Improving the ANN classification accuracy of landsat data through spectral indices and linear transformations (PCA and TCT) Aimed at LU/LC Monitoring of a River Basin

In this paper an efficient Artificial Neural Networks (ANN) classification method based on LANDSAT satellite data is proposed, studying the Cervaro river basin area (Foggia, Italy). LANDSAT imagery acquisition dates of 1984, 2003, 2009 and 2011 were selected to produce Land Use/Land Cover (LULC) maps to cover a time trend of 28 years. Land cover categories were chosen with the aim of characterizing land use according to the level of surface imperviousness. Nine synthetic bands from the PC, Tasseled Cap (TC), Brightness Temperature (BT) and vegetation indices (Leaf area Index LAI and the Modified Soil Adjusted Vegetation Index MSAVI) were identified as the most effective for the classification procedure. The advantages in using the ANN approach were confirmed without requiring a priori knowledge on the distribution model of input data. The results quantify land cover change patterns in the river basin area under study and demonstrate the potential of multitemporal LANDSAT data to provide an accurate and cost effective means to map and analyze land cover changes over time that can be used as input for subsequent hydrological and planning analysis.