Land Cover Analysis - Great Lakes Project

Land Cover Analysis

Data Development Methods

The project focused on the coastal area of the Great Lakes region which included the states of Illinois, Indiana, Michigan, Minnesota, New York, Ohio, Pennsylvania, and Wisconsin.
Mapping boundaries were determined by the combined extent of estuarine drainage areas (EDAs), Coastal Drainage Areas (CDAs), Coastal Zone Management Act boundary definitions by state, and coastal counties.
Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper (ETM) imagery were used. This imagery was chosen because it provided a cloud-free image mosaic with three seasons of imagery.
Land cover change requires two dates of imagery to make a comparative assessment. Imagery for the earlier date was chosen based on obtaining a cloud-free image mosaic of leaf-off imagery.
The imagery were assessed for spatial (horizontal) accuracy.
Statistical techniques were used to identify and remove cloud cover. After removing these clouds, cloud-free pieces of overlapping images were reinserted into these holes, forming a cloud-free mosaic.

The image area was separated into 233 separate spectral clusters using supervised classification to create a signature file. The signature file was then run through a supervised classification process.
The resulting clusters were labeled with primary land cover classes such as forest, agriculture, and developed by overlaying field-collected data. These land cover classes were further refined into the Coastal Change Analysis Program (C-CAP) land cover classification scheme, developed for the contiguous U.S.
Wetland classes were refined using National Wetland Inventory (NWI) data and developed classes using rasterized TIGER 2000 data. Cultivated and grassland classes were differentiated using crop rotation indicators. If agricultural land indicated crop rotation through significant spectral change or bare soil in at least one season of imagery, it was called cultivated; otherwise, it was called grassland.
Draft classifications were developed prior to going to the field.

Analysts visited areas of uncertainty or confusion that were encountered in the initial classification to determine if the preliminary identification of land cover classes was correct.
Project partner expertise and local knowledge were critical components in the characterization of land cover for the Great Lakes. Analysts met with project partners, who provided local expertise and access to state, federal, agricultural, and closed lands, as well as to reserves.
Project partners accompanied Center staff and assisted with the completion of aerial and ground surveys. Aerial surveys played a crucial role as many areas were inaccessible via ground transportation.

During the final land cover class development, many versions of map products were created to deal with class confusion issues.
Some area-of-interest editing was necessary to ensure the best accuracy of products.

Center staff generated stratified random samples of field points to visit and verify. Center staff then made coordinated field visits to examine as many points as possible.
An interactive laptop and Global Positioning System interface was utilized for accuracy point verification. This process uses the satellite imagery as a visual backdrop to aid Center staff as they navigate to each point and examine whether that point has been correctly classified on the satellite imagery.
Due to their expertise and local knowledge of both the area and land cover found throughout the Great Lakes, local experts assisted with the final ground and aerial surveys.
Throughout the final accuracy assessment, Center staff identified what could be improved upon and noted any errors in the land cover classification. These areas were then corrected.

Final products include C-CAP land cover and change data for the coastal region of the Great Lakes. These data sets are available individually for downloading.
Complete information about the data development methods can be found in the metadata file that accompanies the data when downloaded.