1. Scene Acquisition Two images were acquired for each area, one from a spring time frame, the second from a summer date. Nearly cloud-free images were acquired in almost all cases. The following is the list of scenes acquired and used to produce the Landcover 2002 product. Path 28 Row 30-31 May 20, 2002 Landsat 5 Row 30-31 July 15, 2002 Landsat 7 base Path 27 Row 30-32 May 13, 2002 Landsat 5 base Row 30 July 16, 2002 Landsat 5 Row 31-32 Sep. 2, 2002 Landsat 5 Path 26 Row 30-32 June 7, 2002 Landsat 5 Row 30-32 Sep 3, 2002 Landsat 7 base Path 25 Row 30-32 May 26, 2003 Landsat 7 base Row 30-32 Aug 19,2002 Landsat 5 Path 24 Row 31 May 27, 2003 Landsat 5 Row 31 Sep. 5, 2002 Landsat 7 base

2. Processing Areas Six processing areas were used, the central ones generally extending along the satellite path from southern Minnesota into northern Missouri, and encompassing the equivalent of three Landsat scenes in a single file. Each Path listed above represented a separate processing area. However, Path 27 was spit into two processing areas, called North and South, because no summer scene was found that spanned the full path without clouds.

3. Within date layer stacking The 30 meter products for each Landsat 7 scene were resampled to match the 15 meter Panchromatic band and layer stacked with it. This resulted in a 7 band stack. In a similar way, the six 30 meter bands of each Landsat 5 image were stacked into a 6 band file, and resampled to 15 meter resolution.

4. Principal Components Analysis Principal Components Analysis (PCA) was run on images for both dates. The resulting PCA images were examined, the best bands selected from each for use in producing the image classification. Typically from four to five principal components were used from each image.

5. Scene to Scene Georeferencing, Layer Stacking All processing and preprocessing steps were done on path-oriented products. This was done to preserve pixel spatial relationships until the last possible moment, in order to minimize the destructive effects of resampling. To accomplish this, one image of each image pair (usually a Landsat 7 image) was chosen as the base image. Landsat 7 scenes provided a 15 meter panchromatic band which, when coregistered with the multispectral bands, provides a basis for more precise georeferencing of the image to the final map space. A 15 meter pixel size was used for all processing. Control points were taken to georeference the Landsat 5 scene to overlay on the Landsat 7 scene in path-oriented space. These were used to resample the Landsat 5 Principal components image into the space of the Landsat 7 image, at 15 meter resolution. The resampled Landsat 5 Principal Components image and the Landsat 7 Principal Components image were layer stacked into a single file, now containing 8 to 9 data layers, in path-oriented space. The edges of these images were clipped so that there remained no points at which fewer data layers were present than were present in the center of the image.

6. Image Classification An unsupervised ISODATA classification process was run on the principal components layer stacks, to generate statistics for 240 classes. These statistics were used in a supervised classification process, using the Maximum Likelihood classifier, to generate a 240 class classification. A 1000 class classification was similarly produced, but was used only for finer discrimination of the coniferous forest class.

7. Class Labelling The ERDAS Imagine Class Grouping Tool was used to group and label the unsupervised classes into their final land use classes. Seventeen land use classes were in the final label set. The Dendrogram ancillary tool was used to bring together classes with similar spectral character. Where useful, class groups were constructed to facilitate further refinement using the Fuzzy Recode tool.

8. Fuzzy Recode The labelled classes were remapped into their final land-use classes using the ERDAS Imagine Fuzzy Recode tool, which is guided by a weighting scheme, and performs a convolution process on each pixel, bringing neighborhood information into the final class assignment of each pixel.

9. Manual Raster Editing Some spectrally confused classes were separated by manual editing processes. Specific edits performed in each processing area were to separate Commercial/Industrial from Barren, to separate Coniferous Forest from Wet Forest, and to produce a cloud mask.

10. Image Recombination, Georeferencing Three separate files for each processing area, the labelled classes, the coniferous forest file, and the cloud mask, were recombined into a single file by an automated process. Control points were picked from the original Panchromatic band to georeference the scene to the final map-oriented framework in UTM zone 15. The recombined image for each processing area was resampled to UTM.

11. Image Mosaicking, Masking These reprojected images were mosaicked into a single file, covering the entire state of Iowa. A cutline was used between Path 27 North and Path 26 to eliminate clouds present in Path 26. A generalized perimeter of the state was developed which provides a buffer of at least 20 km around most of the state. Data outside this perimeter was masked to the background value of zero.

Entity_and_Attribute_Overview:

Landcover 2002 Classes

1	Open water
Spatial/spectral areas of open water, generally without any vegetation present
2	Wetland 
Spatial/spectral areas of marsh land, and sometimes forested wetlands.  Also includes areas of saturated soil. This class generally reflects the presence of both a wetness signature and a vegetation signature.
3	Wet forest
Edited class of mixed pixels of forest and water.  These pixels spectrally appeared with the coniferous forest class, and were manually edited into this class in areas that clearly represented waterways and not coniferous forest. 
4	Coniferous forest
Spatial/spectral areas of evergreen forest with editing.  Some conifers (especially cedar) occur frequently in stands too thin to be separated spectrally.  Also may include some deciduous forested areas in terrain shadow or in a floodplain situation.
5	Deciduous forest
Spatial/spectral areas of broadleaf deciduous forest and trees.
6	Ungrazed Grasslands 
Spatial/spectral areas of ungrazed grasses.  Includes rural road and ditch complexes, grassed waterways, some grassland/forest edge areas, and some tracts of grasses that are spectrally separable, and appear to be unmanaged.  This is the catch-all class for grasslands that are not otherwise separable into more detailed classes. Field observation suggests that areas with more native character are more likely to be assigned to this class.
7	Grazed grasslands 
Spatial/spectral areas of grasslands that show a healthy vegetative signature in spring, generally due to absence of senesced vegetation.  Some of the more lush areas of grazed grasses will appear in the Alfalfa class.
8	CRP 
Spatial/spectral areas of unmanaged grasses in heavy stands.  This class reflects the presence of senesced grasses in the spring, masking any lush vegetation response.  Field observation suggests that planted areas of native grasses and also areas of aliens such as brome grass, etc. are both included this class 
9	Alfalfa, winter wheat, lush grass
Spatial/spectral areas of very lush vegetation, usually consisting of alfalfa fields, but sometimes including other lush grasses, such as winter wheat, lush areas of grazed fields, and golf courses.
10	Corn
Spatial/spectral areas of row crop planted to corn in 2002.  Will include small amounts of spectrally confused areas planted to soybean or other crops.  In the very southeast corner of Iowa where only a single (Spring) date of image was used for the classification, all rowcrop areas are assigned to corn.
11	Soybeans
Spatial/spectral areas of row crop planted to soybeans in 2002.  Will include small amounts of spectrally confused areas planted to corn or other rowcrops.
12	Other agriculture
Spatial/spectral areas of row crop planted to other crops besides corn and soybeans in 2002.  In general these areas were characterized by early harvest (or no crop planted), and presented a bare soil aspect in the summer image.  This class may include some areas that are totally barren of vegetation.
13	Roads
Spatial/spectral areas that are primarily parts of major roadways or city streets.  Roadway classes that occurred nearly exclusively in rural areas were generally assigned to the Ungrazed grasslands category.
14	Commercial/Industrial areas
Spatial/spectral areas that are largely covered with broad expanses of impervious surfaces, such as asphalt and concrete, usually in urban centers.  This includes both building roofs, parking areas, and some streets or highways.  May include some areas of exposed rock or sand, such as quarries or sandbars that were missed in the edit.
15	Residential areas
Spatial/spectral areas of mixed pixels containing both vegetation and extensive impervious surfaces, and generally occurring almost exclusively in urban areas.
16	Barren (quarries, sandbars, etc)
Edited areas that are largely covered with exposed rock or sand, such as quarries or sandbars.  These classes were spectrally confused with the commercial/industrial class, and were assigned to this class by a manual edit process.
17	Missing data (clouds, shadow, pixel dropouts)
Edited areas of missing data, usually due to the presence of cloud or shadows in the imagery.  Where possible, areas where clouds or shadow were present in one of the multitemporal images, but not both, were filled in with an interpretation derived solely from the unaffected image.  These pixels were assigned to this class by a combination of manual editing and automated process.

This digital, geographically referenced data set was developed by the Iowa Department of Natural Resources to carry out agency responsibilities related to management, protection, and development of Iowa's natural resources. It resides in the Natural Resources Geographic Information System library. Although efforts have been made to make it useful to the Department, the Department assumes no responsibility for errors in the information. Similarly the Department assumes no responsibility for the consequences of inappropriate uses or interpretations of the data made by anyone to whom this data has been made available. The Department bears no responsibility to inform users of any changes made to this data. Anyone using this data is advised that precision implied by the coverage may far exceed actual precision. Comments on this data are invited and the Department would appreciate that documented errors be brought to staff attention. The development of the coverages in the NRGIS Library represents a major investment of staff time and effort. As a professional responsibility, we expect that the Iowa DNR will receive proper credit when you utilize these GIS coverages and that the original documentation will remain with any redistribution of these coverages.

Land Cover of the State of Iowa in the Year 2002

Metadata: