Spectral-Temporal Analysis by Response Surface applied to detect deforestation in the Brazilian Amazon

Abstract

Spectral-Temporal Analysis by Response Surface (STARS), which exploits both multispectral and multitemporal information using fitted response surfaces, was used to describe deforestation patterns in the Brazilian Amazon. The STARS was conducted upon a MODIS dataset formed by 21 selected cloud free images (eight days composition) acquired from August 2003 to August 2004. The Multi-Coefficient Image (MCI) resulted from the STARS was used as input attributes for the three tested classifiers: Instance Based K-nearest neighbor (IBK), Decision Tree (DT) and Neural Network (NN). The IBK classifier presented the highest accuracy (K=0.93) in detecting deforestation also indicating the deforestation period (early or late in the year). The results showed that the STARS is promising to describe spectral change patterns over time, allowing detection of the deforestation process which occurs in the Brazilian Amazon.