Geoscience Reference
In-Depth Information
10.3.3 Pre-Classification Image Processing
Landsat TM data were orthorectified to a planar projection (UTM) by the distributor
prior to download. Using ENVI (Research Systems Inc.) image processing soft-
ware, the seven bands of the data were projected to coordinate system: UTM, Zone:
17 N, Datum: NAD 83. Erdas Imagine “Model Maker” was used to stack the separate
bands (minus thermal band 6) to produce one six-band multispectral image. To
reduce file size and processing time, the image was subset to a 1 km buffer around
the study area. The buffer was used to minimize edge effects in the classification.
Multispectral images can be radiometrically enhanced to correct for atmospheric
attenuation effects and allow for greater separability of reflective materials. Although
atmospheric correction is critical for detection of subtle differences in land cover
classes that have closely related spectral characteristics, land cover types in an
Anderson Level I classification are dissimilar enough that radiometric correction is
often not necessary for discrimination between classes (Jensen
2005
). After visual
examination of the Landsat 5 data, it was decided that the negligible amount of
visible haze in the image would not influence the separation of the six land cover
types. The data were not resampled to correct for atmospheric effects.
10.3.4
Image Analysis
The study area was classified into six land cover types with minor modifications
to the seven Anderson Level I classes (Anderson et al.
1976
) using standard USGS
methods. Considering the unique character of the study area and the objectives of
this study, the following classes were chosen for the classification: impervious,
grass, agriculture, bare soil, forest and water/wet vegetation. ENVI software was
used for all digital image processing and accuracy assessment, with the exception of
selected preprocessing techniques that used Erdas Imagine (Leica Geosystems)
software.
A common issue with land cover classification in agricultural regions is confu-
sion between agriculture and urban land cover types. Barren agricultural fields are
often misclassified as impervious surfaces and vice versa. This is due to the highly
reflective nature of dry bare soils, which have similar spectral signatures to that of
impervious surfaces. Likewise, grassy agricultural vegetation such as winter wheat
has reflectance values similar to those of turf grass commonly used on golf courses
and in urban and residential landscaping. Because these misclassifications will
ultimately affect the accuracy of the analysis, agricultural land was classified
separately from non-agricultural land. This was accomplished by stratifying the
data based on agricultural information in GIS parcel data from Lucas, Wood and
Ottawa Counties. Using ENVI image processing software, the agricultural parcels
were subset from the data based on each parcel's Current Agricultural Use Value
(CAUV), a tax credit for agricultural land that has proven to be a good indicator of
