Land cover/use analysis and modelling

The land cover/use databases are digital maps that provide the basic information for knowing and managing the territory at various scales, with particular reference to agriculture, food security, natural resources and environment protection. This chapter describes the production process of land cover/use databases by photo-interpretation on the screen or by semiautomatic classification of remote sensed images. After a review of supervised and non-supervised approaches, main supervised parametric and non-parametric machine learning classifiers are applied to Sentinel 1 and 2 satellite images, repeating the classification for different training and test sets. The land use assessed on a grid of geo-referenced points in the north of Tuscany Region is used as ground truth. The results are discussed focusing on Cohen’s Kappa, the overall accuracy and the dispersion of the distribution of the overall accuracy with different training and test sets. The quality control and the validation of land cover/use databases are addressed, taking into consideration both positional and thematic accuracy and alternative sampling strategies for collecting reference data, including adaptive sampling techniques. Since land use change is extremely important for environmental purposes, pros and cons of approaches for developing and assessing the accuracy of land cover change databases are discussed.