Land Cover and Crop Type Classification using Machine Learning Techniques on Satellite Multispectral Data

As the global population grows in the near future, being expected to reach the number of 9.7 billion by 2050, agriculture will become more important than ever. Worldwide, in fact, the food production industry is already under severe pressures. On one hand the change in climatic patterns caused by global warming triggers mutations in ecosystems with the development of new plant diseases and stronger pests, with significant impacts on production. On the other the public opinion is day by day less in favor of unsustainable production methods as the food sector is responsible for intensive exploitation practices and about one third of global greenhouse gas emissions. The industry will have to inevitably reinvent itself in order to increase its productive capacity and be able to meet the expected growing demand, lowering costs and reducing environmental impacts. One of the most promising innovations, whose aim is to guide agriculture in this transition, is the application of satellite data to map and monitor every key aspect of crop growth maximizing yields, reducing water wastes, reducing fertilization pollution, lowering crop management costs and much more. On this path, the aim of this study is to examine the possibility of obtaining land cover and crop type information by means of multispectral data provided by Sentinel-2 mission of the European Copernicus program, which are fundamental knowledge in order to identify characteristics of a selected area allowing its management in a way completely tailored on its specifics. In order to do that two datasets on which to train and evaluate models, are built, one for the task of land cover classification, one for the task of crop type classification, by computing time series of spectral indices associated to points included in a third dataset from Eurostat called LUCAS used for the labels.