SOIL SEALING AND LAND USE CHANGE DETECTION APPLYING GEOGRAPHIC OBJECT BASED IMAGE ANALYSIS (GEOBIA) TECHNIQUE

On the other hand, for the compactness criterion, the suitable weight was 0.3 in most of the classes except for the classes Olives, Vineyards, Mixed Olives-Vineyards and urbanization for which the weight 0.5 was more suitable. Ten land use and land cover classes were recognized during the classification progression which are urban, water bodies, tree lines, woodland, pasture, bare soil, agriculture fields, olives, vineyards and mixed vine-olives. Different features and values were used for the recognition of classes during the classification process. To generate the final LULC map, the classified tiles were exported to a GIS environment in a polygon shapfile form and then went through mosaicing process to form one polygon layer with all classes. The calculated overall map accuracy is 77% with a kappa value of 0.73 which are both within ranges of fair accuracy. The producer's accuracy states how well the map producer recognized a land cover type on the map from the remote sensing imagery data. Results show that the highest producer's accuracy was for pasture class (94%) while the lowest was for the vineyards class (44%). Comparing the old and the improved (GEOBIA) maps of LULC shows that, regarding the agricultural area, 50% of the detected change using the original data was misclassified compared with the improved classification of aerial photographs. The results revealed that the urban area was underestimated in the old LULC map of 1954. This leads to an important finding of this research that modeling results are greatly correlated to the historical input data a