Decision Tree Classification of Land Use and Land Cover of the Tapajos National Forest Region, Brazilian Amazon

Decision Tree Classification of Land Use and Land Cover of the Tapajos National Forest Region, Brazilian Amazon

The Amazon rainforest covers an area of approximately 5 million km2 and is responsible for harbouring much of the planet biodiversity. Despite its importance this region suffers constantly with the deforestation process and is a source of study and the center of attention of the scientific community worldwide. The Tapajos National Forest is an important reference unit for conservation of tropical forest resources and is often the target of several studies. However, there are few studies that integrate different information from data collected remotely in the mapping land use and land cover in the Tapajos National Forest. In this context, the main objective of this study was evaluate the use of the technique of decision tree for map the land use and land cover in the Tapajos National Forest region, including classes of forest degradation and regeneration. For this, we used data mining technique, known as decision tree, and as input data for creating the decision tree we used different information obtained from an optical image of the sensor TM of the satellite Landsat 5 and this image is from the year 2009. Therefore, the data that were used in the decision tree were the six bands of Landsat 5 TM sensor of the year 2009, the three-fraction images (soil, shade and vegetation) obtained by the Linear Spectral Mixture Model, the three vegetation indices, Normalized Difference Vegetation Index, Normalized Water Index and Soil-Adjusted Vegetation Index. Through this work we concluded that the use of decision tree enabled the integration of the information obtained from the image Landsat 5 TM. Moreover, the classification of land cover and land use the Tapajos National Forest showed satisfactory results with Kappa index of 0.79. Approximately 81.2% of the pixels were classified correctly and approximately 18.8% of the pixels were classified incorrectly by the decision tree. The largest classification errors occurred between classes of pasture, regeneration, forest and degraded forest. The classes that showed the best results in classification were the classes water, cloud and cloud shadow.