Detecção de mudanças na costa de manguezais da Amazônia a partir da classificação de imagens multisensores orientada a objetos

Abstract

Mangroves presents great importance to the ecological balance, and a nursery conducive to the development of various animals and plants. In recent years, degradation of mangroves has been occurring more frequently due to the plundering of their natural resources, land planning and poorly planned tourist activities. By remote sensors can map large areas of the area more quickly and efficiently. The objective is to map the distribution of mangrove areas to the east of the Amazon River into the Bay of San Marcos in 1996 and 2008 from remote sensing data. The mapping, change detection and quantification was performed by ALOS / PALSAR, JERS-1, SRTM and Landsat 5 TM. In order to classify the images, we used the software Definiens Ecognition 8, which uses the logic of object-oriented classification. In the classification of the mangrove was an elaborate process tree that stores all the elements or rules (segmentation, algorithms, classes and attributes) needed to obtain the final classification. The result of the quantification of the mangrove was 6705,05 km ² (1996) and 7423,60 km ² (2008) which shows a net increase in mangrove area of 718,55 km ². The change detection map allowed an overall increase of 1931,04 km ², a total erosion of 1212,49 km ², remaining an area of 5492,56 km ² of mangrove unchanged. To statistically validate the results, we elaborated two confusion matrices containing the rights and wrongs of the classification. The error matrix for validation of the classification of classes mangrove swamp, upland, water mass, secondary vegetation, fields and lakes showed an overall accuracy rate = 96.279%, Kappa = 90.572% and 92.558% = index Tau, which showed the classification efficiency of mangroves in relation to other classes used in processing. The error matrix for validation of classification and Non-Change Change of mangrove area showed high accuracy Global = 83.33%, Kappa = 66.10% and 66.66% = index Tau. Therefore, we conclude that the method of object-oriented classification logic is excellent for mapping mangroves and very good for the detection of changes in tropical coastal areas. Regarding the expansion of mangrove areas, it is observed only in the Amazon region, as opposed to what is observed in other large systems of mangroves, such as the Gulf of Papua New Guinea and the Sundarbans in Bangladesh and India. The results were used to compose a mosaic of regional and global mapping of mangrove and ratify the large expanse of mangrove forests in Amazonian Brazil as one of the best preserved of the planet.