General Phase 2 Land Cover Classification Process for Iowa Gap - 2/2001

Preprocessing

Use Phase 1 land cover grid extents to make a polygon clip coverage in Arc/Info. In Imagine, subset each input TM scene with the clip coverage. Check that each referenced TM scene lines up exactly at the pixel level with the Phase 1 image; adjust if necessary within the map model. Recode the Phase 1 image classes to 1 - 6 or 1 - 7, depending on presence of clouds. This allows for correct overlaying of subsequent classifications.

Wetlands

Mapjoin NWI county basin model results to cover the scene. Recode text to numeric field for importing into Imagine. Model aggregates wetlands into temporary, seasonal, semi-permanent, permanent and water.

Mapjoin original NWI county data to cover the scene. Recode text to numeric field for importing into Imagine. Original data is used to add back forested, shrub and saturated wetlands.

For north and south state edge scenes, mapjoin adjoining state's original NWI quads that overlap Iowa's border. Recode text to numeric field for importing into Imagine. Original data is used to write herbaceous, shrub and forested wetlands onto Phase 1 land cover.

Convert the model results and original NWI data from an Arc/Info coverage to an Imagine .img file. Align the pixels to the Phase 1 land cover image by changing the map model. Overlay basin NWI data, then original NWI data, then Minnesota or Missouri NWI data, if applicable, onto Phase 1 land cover.

Clip land cover to Iowa border plus Minnesota or Missouri NWI quad boundaries, if applicable.

Clouds

If there is a cloud class from Phase 1, determine which dates available are cloud-free for those areas.

Create a mask of the cloud and cloud shadow areas from the to-date land cover for each cloud-free date.

Run unsupervised classification on the cloud mask(s) with 80 classes. Use field data, DOQs and CIR bands to assign alliances to classes.

Overlay final alliances onto to-date land cover or recode general classes into corresponding classes in the to-date land cover.

Forest

Use to-date land cover to mask out all but trees in the earliest date available (April or early May). Make sure skip factor=1, rerun statistics.

2530

Run tasseled cap transformation on early tree mask; keep first three bands. Run unsupervised classification on the tree mask with 80 classes.

2630

Instead of the tasseled cap process, display the early tree mask in the viewer (bands 4,5,6) for the CIR effect. Use the spectral profiler to get several profiles each of upland deciduous forest, cedar forest and pine forest. In the graph, check the values for band 5 to see if there is enough difference (100 vs 75) between the deciduous and evergreen sites. If so, use the decid.gmd model and set the minimum cut off value for deciduous trees to just above the evergreen value in band 5. The model result will have the deciduous forest pixels ranked 1. Overlay that on the to-date land cover; the pixels left in the Phase 1 woody class will be the evergreens and mixed forest.

Use those remaining pixels to make a mask of the April tree mask. Do an unsupervised classification on all 6 bands with 50 classes.

Use CIR bands displayed under the classification to pick out the evergreens and evergreen/deciduous mixes. Use sample sites to check evergreens. Tree classes left after evergreens are done are upland deciduous; there may be some deciduous trees left because the cutoff for deciduous forest in the model was an estimate.

Overlay the forest classes onto the to-date land cover.

Woodland

Found in the herbaceous class in 2630 in the early Spring date (40691) using bands 3,4,5 once the obvious grass and shrub alliances were assigned. Unsupervised classification was run with 50 classes.

Grasslands

Use to-date land cover to mask out all but herbaceous in early and late Spring dates (April and May) or early Spring and Fall dates (April and October). The choice of which to use depends on the degree of separation between warm and cool season sites when sampled with the spectral profiler. Use May for cool vs warm grass; if no May date exists, use the October date.

Run tasseled cap transformation on the two dates and keep the first three bands of each. Layer stack the 6 bands together and rerun the statistics for skip = 1. Run unsupervised classification on the herbaceous mask with 80 classes.

Use CIR bands displayed under the classification to pick out cool and warm season grasses. Use sample sites to find shrubland, woodland, savanna, and check grasses. Use DOQs to find grass/other mixes.

Overlay grassland, shrub and grass/mix alliances onto the to-date land cover.

Urban

Use to-date land cover to mask out all but artificial in a late spring or early Fall date (October). It is better to have contrast with growing vegetation.

2530

Run principal components analysis on the two dates, keep the first three bands from each and combine the two images into one, 6 band image. Run unsupervised classification on the artificial mask with 20 classes.

2630

Use the DOQ and CIR bands displayed for the Fall date (October) and find areas of mostly pavement and mostly vegetation. Use the spectral profiler in these sites to see which 3 bands provide the best separation between high and low vegetation. Subset out the 3 best bands and run an unsupervised classification on that image with 20 classes.

Use DOQs to differentiate between artificial/high and artificial/low vegetation in the resultant classified image.

Overlay artificial/high and artificial/low vegetation alliances onto the to-date land cover.

Cropland, Barren, Water

These Phase 1 classes will not be further delineated unless there is time at the end of the project to subdivide the Barren class.

Return to the Iowa Gap Home Page.