Classification of Landsat 8 images in the Segura Hydrographic Demarcation

Abstract

This work presents a cartography of land uses in the Segura Hydrographic Demarcation obtained by classifying 2017 Landsat 8 images. The classification was carried out using two classifiers: Maximum Likelihood (ML) and Random Forest (RF). Training areas were obtained from historical high resolution imagery until 2016. Prior to classification, a cross validation analysis of the training areas was carried out to determine which of them may have undergone a change of use between 2016 and 2017. The results obtained with ML and RF, both with the original set of training areas and with the one obtained eliminating the problem, are compared to determine the best option. In the case of ML, the results improve after eliminating the changing training areas, from 77.7% to 81.4%; however, with RF this improvement is not so important, going from 84.1% to 85.1%. Therefore, it can be concluded that, with both methods, the classification is more exact when the modified training areas are used and, although the results obtained are quite acceptable for both ML and RF, the latter performs a more accurate classification.

Keywords

• Random Forest
• Máxima verosimilitud
• usos del suelo
• teledetección
• Landsat 8