Machine Learning for Quantification of Land Transitions in Italy between 2000 and 2018 and Prediction for 2050

Italy is one of the main Mediterranean countries with its landscapes undergoing chronological changes because of intense urbanization, socioeconomic, and anthropogenic activities. In this study, an attempt has been made to analyze land use/land cover changes (LULCC) in Italy between the years 2000 and 2018 using Copernicus CORINE Land Cover (CLC) datasets of corresponding years. In addition, the Cellular Automata Markov model has been employed to predict the LULC of Italy for the year 2050. The CLC datasets were reclassified into 8 major land cover/use classes: built-up, agriculture, forests, natural vegetation, bare land, water bodies, ice/snow, and wet-land. Resultant LULCC detection maps indicate a significant increase in bare land by 36.4 % in the period 2000-2018 and, according to the prediction, a further increase by 46.7 % in the period 2018-2050 will be experienced. Built-up increased by 14.7 % between 2000-2018 and 19.8 % between 2018-2050. Significant decreases in natural vegetation and ice/snow in the period 2000-2018 have been quantified in 18.9 % and 6.5 %, respectively. According to the present prediction, those classes will further decrease in the period 2018-2050 to 24.5 % and 7.1 %, respectively. To validate the quality of the present research, an accuracy assessment of the reclassified maps of the years 2000 and 2018 has been carried out. The accuracy assessment involves comparing classified maps with the ground truth image through 576 randomly selected samples. Reclassified maps of 2000 and 2018 showed an overall accuracy of 89.9 % and 89.6 % with a Kappa index of 0.884 and 0.880, respectively. The findings in this study can be critical for warning policymakers, planners, and other associated development workers to adhere to the best suitable land-use management option for the Italian territory.