Integration of data fusion techniques and Artificial Neural Networks for land cover mapping

This study concerns the mapping of land cover changes connected with the recent realisation of the reservoir of Bilancino (Northern Tuscany, Italy) has a fundamental step in the environmental monitoring and natural hazard prediction over the whole area. The realisation of this large reservoir (maximum capacity ca. 69x10^6 m^3), created for assuring regular water supply and flood control to the city of Firenze, induced important environmental modifications over large part of the Sieve river valley. To quantify and understand these consequences, and also to attempt short-term predictions of their evolution. AIMA ortophoto and Landsat TM data have been analysed to figure out the state of the environment before the impoundment of the reservoir while recent Landsat ETM and ASTER data showed the present conditions. In this context Landsat ETM images of different seasons (June 2000 and February 2001) have been acquired and processed: different data fusion techniques have been tested to obtain a multispectral data set of 15 m spatial resolution to be compared with a VNIR ASTR image of October 2001. Three image fusion techniques have been applied: Hue Saturation Value (HSV), Color Normalized (CN) and Principal Component Substitution (PCS). Artificial Neural Networks methods have been applied to assess the confidence and the consistency of the different data fusion techniques in detecting spectral information of water and its variation through seasons in different land cover data sets. Results show that remote sensing data, calibrated and validated with ground surveys, could represent a valuable tool able to provide models and indicator-based approaches the necessary parameters to understand, monitor and predict rapid environmental changes.