Methods to Improve the Accuracy of Remote Sensing Data Classification, Change Detection,and Urban Growth Modeling

Accuracy assessment as an indisputable complementary of classification process validates land cover and land use (LCLU) change detection analysis. The simultaneous selection of an appropriate classification method and optimal bands, as well as, the inclusion of the vegetation index (e.g. NDVI) to the composites with poor spectral separability are fundamental factors improving classification accuracy assessment of remotely sensed data. The combination of different multi-spectral bands from Landsat ETM+ data in order to utilize the most effective composites, under performance of per-pixel classification techniques of ML and/or SVM, may not only reduce data redundancy, but also substantially discriminate each of the land-cover categories from all others. In this way, we could reduce the number of spectral bands to minimal band subsets and conversely augment overall kappa value to 95%. It is worth mentioning that classification accuracy assessment in this work was not confined to only kappa coefficient of agreement, or rather it has gone a step further by estimating overall tau coefficient based on equal probabilities of group membership. Basically, a variety of several statistical measures of accuracy assessment of remotely sensed data could provide this facility to analyze the classification quality from different perspectives. For the purpose of having carried out the second step of this study, we analyzed spatial and temporal dynamics of land use change using multi-temporal Landsat imagery. In addition to pre-classification change detection techniques (e.g. principal component transformation (PCT) and band combination (BC)), we applied post-classification comparison through the integration of remotely sensed data and GIS. On the strength of obtained results, it was realized that a lack of the efficient urban planning and not enough sustainable management of forest ecosystem have pushed Pisa Province towards the expansion of built-up structure (with urban sprawl index of about 0.3 from 2000 to 2006) and also deforestation (with annual rate of 1% from 1972 to 2006). In reality, significant tendency of urban development in the present century has made a serious issue in territorial and urban planning discussions of European cities. Mutual dependence between urban expansion and natural resource change has imposed a heavy pressure on air quality, water bodies, and landscape of Europe. With respect to the importance of modeling to determine dynamics of urban growth, the last step of this study was allocated to evaluate the performance of SLEUTH model (formerly, the Clarke Cellular Automaton Urban Growth Model) for the first time over historical Italian cities located in Pisa Province and its surroundings. The capability of SLEUTH to simulate land use conversion and urban growth, based on historic growth patterns, has allowed us to predict the cumulative trends of the area towards urban development over the coming decades. It is postulated that there is an essential demand for sound land-use planning to handle the rapid changes before urban sprawl engulfs the whole landscape of the area.