Classification of Integrated Crop-Livestock Systems using PlanetScope time series

The new generation of orbital platforms has increased the opportunities for land cover classification using time series of satellite images in the last few years. In this study, we assessed the performance of high spatial and temporal resolution PlanetScope (PS) time series to map integrated crop-livestock systems (ICLS) and different land covers in the western region of São Paulo State, Brazil. To achieve this goal, 10-day and 15-day composite time series of the vegetation indices on both pixel and object-level were extracted from the PS images. The land cover classifications were performed using the Multi-Layer Perceptron (MLP) classifier, which achieved overall accuracies greater than 98.0%. The 10-day composite PS time series slightly outperformed the 15-day composite, returning overall accuracies of 99.1% and 98.6%, respectively. Although our method improved the discrimination of land parcels with ICLS, prediction maps returned misclassifications due to the hybrid unit of analysis, which will be improved in future works with the use of new deep learning algorithms that fully explore the temporal domain of the time series.