Land Cover Classification in Amazon using Alos Palsar Full Polarimetric Data Luciano Vieira Dutra Graziela Balda Scofield Sumaia Resegue Aboud Neta Rogério Galante Negri Corina da Costa Freitas Daniel Andrade Instituto Nacional de Pesquisas Espaciais - INPE Caixa Postal 515 - 12245-970 - São José dos Campos - SP, Brasil {dutra, graziela, sumaia, rogerio, corina, andrade}@dpi.inpe.br.br Abstract. The ALOS PALSAR sensor can provide full polarimetric SAR data (HH, HV and VV) but the full polarimetric mode is only available experimentally. Here, several supervised classifiers have been studied to determine how much the use of full polarized (HH, VV and HV, no phase information) PALSAR data information can improve, or not, the overall classification accuracy in comparison with the standard products, which, for PALSAR instrument, is the HH (like JERS-1) or the dual polarization product HH-HV. The study area, Tapajós National Forest at the south of Santarém City, in the Brazilian Amazon, Pará State, has being object of intensive scientific observation for more than 15 years. Several types of supervised classifiers are tested for having, as much as possible, an assessment rather independent of the classifier type. Initial results indicate that, no phase considered, the dual polarization product HH-HV is the better channels combination for mapping the set of tropical classes composed by the primary forest, secondary forest, bare soil, agriculture and degraded forest. Also, it was observed that one year regeneration areas are not discriminated in any PALSAR combination, which indicates the utility of maintaining the complementary use of optical images when possible, because, in the optical combination, the ‘one year regeneration’ class still shows different from secondary forest. Region based classification, particularly one developed to take in account as much as possible the radar statistical behavior, generally presented better performance. Palavras-chave: Alos PALSAR, polarimetry, radar land cover, Tapajós, SAR. 1. Introduction The importance of Microwave Remote Sensing for tropical forest monitoring is well known. The new generation of orbital SAR plataforms is capable of delivering several types of polarimetric products, but the full polarimetric mode is of restricted use because of operational limitations. This investigation is focused on how the use of full polarimetric PALSAR data information can improve, or not, the overall classification accuracy in comparison with the standard products. Particular attention is given to the discrimination capability of perceiving land cover alterations of great importance on tropical environment monitoring, like recent deforestation, degraded forest and regeneration, as a function of the channels set used. Several types of supervised classifiers are tested for having, as much as possible, an assessment rather independent of the classifier type. The classification results improvements are statistically tested for significance. The study area, Tapajós National Forest at the south of Santarém City, in the Brazilian Amazon, Pará State, has being object of intensive scientific observation for more than 15 years. 2. Materials And Methods 2.1. Materials The Tapajós National Forest (FLONA), Figure 1, is located at the south of Santarém City, in the Brazilian Amazon, Pará State. A polarimetric (PLR) scene (level 1.5), from October 21, 2006, was obtained via the User Remote Sensing Access (URSA) from the Alaska Satellite Facility (ASF). Field work was conducted to collect ground data on October 2005, during a L 7259