Identification of clear felled areas using SPOT P and Almaz-1 SAR data

Abstract

Almaz-1 S-band (3.125GHz) satellite SAR data from two subsequent years in combination with optical data have been analysed to determine the separability between recently clear felled areas and forested areas. The test area is situated in northern Sweden within the boreal conifer belt and consists of 32ha of recently clear felled areas and 464ha of forested areas. The basic assumption in this study is that, in an operational case, optical data before the clear felling, and radar data from both before and after the felling, will be available. The optical image is used for segmentation of the forest landscape into homogenous segments. For each segment statistical features based on the first and second order histogram are computed in the two radar images. The changed areas are identified by linear discriminant analysis with cross-validation. Classification based on segment mean values from the Almaz-1 images performed better than using a simple square neighbourhood mean. When using the segment mean values 61.4% of the clear felled areas were correctly classified, under the premise of equal errors of omission and commission. The textural features derived from the second order histogram did not improve the classification result.