Geoscience Reference
In-Depth Information
availability, and cloud cover. To minimize the potential effects of topographic shad-
owing in vegetation classifications we calculated the NDVI (Normalized Difference
Vegetation Index) to enhance vegetation signatures and provide some shadow cor-
rection in all images (Lillesand and Kiefer
2004
), and then incorporated these as
a layer into each multiband image. In doing this, the original spectral patterns
of the various features could be examined, with the NDVI values incorporated
into the spectral signature as another band. This allowed the NDVI values to be
compared to other spectral values for any pixel simultaneously and thus identify
some areas where shadowing may have lowered vegetation signatures within the
same class.
We processed images using the isocluster algorithm for unsupervised cluster
classification in Erdas Imagine 8.3. To ensure all spectral variability was captured,
analyses were repeated with increasing numbers of clusters until the number of
requested classes exceeded the number derived from cluster statistics. This resulted
in between 45 and 90 unsupervised classes which we identified based upon spec-
tral signatures and supportive field data available for some years. Supportive data
used to identify and merge classes included a series of ground photos (by Ross
Gilcrist of Weyerhaeuser Corporation) taken at the same locations yearly from
1980 to 1999, and an ARC/INFO coverage of tree planting data for a portion
of the study area supplied by the Weyerhaeuser Corporation. Aerial photos from
1996 and field work conducted in July 1997 were also used to aid in confirming
classifications.
We merged clusters into seven landcover vegetation and non-vegetation classes to
represent the major habitat components of deer and elk described by Witmer et al.
(
1985
) and the dominant vegetation and potential forage production (Washington
Department of Fish and Wildlife, unpublished data):
•
Non-Veg: Non or very low vegetation features including water, bare soil, rock,
soil/sparse-vegetation. Estimated forage biomass
=
0 kg/ha.
•
Soil-Veg: Bare soil areas having some degree of vegetation influence on spec-
tral signatures (or evident in ground photos) but still dominated by bare soil.
Estimated forage biomass
=
3 kg/ha.
•
Grass-Forb: Early successional class following site preparation for replanting.
Consists of mostly herbaceous vegetation although very early stands may have a
significant (@ 50%) bare soil component. Estimated forage biomass
=
11 kg/ha.
•
Shrub-Seedling: Follows planting of seedling trees and consists of deciduous
shrubs (and conifer species in later stages) overtopping Grass-Forb. Conifer and
shrubs are generally less than 1 m tall. Little bare soil influence on spectral
signatures. Estimated forage biomass
=
56 kg/ha.
•
Shrub-Seedling: Representing more advanced regeneration and co-dominance of
tall shrubs, conifer saplings, and herbaceous forage. Estimated forage biomass
=
l97 kg/ha.
•
Closed Sapling-Pole (CSP): Consists of forest stands approximately 12 years
of age or older and having canopy closure of 60% or greater. Estimated forage
biomass
=
6 kg/ha.
