Extracti G Buildi Gs I the City of Lisbo Usi G Quickbird Images a D Lidar Data

The methodology is based on semi-automatic extraction of features, using information from a QuickBird image and LIDAR data. The QuickBird image is suitable to delineate the different elements on the surface based on their spectral characteristics. However, this task is much more successful and richer when altimetric information is also used in the extraction process. Thus, with the introduction of LIDAR data, we intend to study the effect of the contribution of information on the height of urban elements, when classifying orbital data. The thematic information under analysis is the land cover class Buildings. Two maps are produced and compared. One map is obtained by a classification based only on spectral data while the other map is obtained using spectral and altimetric data. The same methodology is applied in both scenarios, using the same training features. To evaluate the quality of the two maps, a comparison with a cartographic layer is carried out. This reference layer is obtained by visual analysis of the QuickBird image and is used to calculate the Overall Accuracy of the maps obtained with the different data sets. * Corresponding author. 1. I TRODUCTIO Remote Sensing is a science, a technique and a technology at the service of Earth observation and, in particular, cities. The cities represented in satellite images are objects in the physical sense of the term, characterized by a wide range of spectral responses. These spectral responses only make sense when coupled with thematic content, interpreted based on the shape and morphology of the different elements that are include in the urban environment. The high rate of changes in cities requires the existence of matching geographic information in order to allow a proper land monitoring and planning. The project GEOSAT explores the potential of high spatial resolution satellite images for extracting spatial information and update existing cartography. The goal is to provide updated spatial data to support the different land-related activities of municipalities. In this context, methods for extracting urban elements relevant to the municipal planning process are tested in the city of Lisbon, Portugal. The city is characterized by a variety of complex objects (surfaces) which are associated with a wide spectral range. Furthermore, the city is also characterized by volume. Consequently, the shadow projected by the buildings multiply the radiometric effects and complicates the detection of elements. One can conclude that, in the urban realm, the spectral recognition of isolated features is not enough. On one hand, other types of data such as LiDAR (Light Detection and Ranging) must be included to separate impervious surfaces having very high reflectance (like harvested urban agriculture areas, for example) from tall buildings that have similar levels of reflectance. One the other hand, urban remote sensing must ally the spectral recognition of features with the spatial recognition (spatial organization of spectral signatures) (Bähr, 2001). These approaches have been grouped under the term Geographic Object-Based Image Analysis (Hay and Castilla, 2008). The work presented takes place in the context of project GEOSAT, for which there exist already several publications. Freire et al. (2008) tested the extraction of geographic objects in two study areas located in Lisbon, using a QuickBird image. The authors concluded that the different land cover classes and urban morphology influence the replicability of the mapping processes in distinct urban contexts. Santos et al. (2009) studied the quality of extraction of red tile roofs. The methodology began with the building extraction, followed by geometric generalization and subsequent accuracy assessment. Metrics that enabled comparison of vector data sets were tested in order to assess the compliance of a spatial data set obtained by semiautomated methods, with another set obtained by visual analysis, regarding the thematic and geometric quality. Freire at al. (2009) employed different GEOBIA approaches to extract agricultural use from a pan-sharpened QuickBird image, and results were compared. The study illustrated the potential and limitations of using VHR imagery and feature extraction methods to detect and monitor informal agriculture within a very heterogeneous urban fabric. 2. DATASET A D STUDY AREA The dataset explored in this paper includes spectral data, acquired by satellite, and altimetric data. The spectral data is a The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXVIII-4/C7 QuickBird image acquired in April 14, 2005. The image has a spatial resolution of 2.4 m in the multi-spectral mode (visible and near-infrared bands), a pixel size of 0.6 m in the panchromatic mode, and a radiometric resolution of 11 bits. The image used in this study has an off-Nadir angle of 12.2o. The altimetric data is composed by two sets. One set is derived from a LiDAR point cloud (Light Detection And Ranging), and the other is derived from cartography. From a LiDAR flight done in 2006, a surface image was produced based on the second return, with 1 m resolution. From 1:1000 scale cartography of 1998, a set of elevation mass points and contours were retrieved. The study area is located in the oriental part of the city of Lisbon and occupies 64 ha (800 m X 800 m) (Figure 1). The area is characterized by a diverse land cover that includes herbaceous vegetation, lawns, trees and agricultural plots, bare soil, single and multi-family housing, a school, industrial properties, and roads and rail networks. Figure 1. Study area in the city of Lisbon