Mapping sealed surfaces for Brussels and surroundings using high and medium resolution satellite data

Sealed surfaces prevent water from infiltrating the soil and have a negative impact on environmental conditions. Mapping and monitoring the extent of sealed surfaces has therefore become important in many applications. A promising technique to obtain information about the distribution of sealed surfaces is to extract its occurrence directly from remotely sensed data and, more in particular, from satellite imagery. To exploit the advantages of both medium and high resolution data the use of sub-pixel classification is a good option. In this study artificial neural networks were used for sub-pixel classification of a Landsat ETM+ image for Brussels and surroundings. Training the networks was accomplished by extracting detailed information about land cover from a classified Ikonos image covering part of the Landsat data. Classification of the Ikonos image resulted in a kappa-index of 0.91. The kappa value was improved to almost 0.95 through post classification. Based on a random sample of Landsat pixels with known proportions for four major land-cover classes -sealed, vegetation, water, bare soil-, neural networks for sub-pixel classification were trained and validated. The average difference between real and estimated proportions of sealed surfaces for a Landsat pixel is about 9%.