Delineation of burnt mountain forest areas by high-resolution satellite images

In this paper we present a remote sensing technique, based on very high spatial resolution Quickbird satellite data, aimed to map burnt forested areas located in alpine environment hit by winter fires occurred in Lombardia Region in the 2005 year. Quickbird satellite images have a spatial resolution of 2.5 m and are characterized by 4 spectral bands covering the regions of blue, green, red and near infrared. Burnt areas were automatically extracted by using an object oriented classification combined with a connectivity algorithm developed with the aim to join burnt isolates pixel with the main body of the area hit by fire. The proposed algorithm is based on the exploitation of a Gaussian function that produces a degree of membership to be burnt for every pixel not classified as burnt by means of the preliminary automatic classification. The membership function is established on the base of the spatial distance and it decrease according the full width at half maximum of the Gaussian function. The produced maps have been compared with the burnt area boundaries obtained by means of field survey based on GPS measurements; this allowed us to estimate the goodness of the proposed method. The comparison between the results produced by the connectivity algorithm and the reference measured in ground showed high degrees of accuracy with errors ranging from 3 to 20%.