An Object Oriented Approach for the Discrimination of Forest Areas under the Criteria of Forest Legislation in Greece Using Very High Resolution Data

The accurate discrimination of forest from natural non-forest areas in Greece presents great interest, since nowadays there is an ongoing effort to develop a Forest Cadastre system. We evaluated the possibility to extract forest areas according to the legislation criteria, in a mountainous area in the Northern-central part of Greece, using an object oriented approach and a very high resolution image. The 240 hectares study area is occupied from deciduous and evergreen forest species, shrublands and grasslands. The segments were classified using two different algorithms, namely Nearest Neighbor, built-in the software eCognition and a logistic regression approach. Furthermore we evaluated for the same task the usefulness of a fused image with the Gram-Schmidt method, classified after the segmentation with the NN algorithm. After the classification of the first level we proceed with a classification based segmentation approach resulting to a second upper level. The later was classified using class and hierarchy related features of the software to quantify the criteria of the Forest law. Logistic regression classification of the original multispectral image proved to be the best method in terms of absolute accuracy reaching around 85% but the comparison of the accuracy results based on the Z statistic indicated that the difference in the results between the three approaches was non-significant. Overall the object oriented approach followed in this work, seems to be promising in order to discriminate in a more operational manner and with decreased subjectivity the extent of the forest areas in Greece.