High Spatial Resolution Digital Imagery
57
5.1.3
Minimum Mapping Unit
Very high-resolution imagery typically provides far more detail than can be
efficiently used for landscape mapping. For example, high quality, large-scale
aerial photography will provide adequate clarity and detail to accurately identify
surface features that are no larger than a few square feet. It is unlikely that a
photo interpreter would be required to delineate image features only a few feet in
diameter. Instead, the analyst will work within the limits of a predefined Mini-
mum Mapping Unit (MMU). This two-dimensional area represents the minimum
size of the polygon delineated within the aerial photo. Land use/land cover clas-
sifications commonly use a MMU of 1 acre to 1 ha for large scale maps and
much larger MMU sizes, such as 10 to 100 ha for very small-scale land cover
maps. The MMU used in this application is very small as compared to most land
cover classification schemes. All reference wetland class maps were subjected to
a nine-pixel (~ 6.8 m2) minimum mapping unit filtering routine. This two-step
routine first uses a simple raster GIS technique to find and remove all raster
polygons that are less than or equal to nine pixels. A raster polygon is a contigu-
ous group of pixels that have the same land cover class designation. The thematic
image is then "sieved" to remove those raster polygons below the nine-pixel
minimum threshold.
The "holes" (removed raster polygons) are then filled by iteratively passing a
33 majority filter over the image until all of the deleted pixels are reassigned to
a new class value.
There were two reasons for applying the MMU filter:
The nine-pixel "sieve" effectively removed the majority of the thematic
noise in the output class maps. With high spatial resolution imagery,
single pixels and small groups of adjacent pixels make up raster poly-
gons that are very difficult to ground truth. The very fine level of detail
provided by these maps is typically not required and may actually de-
grade the interpretability and the overall accuracy of the classification.
Output products include ArcView shapefiles. Preliminary examination of
shapefiles created from the full resolution class maps suggested that the
interpretability of these vector files would be very difficult. File size was
also a problem with the full resolution maps. The filtered class maps
stored the data with greatly reduced file sizes and appeared to offer more
easily interpreted data.
Figure 35 displays the result of the MMU filter on the Knapps Narrows class
map. Employing an extremely small MMU for this project maintained the detail
provided by the classified CAMIS images.
Previous Page
