Fine-scale mapping of a grassland from digitized aerial photography: an approach using image segmentation and discriminant analysis
Conventional methods of classification from remotely-sensed images seldom discriminate accurately among the land cover categories that are relevant in ecological applications. In the present study, we apply an image segmentation technique to a high-spatial-resolution (13.5cm), digitized, aerial, colour-infrared photograph of an annual grassland and subsequently identify land cover categories through field inspection and linear canonical discriminant analysis of the image. We show that per-segment statistics can be used to discriminate among four land cover categories bunch grasses, dense cover of annuals, sparse cover of annuals and bare ground while conventional per-pixel statistics produce low separabilities for the same categories. We also show that soil disturbances by pocket gophers (Thomomys bottae) can be identified and they are significantly concentrated in areas covered by bunch grasses at the time of image acquisition. We conclude that image segmentation and linear canonical discriminant analysis of high spatial resolution imagery provides an adequate tool for monitoring a grassland's patch dynamics at a scale and with a legend that are compatible with outputs of spatially-explicit ecological models.