EXTENDING LANDSCAPE CHARACTERIZATION CAPABILITIES OF FRAGSTATS (FOSS) TO ADVANCED REMOTE SENSING CLASSIFICATIONS
FRAGSTATS is FOSS for calculating landscape metrics, a type of spatial statistic designed to capture aspects of landscape structure patterns and quantify those patterns within a designated boundary. Traditionally, FRAGSTATS has been used to characterize hard-classified remote sensing images where each pixel is assigned a single land cover class value. However, recent advances in remote sensing have led to spectral unmixing techniques, whereby images can be classified with the percentages of multiple land cover classes present in each pixel. Spectral unmixing is particularly useful for moderate- and low-resolution imagery and is becoming widespread due to its ability to more accurately capture land cover mixtures. However, these unmixed data pose a problem for open source statistical packages, such as FRAGSTATS, which are programmed to run on hard-classified raster datasets with a single value for each pixel.
This research proposes a modification scheme for spectrally unmixed data using a moving window classification whereby the fractional vegetation measurements are preserved and utilized to better determine landscape structure patterns. In contrast to the current method of forcing user-defined threshold breaks onto unmixed data to hard-code them, this modification calculates a value for each pixel based on its fractional value as well as the fractional values of the surrounding neighbor pixels. In this manner, pixels to be included in the statistical analyses are determined based on the fractional percentages within the context of the landscape patch, thus preserving the benefits of the unmixing scheme. The adapted raster can then be analyzed by FRAGSTATS to calculate a host of landscape metrics.
Amy Frazier - University at Buffalo