Caveats include: Maps in this series are naturally correlated since all came from one map. No warranty or claim is made of the utility of this map for any particular purpose, this is considered to be a research dataset. The forest fragmentation map portrays relative fragmentation, at one scale, considering only forest and non-forest cover types.
Land cover data (MRLC) was obtained from EROS Data Center in binary format.
The image was subdivided into sixteen (16) overlapping rectangles using an in-house software tool named SPLITTER.C. The rectangles overlapped to avoid artifacts near image boundaries during the spatial filtering operations.
Each of the rectangles was then processed via spatial filtering to estimate the index as described below. The spatial filtering program is an in-house software tool named SPATCONV.C (Riitters et al. 1997)
After spatial filtering, a map of index values was constructed by reassembling the 16 rectangles into a single image (via an in-house software tool named LUMPER.C, which removed the overlapping parts of rectangles).
The "header file" information for the derived map is as follows. This information is needed for some image import filters. Ulxmap and Ulymap refer to the center of the upper-left pixel. nrows 13240 ncols 13265 nbands 1 nbits 8 layout bsq skipbytes 0 ulxmap 1154670.000000 ulymap 2064600.000000 xdim 30.000000 ydim 30.000000
Spatial filtering proceeded as follows.
Pixels were re-classified into forest and non-forest classes by using the following scheme:
Original MRLC class Forest/non-forest class 11: open water missing 12: perennial ice/snow missing 21: low intensity developed non-forest 22: high intensity residential non-forest 23: high intensity commercial/industrial non-forest 31: bare rock/sand/clay non-forest 32: quarries/strip mines/gravel pits non-forest 33: transitional barren non-forest 41: deciduous forest forest 42: evergreen forest forest 43: mixed forest forest 51: deciduous shrubland forest 52: evergreen shrubland forest 53: mixed shrubland forest 61: planted/cultivated (orchards, vineyards, groves) non-forest 71: grassland/herbaceous non-forest 81: hay/pasture non-forest 82: row crops non-forest 83: small grains non-forest 84: bare soil non-forest 85: other grass (lawns, city parks, golf courses) non-forest 91: woody wetland forest 92: emergent herbaceous wetland non-forest
A 9x9 pixel quadrat was centered on each pixel of the original land cover map. An attribute adjacency table was then tabulated, considering only pixel pairs in cardinal directions (i.e., four neighbors per pixel), and counting each pixel pair once. The fragmentation index was then computed from the attribute adjacency table as the frequency of forest-forest pairs, divided by the sum of frequencies of forest-forest plus forest-nonforest pairs. This yields a value of 0 when no forest pixels are adjacent to another forest pixel, and a value of 1 when all forest pixels are adjacent in all cardinal directions to other forest pixels. Thus, the range is from zero (high forest fragmentation)to one (low forest fragmentation). The index is not independent of the amount of forest present in the window. If the center pixel was "missing" in the land cover map, then the index was assigned a "missing" value. Note that the index is defined for all other land cover types.
The calculated values were discretized to the range [1,255] and stored at 30-meter spatial resolution. Thus, a pixel value in the new map represents the index for the surrounding 9x9 pixel window in the original land cover map.
The transformation used to discretize the values was: D = ( C * 254 ) + 1 where D = discretized value in range [1,255] C = calculated value in range [0,1] If needed, the original calculated values can be approximated by applying the backtransformation: C = ( D - 1 ) / 254 The backtransformed values will be in steps of size approximately 0.00394 as a result of the discretization process.