A semi-operational approach for land cover mapping in the Mediterranean

摘要

Classification of remotely sensed imagery is considered an appropriate technique to obtain land cover information. Recent improvements in the spatial resolution of sensors make possible the extraction of detailed thematic maps. However, high spatial resolution may reduce the accuracy of per-pixel classification methods, as the spectral within class variability increases. Similar spectral characteristics of Mediterranean land cover types and high land fragmentation make the discrimination of land cover categories difficult. An object-oriented classification approach is thought to be a solution.The aim of this study was to develop a methodology for the semi-operational land cover mapping in the Mediterranean to support policy-making and planning, carried out in the framework of the Geoland project (SIP3-CT-2003-502871), in the context of "Global Monitoring for Environment and Security" (GMES) initiative.A two-step approach was developed on the basis of object-oriented classification to exploit the potential of the different available data. In the first step only QuickBird multi-spectral data were used to extract land cover classes at coarse level, while in the second step the categorization of the vegetated areas was refined using both multi-spectral and panchromatic images, as well as DEM data.The developed methodology was implemented to map a heterogeneous, highly fragmented area in Crete (Greece). The analysis of the confusion matrix revealed some patterns of common misclassifications, mainly due to the high spatial resolution of the imagery. The overall accuracy shows an acceptable performance of the developed object-oriented approach, given the heterogeneity of the environment and the complex classification needs.