień M., (eds) resolution (VHR) digital sensors and object-based image analysis tools, a wealth of new applications or new ways to approach problems in the urban context of remote sensing evolved, as for example demonstrated by De Kok et al. (i), Meinel et al. (ii), or Small (iii). Airborne scanners, e.g. the High Resolution Stereo Camera (HRSC) or the Leica Airborne Digital Sensor (ADS 40), but even more sensors on satellite platforms such as Ikonos or Quickbird have started to re-shape the focus in remote sensing based urban monitoring. Nevertheless, due to the complicated geometry of urban surfaces and the spectral limitations of VHR data there has not been a breakthrough in such concepts, even that major advances are obvious. Hyperspectral airborne data offer an alternative in mapping urban imperviousness on the basis of spectrally driven analysis concepts. While sensor limitations will not allow for comparable geometric resolutions as with multispectral cameras, a high spectral and – relatively speaking – moderate geometric resolution offers different analysis concepts compared to geometric very high resolution data. Examples of successful applications of new methodological concepts include studies by Heiden et al. (iv), Herold et al. (v), or Segl et al. (vi). However, to our knowledge there has been no research on comparing the concepts of geometric and spectral VHR data concerning their potential in mapping urban imperviousness. This research hence illuminates the differences in imperviousness estimates from spectral and geometric very high resolution data, exemplified on the basis of HyMap and Quickbird imagery. In this context, a focus was put on a straightforward analysis scheme that does not require a comprehensive endmember collection strategy and may hence be largely automated in the future. METHODS HyMap data were acquired during the HyEurope 2003 campaign over Berlin, Germany, with a geometric resolution of 3.9m. Two subsets representing densely built-up urban areas, open residential, areas, and industrial areas were extracted from a flight line covering central Berlin from SSW to NNE. Figure 1: Test areas (Berlin-Schoeneberg, Central Berlin) in HyMap data (R-G-B 29-80-15). Secondly, a multispectral and panchromatic Quickbird frame covering the SE-quarters of the city was employed to compare the analysis opportunities connected with geometric VHR data. Digital vector information from the UEIS served for comparison purposes.