Up to nowadays, satellite data have become increasingly available, thus offering a low cost or even free of charge unique tool, with a great potential for quantitative assessment of urban expansion and urban sprawl, as well as for monitoring of land use changes and soil consumption. This growing observational capacity has also highlighted the need for research efforts aimed at exploring the potential offered by data processing methods and algorithms, in order to exploit as much as possible this invaluable space-based data source.The work herein presented concerns an application study on the process of urban sprawl conducted with the use of satellite ASTER data. The selected test site is highly significant, being it a coastal zone (with the presence of sand and rocks) characterized by a fragmented ecosystem and small towns, with an increasing rate of urbanization and soil consumption. In order to produce synthetic maps of urban areas, ASTER images were classified using two automatic classifiers, Maximum Likelihood (MLC) and Support Vector Machines (SVMs) applied by changing setting parameters, with the aim to compare their respective performances in terms of robustness, speed and accuracy. All process steps have been developed integrating Geographical Information System and Remote Sensing, and adopting free and open source software.Results pointed out that the SVM classifier with RBF kernel was generally the best choice (with accuracy higher than 90%) among all the configurations compared, and the use of multiple bands globally improves classification. One of the critical elements found in this case study is given by the presence of sand and sand mixed with rocks. The use of different configurations for the SVMs, i.e. different kernels and values of the setting parameters, allowed us to calibrate the classifier also to cope with a specific need, as in our case, to achieve a reliable discrimination of sand from urban area.
Read full abstract