Mapping seasonal vegetation changes with multi-temporal radar segmentation

The wet/dry cycle particular to the Northern Australian climate creates a distinctively harsh environment for the vegetation species present. At a particular location a species adapted to immersion in water may flourish whilst the site is inundated but as the water dries up will be replaced by a species capable of surviving in the drier conditions. In this way particular species continually replace one another as the cycle progresses. For this highly complex environment single date image classification does not do justice to the constantly changing patchwork of vegetation as the drying cycle progresses. As moisture is a major determinant of the dielectric properties of most non-metallic materials, and vegetation in particular, any change with time may be mapped. A multi temporal approach to the situation is well suited to highlight not only the changes in vegetation community between seasons, but also that occurring between years. A region of the South Alligator River floodplain, within Kakadu National Park was surveyed to ascertain if these changes could be accurately mapped. Analysis of three multi temporal radar images obtained by Radarsat during the period of 1998-1999 were co-registered and subjected to GMRFM segmentation. The results of this segmentation were then used in a statistical clustering algorithm in order to regroup spatially separate segments into a smaller number of classes. Subsequent air photo interpretation and ground truth information have shown that the technique is invaluable in determining the timing and location of ecological classes in a highly variable climate.