| Abstract
|
SILC3REGIONZ is an ArcInfo 30 meter region grouped grid (raster file) of ground feature polygons derived from nine Landsat Thematic Mapper (TM) images across west-central Montana. Each of the nine Landsat TM images were processed separately and were then edgematched together to form one GIS region database.First, each Landsat image was delineated into regions in an unsupervised classification. Second, each region was assigned a land cover type label in a supervised classification. Regions classified as conifer cover types were also assigned labels for tree size class and canopy cover. A total of 31 land cover type classes, 5 tree size classes, and 5 conifer canopy cover classes were labeled across the nine scene area. Accuracies were estimated for the land cover type and tree size classifications of each Landsat image. Third, the regions from all nine Landsat images were edgematched together into the SILC3REGIONZ database. Below is a description of how the original Landsat images were delineated into ground feature regions, how the regions were labeled to land cover type, tree size, and tree canopy, and how the nine Landsat images were edgematched into one database.Image Landsat Image-Date Covertype-Acc Treesize-Acc SCENEPOLY-ID
--------------------------------------------------------------------------
P40R27 7 7/27/99 69.8% 64.0% 49000001 ..
P40R28 5 7/26/96 69.0% 57.1% 50000001 ..
P39R27 5 7/01/95 73.5% 61.8% 51000001 ..
P39R28 5 6/28/94 74.7% 69.6% 52000001 ..
P39R29 5 6/28/94 75.7% 54.7% 53000001 ..
P38R27 5 7/12/96 75.9% 55.9% 54000001 ..
P38R28 5 7/12/96 74.6% 64.5% 55000001 ..
P38R29 5 7/12/96 66.0% 57.9% 56000001 ..
P37R29 5 7/24/00 64.9% 66.9% 57000001 ..Each image was run through an unsupervised classification to classify the approximately 56 million pixels into 130-150 spectral classes. A potential natural vegetation (PVT) classification provided by Region One of the USDA Forest Service was included as a seed source in the unsupervised classification. Using an iterative classification process with spectral signature files and PVT seed, each pixel was assigned to a spectral class. The pixels were then merged into regions (raster polygons) through a multi-stage merge algorithm using a variable minimum map unit based on heterogeneity of the surrounding pixels. Regions larger than 202.4 hectares were divided until all regions in the grid were less than 202.4 hectares (500 acres). The final unsupervised classification resulted in the creation of regions ranging in size from 0.1 hectare to 202.2 hectares. Each region was assigned a set of attributes for TM channels 1-7 and panchromatic from the imagery, along with elevation, aspect, and slope from the 30 meter digital elevation model (DEM); these attribute values were calculated based on the mean values of the 30 meter pixels within each region.Vegetation overstory (land cover type) was labeled using a supervised classification. The land cover type supervised classification was performed using a combination of manual and training data classifications. The following classes were manually labeled: urban, irrigated and non-irrigated agriculture, recently burned areas, water, mines, snow, and clouds. The other land cover types were labeled using training data and spatial classifiers. The training data were screened for data attribute accuracy, positional accuracy, and agreement with the imagery. The land cover type labels were filled into the COVERTYPE field of each scenes region grid.COVERTYPE DESCRIPTION
------------------------------------------------
1100 Urban or Developed Lands
2010 Agricultural - Dry
2020 Agricultural - Irrigated
3130 Very Low Cover Grasslands
3150 Low / Moderate Cover Grasslands
3170 Moderate / High Cover Grasslands
3610 Mesic Shrublands
3625 Sagebrush / Xeric Shrublands
4101 Aspen
4150 Mixed Broadleaf / Cottonwood Forest
4203 Lodgepole Pine
4204 Whitebark Pine
4205 Limber Pine
4206 Ponderosa Pine
4212 Douglas-fir
4214 Rocky Moutain Juniper
4215 Western Larch
4216 Utah Juniper
4223 Douglas-fir/Lodgepole Pine
4230 Douglas-fir/Ponderosa Pine
4237 Subalpine Fir/Spruce
4241 Mixed Upper Subalpine Conifer Forest
4242 Mixed Lower Subalpine Conifer Forest
4244 Mixed Xeric Conifer Forest
4400 Burns
5000 Water
7300 Rock
7500 Mines/Quarries
9100 Snow
9800 Cloud
9900 Cloud ShadowConifer tree size and canopy cover labels were assigned using a training data classification. The training data were screened for data attribute accuracy, positional accuracy, and agreement with the imagery. Due to a high level of disagreement between the training data tree canopy call (see accuracy assessment), and the imagery, the tree canopy data was recoded based on MNDVI splits. The tree size labels were filled into the TREESIZE field and the tree canopy cover labels were filled into the TREECANOPY field of each scenes region grid.TREESIZE DESCRIPTION
-------------------------------------
1 Sapling 1.0-4.9" DBH
2 Pole 5.0-8.9" DBH
3 Medium 9.0-14.9" DBH
4 Large 15.0-20.9" DBH
5 Very Large 21.0-99.0" DBHTREECANOPY DESCRIPTION
-------------------------------------
1 1-9% Cover
2 10-24% Cover
3 25-44% Cover
4 45-64% Cover
5 65-100% CoverA cross validation (leave-one-out), accuracy assessment for land cover type and tree size of each scene was performed using the entire training data set. No estimate of accuracy was performed on the manually classified land cover types.Edgematching was done by selecting regions from only one of the scenes in overlap areas. First, the imagery was visually analyzed to select cloud free or snow free areas in overlap zones. Next, the images were ranked based on overall image quality. Finally, after the land cover type classification, regions were selected that created a consistant covertype pattern across the transition zone from one scene to another. The end result was a region grid with 4,820,703 regions.Image Landsat Date Priority Number of Regions Used
-------------------------------------------------------------
P40R27 7 07/27/99 1 754,813
P40R28 5 07/26/96 2 580,429
P39R27 5 07/01/95 3 290,811
P39R28 5 06/28/94 4 315,795
P39R29 5 06/28/94 5 472,441
P37R29 7 07/24/00 6 684,885
P38R27 5 07/12/96 7 683,615
P38R28 5 07/12/96 8 628,551
P38R29 5 07/12/96 9 409,363
-------------------------------------------------------------
Total Regions in Database 4,820,703To help track similarities and differences in map legends between this classification and previous ones (SILC1 and MT-GAP) cover type names and codes numbers were not changed if the actual types were mapped using similar training data. Please see the individual scene report documents for further details.
|
| Purpose
|
This land cover classification from the nine Landsat TM scenes was prepared for the Eastside Assessment Area under contract to the USDA Forest Service. These data were produced to delineate ground features and map existing land cover in a standardized, consistent manner across the project area. This land cover grid is suited for analysis at the regional, sub-regional, and landscape levels; it can also provide support for many management disciplines, including timber, wildlife, fisheries, and recreation.
|
