Electric Peak, photo Sherry Ballard, California Academy of Sciences	West-Central Montana SILC3 Covertype for Electric Peak

The satellite imagery land cover classification system (SILC) was started in the early 1990s to create regional land cover type, tree size, and tree canopy GIS databases for Montana and Idaho. The University of Montana developed the system with Region One of the U. S. Forest Service, the USGS Gap Analysis Program (GAP), and the state of Montana. The SILC land cover type databases were developed for modeling at the regional or sub-regional level (1:500,000 or 1:100,000 scale). The SILC process is divided into two stages. First, an image segmentation is used to delineate ground features and create a region grid GIS database. Second, segmented regions are labeled in a supervised classification. In early SILC projects a second GIS database was created for riparian areas and were classified in pixel classification. In later SILC projects riparian areas were included in the classification by using a variable minimum map unit (MMU) (1 acre for riparian and 5 acres for upland areas).

Currently, new image segmentation processes are being developed with Landsat TM-7 imagery resampled to 15 meters. The lab is exploring two new types of image segmentations: first is a new SILC approach using ISODATA cluster analysis and merge techniques; second is the use of eCognition software for multi-resolution segmentations. These new techniques are being used for sagebrush canopy cover and species mapping near Billings, Montana.

For more information go to: Mapping Sagebrush/Grasslands from Landsat TM-7 Imagery

Additional research is being done in western Montana in predominately forest types to improve tree size and tree canopy segmentations. Another study area is in the Eastern Cascades on the Wenatchee National Forest. In both areas improved image segmentation and supervised classification systems are being tested.

Another area of research is using IKONOS 4-meter data at Pinnacles National Monument. IKONOS imagery presents unique issues of scale and resolution and may require multi-scale segmentation. Ecognition software is being used to experiment with multi-resolution segmentations to create variable map units in different vegetation communities. Use of Landsat TM data resampled to 15 meter data or IKONOS imagery with 4-meter data may help improve classifications of difficult vegetation types like sagebrush or noxious weeds.

SILC1 Through SILC2 Projects:

For more information see final report PDF (file size 67 MB): Current Vegetation Map of Northern Idaho and Western Montana

Research done during the SILC1 project led to a revised image segmentation process. In SILC1 only 25-40 spectral classes were used in the unsupervised classification. A new segmentation process was developed with 100-130 spectral classes. An iterative ISODATA cluster analysis was used to create the spectral signature classes which were then used to classify the imagery pixels into one of the spectral classes. The iterative ISODATA cluster analysis with more spectral classes helped break up very large regions created in the SILC1 process. A second generation SILC2 process was developed from this research.

A second important feature of the SILC2 process was the combination of the upland and riparian image segmentations. First, potential riparian areas were identified from DEM modeling. A mask was used in the merge process to restrict merging in potential riparian areas. This allowed riparian areas to be included with upland areas in the final image segmentation database. The new SILC2 process allowed potential riparian areas to have a 1 acre MMU and upland areas to have a 5 acre MMU. A single GIS database could be created for all areas within the Landsat image.

SILC2 - Little Missouri National Grasslands Classification
The SILC2 process was used on three Landsat TM-5 images in the Little Missouri National Grasslands for the Custer National Forest in 1997. A quarter section of a fourth Landsat TM-5 image was resampled to 15m and was also classified. The Custer SILC2 classification focused on rangeland areas verses SILC1 which was predominately forested areas. In particular more grassland and shrubland types were classified. In the new image segmentation process small features such as woody draws and riparian areas were captured much better than the SILC1 process. In the supervised classification a single nearest member euclidian distance classifier was used to classify both upland and riparian cover types. A total of 36 land cover types (including riparian types) were mapped at 30 meters.

For more information see final report PDF (file size 2 MB): Land Cover of the Little Missouri, Grand River, and Sheyenne National Grasslands

SILC2 - Salmon-Selway Classification
The SILC2 process was used to classify two Landsat TM-5 images in the Salmon-Selway wilderness for the Craighead Wildlife-Wildlands Institute in 1998. The northern Landsat image was from 1991 and the other from 1995. During the Salmon-Selway classification, a new supervised classification was developed using a nearest neighborhood distance weighted euclidian classifier combined with a spatial classifier. A range of nearest neighborhood sizes from 5-25 were evaluated with the training data set. The expanded neighborhood sizes combined with a spatial classifier improved classification accuracy over 10 percent. A total of 24 land cover types (including riparian types) were mapped at 30 meters.

For more information see final report PDF (file size 23 MB): Vegetation patterns of the Salmon-Selway ecosystem

Metadata link: SALSELZ Region Grid Metadata

SILC2 - Central Idaho Classification Project (CICP)
The SILC2 process was used to classify six Landsat TM-5 images in central Idaho for the Boise National Forest in 1998. The classification was based on Landsat TM-5 imagery from 1993-1995. During the central Idaho classification, more research was done on using nearest neighborhood distance weighted euclidian classifiers and spatial classifiers in the supervised classification. Supervised classifications were run using four different distance weighted euclidian classifiers and two different spatial classifiers. Each of the six Landsat images was classified separately and then edgematched together. A total of 42 land cover types (including riparian types) were mapped at 30 meters.

For more information see final report PDF (file size 23 MB): Landsat Vegetation Mapping of the Southwest and Central Idaho Ecogroups

Metadata link: CICPSILC2Z Region Grid Metadata

SILC2 - Eastern Montana Classification (EMONT)
The SILC2 process was used to complete the classification of 20 Landsat images in eastern Montana for the Montana Department of Fish, Wildlife, and Parks in 1998. The classification was based on Landsat TM-5 imagery from 1989-1994. The project required help from state, federal, and conservation agencies to acquire training data across eastern Montana. A total of 62 land cover types (including riparian types) were mapped at 30 meters.

Please see the MTGAPVEG metadata link below for more information about the classification

SILC1 and SILC2 combined land cover GIS databases (assembled from above projects):

Montana Gap Analysis Project Land Cover
Land cover was mapped across most of Idaho and all of Montana between the SILC1 and SILC2 projects. The SILC1 and SILC2 GIS databases were edgematched together for the Montana Gap Analysis Project. Some of the SILC1 and SILC2 land cover type classes were combined to improve accuracy for MTGAP. A total of 33 Landsat TM images classified in SILC1 and SILC2 were used to create a 30 m land cover layer with a minimum map unit of 2 hectares. The final MTGAP GIS database had 50 land cover type classes and was resampled to 90 meters.

Metadata link: MTGAPVEG Region Grid Metadata
MTGAPVEG 90m Land Cover: MTGAPVEG Land Cover in Imagine 8.5 format (13 MB)

MTCOV30 Metadata link: MTCOV30 30m Region Grid Metadata
MTCOV 30m Land Cover used to create MTGAPVEG: MTCOV30 Land Cover in Imagine 8.5 format (58 MB)

FIRERISK 2000 - Northern Idaho and West-Central Montana Land Cover
A land cover type map of northern Idaho and west-central Montana was assembled from parts of the SILC1, Central Idaho, and Eastern Montana classifications. The land cover map was assembled for the FIRERISK analysis during the 2000 fire season. FIRERISK is a 30 meter grid that maps unique combinations of 7 input variables, including land cover type, conifer tree size and canopy cover, potential vegetation type, aspect, slope, and elevation breaks. The land cover layer used to create FIRERISK was assembled from 23 Landsat 5 TM scenes acquired from 1991-1995 and has 52 cover types.

Metadata link: FIRERISK Grid Metadata
FIRERISK 30m Land Cover: FIRERISK Land Cover in Imagine 8.5 format (40 MB)

SILC3 Projects:

SILC3 - Second Little Missouri Grassland Classification
The third generation SILC3 process was developed during the Little Missouri Grasslands research project for Region One of the U.S. Forest Service in 2000. Woody draws surrounded by large grassland areas were segmented using a variable minimum map unit based on the amount of spectral variation with a local area. Additionally, a potential vegetation type layer was used as a seed source in the image segmentation process. Masks were used in the merge process to restrict merging of unique features in homogenous areas. In particular small meadows, rock outcrops, and aspen stands were preserved in upland areas. The segmentation of riparian areas was also improved to separate unique stream reaches. The new variable minimum map unit ranged from 0.1-5 acres depending on the amount of spectral diversity across an area of the image. A total of 31 land cover types (including riparian types) were mapped at 30 meters.

Metadata link: LITMISS_Z Region Grid Metadata

SILC3 - Second West-Central Montana Classification
The SILC3 process was used to re-classify nine Landsat scenes in west-central Montana for the U.S. Forest Service in 2001. Seven Landsat TM-5 images from 1994-1996 and two Landsat TM-7 images from 2000 were classified and edgematched together. To improve accuracy, some SILC1 and SILC2 land cover type classes were combined and riparian classes were dropped from the map legend. In past projects riparian land cover types were often highly confused with similar upland land cover types. Riparian types could be identified using riparian zone modeling such as valley bottom modeling or DEM modeling. Significantly more training data (4-10 times) were used in the SILC3 classification than the SILC1 classification. Specifically, Forest Service Timber Stand Management Record System (TSMRS) timber stand data were used to improve the classification of conifer forest types and tree size/canopy classes. A total of 31 land cover types were mapped at 30 meters.

For more information see final report PDF (file size 550 KB): Landsat Vegetation Mapping of the West-Central Montana

Metadata link: SILC3REGIONZ Region Grid Metadata

South-West Montana Re-classification of SILC3 for Sagebrush Canopy Cover Types
Three of the SILC3 west-central Montana Landsat scenes were re-classified for sagebrush canopy cover classes for the Dillon Resource Area Office of the Bureau of Land Management in early 2002. In SILC3 all xeric shrublands and sagebrush types were classified as one combined type and had no canopy cover information. For this project five Sagebrush/Xeric Shrubland canopy cover classes were added to the map legend and a new classification was run for the three SILC3 scenes.

For more information go to the following link: Southwest Montana Sagebrush Canopy Cover Classification

SILC3 - Second Ashland Montana Classification
The SILC3 process was used to re-classify 3 Landsat scenes in southeast Montana for the Custer National Forest in 2002. Three Landsat TM-7 images from 2000 were segmented by WSAL and then classified by the Custer National Forest. To improve accuracy some SILC1 and SILC2 land cover type classes were combined and riparian classes were dropped from the map legend. Custer National Forest staff were trained by WSAL to use a set of supervised classification programs and ARC/INFO amls. Existing and new training data were assembled by the Custer National Forest and then used in the supervised classification. A total of 24 land cover types were mapped at 30 meters.

Metadata link: CUSTSILC3Z Region Grid Metadata

MTSILC3 Combined SILC3 and SILC2 Land Cover Grid
The SILC3 classification grids were edgematched together with 5 SILC2 classification grids to create a sixteen scene 30 m region grid database in 2002. The MTSILC3 region grid covers most of Montana and represents the most current land cover type mapping for Montana. The large region grid was created so users could query regions created from all of the individual image databases. The MTSILC3 region grid was created for the Northern Region of the U.S. Forest Service and is currently being updated to reflect vegetative changes since the individual scenes were classified.

Map Legend: Blue West-Central Montana SILC3, Orange Custer SILC3, and LtGreen SILC2

Metadata link: MTSILC3Z Region Grid Metadata
MTSILC3COV 30m Land Cover: MTSILC3COV Land Cover in Imagine 8.5 format (47 MB)