Mapping Land Cover on Barrier Islands Using Airborne Videography

By

D. Ann Rasberry, Paula G. Becker, Rebecca A. Gast, and Timothy A. Palmer

Following the methods described by Slaymaker et al. (1996) the Mid-Atlantic GAP Project is using airborne videography to facilitate the land cover classification of Landsat Thematic Mapper (TM) multi-temporal hyperclustered imagery for Maryland, Delaware, and New Jersey. Video was flown throughout the project area in May-June, and October-November, 1996 as well as in early September in southern New Jersey to capture specific freshwater wetlands while still vegetated. This poster describes the process of creating the land cover classification for the portion of Assateague/Chincoteague Maryland/Virginia found in TM scene 1434.

The GPS points from the flightline were displayed over the hypercluster. The video was then viewed in the context of the imagery to identify sites for field visits. These sites were used to develop a key for interpretation of the remainder of the video. As many classes (species, communities, alliances, etc.) were selected as possible. Following are examples of the video frames acquired during the flights over the barrier islands. These are shown in pairs; the one on the left is the wide angle frame and represents approximately 0.5 kilometer on the ground. The right frame is the 12x zoom through the middle of the wide angle and is approximately 30 meters on the ground.

FOREST

Pinus taeda/Quercus (nigra, falcata) Forest Alliance. This is an upland forest. Note the difference from a woodland: forests have a more closed canopy and are usually taller, although on the barrier islands they tend to be stunted.

WOODLAND

Pinus taeda/Hudsonia tomentosa Woodland. This is a good example, and is distinguished from a forest by the greater degree of canopy opening, in this case, with patches of bare sand and beach heather.

SHRUB

Myrica cerifera/Hydrocotyle sp. Shrubland. This alliance can form dense thickets. This shot shows the bands of communities common on these barrier islands. This alliance is the broad dark band in the center of the wide frame. The zoom shows the density of the thicket. The dead branches are Baccharis halimifolia, a deciduous shrub.

DWARF SHRUBLAND

Hudsonia tomentosa Dwarf Shrubland. This alliance tends to be small and patchy except in a few places on these islands. This shot shows it nestled in another community, probably Myrica cerifera/Baccharis halimifolia. A patch of this size would fall out in our current mapping method. The zoom shows the typical dune top, almost purely H. tomentosa and associated grasses.

HERBACEOUS

Ammophila breviligulata Herbaceous Alliance showing abrupt demarcation at dune lines.

SPARTINA MIX

High and low marsh. This is primarily Spartina atlerniflora (nearest the standing water and the lowest areas), with patches of S. patens and Distichlis spicata on the higher or drier areas. This time of year (spring) there is very little vegetative difference in these grasses.

This is a portion of Landsat TM scene 1434, 12 band hypercluster with the video flight line flown 20 May 1996. Field visits were made and the video prints were marked to indicate the species located on the print. A field form was developed to collect additional information about the site and if ground based photographs were taken this was noted on the form.

Viewing the video along with the flight line displayed over the hypercluster allowed identification of points where the land cover appeared similar to that of sites previously established from field visits. The spectral values for a 5 by 5 pixel areal around each point were recorded for each map class. Data were entered into a Corel QuattroPro spreadsheet and manipulated to determine minimum, maximum, majority (mode), and the number of different values (diversity) for each 25 pixel set. Spectral values were weighted based on the percent of each value in each map class.

Inference rules were written based upon spectral classes and ancillary data. The most useful ancillary daya for these barrier islands was the U.S. Fish and Wildlife Service's National Wetlands Inventory. This layer allowed for separation of map classes based upon upland and lowland (soil) types.

The final map is a result of pixel by pixel classification of the image using the inference rules. A second iteration does a comparison of each pixel to its neighbors to make sure that the likelihood that the pixel was correctly assigned to its map class is maximized.

DISCUSSION: The barrier islands of Assateague and Chincoteague proved to be a challenge to map using the GAP methodology. We found the video protocols to generally work as described by Slaymaker et al. (1996), but we had to modify them slightly; e.g., some video species identification was done on the wide angle prints because the zoom frequently fell completely within one small patch alliance which was very different from those around it. Additionally, rather than topology and/or elevation to assist with class discrimination, in this area the National Wetlands Inventory data proved to be invaluable in separating wet and dry vegetation which tended to be spectrally indistinct on the hypercluster. Of greater concern was the 30 meter pixel size of the TM imagery. These islands are rarely greater than a mile across, yet the vegetation and hydrology changes rapidly. The Mid-Atlantic GAP Project is mapping at a 2.5 hectare minimum mapping unit for the project but treated this area as a special case and did a per pixel classification. Even at this resolution, while distinct alliances could readily be identified from the video, these had to be aggregated into more general map classes. We have called these map classes Land Cover Types, because the term "cover type" nicely describes the clumping of the more finely separated alliances. Several of the alliances which occur along these barrier islands were not vegetated at either of the times we flew video over the area. As a result of several field visits during the summer we were able to change the map class from "Scum" to include these areas in the cover type "Tidally-influenced tall grassland." Similarly, the four forest alliances which occur were not separated because the resolution of the hypercluster along with the rapid changes in topology and hydrology meant that the NWI data set was inadequate in this area. Considering we used NWI data created from photography flown in 1981 in conjunction with video flown in 1996 in order to classify TM imagery from 1991 we believe this classification of these barrier islands to have turned out very well.

REFERENCES:

NBS/NPS Vegetation Mapping Program. 1995. Vegetation classification of Assateague National Seashore. Unpubl. Rep. The Nature Conservancy. Boston. 92 pp.

Slaymaker, D.M., K.M.L. Jones, C.R. Griffin, and J.T. Finn. 1996. Mapping deciduous forests in southern New England using aerial videography and hyperclustered multi-temporal Landsat TM imagery. Pages 87-101 In: J.M. Scott, T.H. Tear, and F. Davis, eds., Gap analysis: A landscape approach to biodiveristy planning. American Society for Photogrammetry and Remote Sensing. Bethesda, Maryland.

Sneddon, L. and M.A. Berdine. 1995. A classification and description of terrestrial community alliances in Maryland: First approximation. Unpubl. Rep. The Nature Conservancy, Boston. 86 pp.

The classification used by the Mid-Atlantic GAP Project follows the National Vegetation Classification Scheme.

Copyright 1997 Maryland Department of Natural Resources. All rights reserved. Mid-Atl. Gap Publ. 97E01