Geoscience Reference
In-Depth Information
Fig. 8.5 Screen shot of Maumee GIS ArcIMS product as viewing at the community scale and
highlighting the SSURGO soils layer
fall into the study area and were used for accuracy assessment of the supervised
classification for the study area. Normally, a minimum of 50 samples for each class
is good enough for the accuracy assessment. However, when a study area is larger
than one million hectare, the minimum number of the sample should be increased
to 75 or 100, thus for this study, 75 samples were used.
To perform a maximum likelihood supervised classification, a training set for
each class of corn, soybean, hay, and wheat was created. By using the training
samples, pixels were selected by using an Area of Interest (AOI) tool for each class.
The training samples were visualized in different colors in terms of cardinal directions
to consume less effort and time in collecting pixels. For each class, about 100 fields of
pixels were collected. Those pixels were the reference for the computer to classify the
entire image. For the forest and water classes, pixels were collected visually. Water is
obvious in a satellite image by its shape and color of navy to light blue with bands 4, 3,
and 2 as red, green, and blue in color composite. Forest is also visible and easily
identifiable in an image by its texture and color of red with the same condition of the
color composite as that used for the water.
After running the supervised classification, sieve and clump functions of ENVI
were applied to the classified image except for the water and forest classes to
smooth isolated pixels. The sieve function identifies an isolated pixel, and the
