High Resolution Mapping of Vegetation Using low Spectral Resolution SPOTMap Imagery : a Study Case in Central Brazil.
Esteves Vieira, Dennis Junio1; Manzon Nunes, Gustavo1; Sanna Freire Silva, Thiago2
1Faculty of Forestry, Federal University of Mato Grosso, BRAZIL; 2Remote Sensing Division, Instituto Nacional de Pesquisas Espaciais (INPE), BRAZIL
Conservation Units (CUs) are one of the best strategies to protect the natural heritage, including conservation of environmental services and biotic and abiotic resources. The identification, delineation and classification of different land cover types on CUs and degraded areas have an important role in the establishment and management of these units, and provide essential information for studies on biodiversity, global biogeochemical cycles and the resulting impact of human activities. Remote sensing is commonly used in the delineation, mapping and monitoring of land cover in protected areas, but it often requires the use of data with high spatial resolution. The high costs associated with this type of imagery can make their use prohibitive, especially for developing countries. An alternative is the SPOTMap product, composed of images with a spatial resolution of 2.5m, and three natural color bands, orthorectified. These images are produced by a process called Supermode, patented by the French Space Agency (CNES), and can be acquired at a lower cost than full multispectral data. Though usually considered inferior for not preserving the spectral information, this product contains a high amount of spatial information, which can be exploited by Object Based Image Analysis (OBIA) methods. Through OBIA, the user can employ spectral, textural, geometric and topological attributes as part of the classification algorithm, approaching the process of human cognition through hierarchical and semantic relationships between classes. This study developed a method, based on OBIA, for the classification of vegetation types using high spatial resolution (HRG) SPOTMap data, using the Araguaia State Park (PEA) conservation unit, in the state of Mato Grosso, central Brazil as a case study. A test area was selected in the central region of the PEA, with approximately 100 ha, chosen as it represents all vegetation types occurring throughout the park. The high-resolution data was subjected to a 5X5 median filtering process using ENVI 4.7, to soften target edges and reduce pixel heterogeneity, improving the results of image segmentation. The multiresolution segmentation algorithm implemented in eCognition8.0 was used to generate image objects, producing two segmentation levels with the following parameters: color (0.8) form (0.2), compactness (0.3), smoothness (0.7), scale - level 1 (80) and scale - level 2 (100). The objects were then classified using a hierarchical network of classes, with three stages. The final decision algorithm used descriptors based on spectral values (brightness, mean and standard deviation) and object shape (edge and compactness) along with spatial context (proximity to other classes). The resulting algorithm allowed the delineation of the following vegetation types: Riparian Forest (forests along river channels), Gallery Forest (forests along narrow streams), Impucas (forest patches on raised terrain), Varjoes (savanna type vegetation on floodable land), Cerrado (dry savanna) and Monchoes (forest patches surrounded by sandy soil). Spatial context analysis was fundamental for the distinction of the Monchoes and Impuca classes, as both are compact, isolated forest fragments, but Monchoes are always surrounded by sand bodies, while Impucas have neighborhood relationships with Varjoes. Varjoes were the dominant vegetation type in the region, occupying approximately 60% of the entire study area. The overall classification accuracy of the present OBIA algorithm was of 90%, with a Kappa index of 0.88. These results show that OBIA methods can be applied to images with high spatial resolution but low spectral resolution, to produce accurate, detailed vegetation maps. This method can be considered a valid alternative for high resolution mapping in developing countries.