| Summary: | The land-use and land-cover map is fundamental for environmental planning, management, and conservation. Remote sensing data is the information basis for the development of such maps, providing speed and cost-effectiveness. In the Amazon region, the constant presence of cloud cover and smoke makes the use of radar images more adequate for operating independently of these barriers. The present research aims to propose a methodology to classify land use and coverage from PALSAR-2/ALOS-2 multitemporal images in the region of Bananal island. The Bananal island is the second largest fluvial island in the world with approximately 750 km long and 100 km wide. The study area is in the confluence between the Javaés and Araguaia rivers, located in the middle Araguaia river basin. The methodology adopted considers the following steps: acquisition of four multitemporal images Palsar 2/ALOS 2 at processing level 1.5 (L band with resolution of 6.25 meters); (b) Probability Density Component Analysis (PDCA); (c) noise reduction using the Minimum Noise Fraction (MNF) transformation; and (d) classification by the Support Vector Machine method considering the temporal signatures. The ACDP and MNF combination allowed the formulation of different features that describe the different classes present in the image. Comparatively, the images were processed using the Gamma adaptive filter and the SVM classifier. The Kappa coefficient results were 0.62 on the SVM/CDP-MNF images and 0.57 on the SVM / Gamma images. The main factor for the decrease of the Kappa coefficient was the confusion between the areas of Savana Rupestre and anthropic use (pasture) in the areas adjacent to the Bananal island
|
|---|
