Seasonality of reflectance in the southern Pantanal of Brazil: spatio-temporal landscape segmentation using correlation length and lacunarity of NDVI images

Detection and quantification of changes in land use/land cover can be thwarted by natural or baseline spatio-temporal variation in reflectance characteristics. This situation is especially true for spatially heterogeneous landscapes that exhibit a strong seasonality in reflectance, such as the Pantanal of Brazil, the largest wetland habitat on the planet. Subject to extensive seasonal flooding by the Rio Paraguai and its tributaries, the PantanaI is an ecotonal landscape, a complex mosaic of shallow lakes, periodically inundated grasslands, and elevated forested patches, which together form a threatened haven of vertebrate biodiversity. The authors examine the efficacy of robust descriptors of spatial structure in segmenting a series of NDVI images of this complex region into appropriate landscapes units. The study area is approximately 65 km by 16 km in extent and transverses four landscapes in the southern Pantanal which each possess different hydrological influences: the lake district of Nhecolandia, the floodplains of the Rio Negro, the Rio Aquidauana, and the Rio Miranda. Scales of fluctuation (SOFs; correlation lengths) and lacunarity indices were measured in NDVI images formed from eight Landsat TM quarter-scenes over two inundation cycles (1985, 1989). Segmentation using correlation lengths in conjunction with NDVI intensity values was successful in separating landscapes into distinctive clusters; lacunarity metrics were too highly correlated with NDVI to enable clear segmentation. Regions of higher NDVI exhibited lower seasonal variation; seasonal variation of SOF showed a similar pattern. Scale of fluctuation metrics merit further consideration for image analysis and spatio-temporal scene modeling.