Data mining applied to temporal dynamics of deforestation pattern: a study case in Southern Amazon forest, Brazil

The Amazon forest is one of the most prominent tropical rainforest worldwide. Providing several benefits, measures to its protection have been included in governmental decision making. In this context, the Brazilian initiative known as PRODES was implemented in 1988. Coordinated by the National Institute of Space Research, INPE, this project monitors annually deforestation in Brazilian Amazon. Even though, it is not enough to avoid deforestation and more analysis are required. In such manner, landscape metrics are commonly used to support analysis of deforestation dynamics and spatial patterns. It is just possible, considering that a real landscape reflects certain spatial patterns and structures. Nonetheless, taking into account Amazon extension, classifying landscape manually is considered a very time consuming task. In this manner, the aim of this study is to automate landscape classification, based on an already visually classified area in Southern Amazon forest. After that, we applied the decision tree to 1985 and 2015 data. Although data mining techniques were used, the final classification was not satisfactory for all the applications. Thus, we propose as further researches alternatives to overcome these issues and to validate the process. Finally, a discussion about the algorithm is also held as well as local temporal dynamics of deforestation