Project Overview: Western Oregon Vegetation Mapping Project
Warren B. Cohen, USDA Forest Service
Thomas K. Maiersperger, Oregon State University
Thomas A. Spies, USDA Forest Service
Doug R. Oetter, Oregon State University
Forest vegetation attributes should be mapped as continuous variables rather than classes, to maintain maximum flexibility for multiple end-users.
The methods should result in a seamless map by minimizing the effects
of variable atmospheric and illumination angles among adjoining images.
Perform first level classification to separate forested
areas from nonforested areas.
Derive and apply predictive equations for
the following forest attributes in the SOURCE SCENE :
- Percent green vegetation cover (%GVC)
- Percent conifer cover (%CC)
- Conifer visible crown diameter (VCDm)
- Local conifer stand age (AGEyr).
Derive and apply predictive equations for the forest attributes in the DESTINATION SCENES .
Assess prediction errors.
The following 1988 Thematic Mapper scenes were used in the analysis: 45/29,
45/30, 45/31, 46/28, 46/29, 46/30, 46/31, 47/28, 47/29. Data reduction and enhancement
was carried out using the Tasseled Cap transformation. A reference database of photo-interpreted
and ground survey data was compliled from the BLM, USFS, and Oregon State Department of
Forestry for model building and validation. Existing GIS layers and DEM data were also
used in the project.
The primary purpose of the first level classification was to separate forest land from non-forest land. An unsupervised classification was performed for each scene to develop shadow, water, cloud, and snow/ice classes. Subsequently,
existing GIS layers were used to assign pixels to urban, agriculture, and other non-forest categories. The remaining forest pixels were modeled in a continuous fashion during regression analysis.
A centrally located Source Scene was chosen (a composite of 46/28 and 46/29 which were acquired on the same date). Relationships were examined between image spectral response and the following forest attributes (as measured in the reference plots): 1) Percent green vegetation cover (%GVC), 2) Percent conifer cover (%CC), 3) Conifer visible crown diameter (VCDm), and 4) Conifer age (AGEyr). Regression models were generated and applied to produce Source Scene estimates for each forest attibute.
- The %GVC and %CC images contain cover predictions only for the pixels labeled as "forest" in the First Level Classification;
- The VCDm and AGEyr images models contain cover predictions only for the pixels labeled as "forest" in the First Level Classification and with values >= 70 in both %GVC and %CC (i.e. closed conifer forest);
- The Stand Replacement Disturbance (1972-1995) layer assesses change only for the pixels labeled as "forest" in the First Level Classification.
Applied radiometric normalization is the process by which land cover information derived for a Source Scene is extended to an adjacent Destination Scene . Within the overlapping, unchanged regions between two scenes, the Source Predictions are used to "train" the Destination Spectra in a regression modeling context. This results is a new predictive model which produces closely matched predictions between scenes, and effectively normalizes between-scene differences in radiometry in terms of the variable of interest. This process was repeated for each remaining destination scene, and the separate pieces were then mosaicked to form a unified map layer for each of the four vegetation attributes.
Predicted versus observed R2 and root-mean-square errors were calculated for each vegetation layer using an independent set of model validation plots. Additionally, a composite forest vegetation map was created, and a classification error matirx was generated
Landsat data were combined with GIS, photointerpreted, and ground reference data to cost-effectively map forest attributes across a large geographic area.
Forest attributes were mapped as continuous variables rather than as classes, to maintain maximum flexibility for end-users.
The applied radiometric normalization technique produced a near-seamless map by effectively minimizing the impact of variable conditions associated with each scene.
Seven digital data layers (Arc/INFO Grid format or Imagine8.3 format) are available for downloading below. Note: some information is currently not available (NA) due to updating or other modification. For any given file, download it to your workspace, uncompress the tar file using "gunzip filename.tar.gz", and then extract the data with "tar xvf filename.tar". Please refer to the metadata for information about each file.
**Requires Adobe Acrobat v.4**
Cohen, Warren. B.; Spies, Thomas A.; Fiorella, Maria. 1995. Estimating the age and structure of forests in a multi-ownership landscape of western Oregon, U.S.A. International Journal of Remote Sensing 16(4):721-746.
Cohen, Warren. B.; Fiorella, Maria; Gray, John; Helmer, Eileen; Anderson, Karen. 1998. An efficient and accurate method for mapping forest clearcuts in the Pacific Northwest using Landsat imagery, Photogrammetric Engineering & Remote Sensing 64:293-300.
Cohen, Warren B.; Maiersperger, Thomas K.; Spies, Thomas A.; Oetter, Doug R. 2001. Modeling forest cover attributes as continuous variables in a regional context with Thematic Mapper data. International Journal of Remote Sensing 22(12):2279-2310.
Cohen, Warren. B.,; Spies, Thomas A.; Alig, Ralph J.; Oetter, Doug R.; Maiersperger, Thomas K.; Fiorella, Maria. In press. Characterizing 23 years (1972-1995) of stand replacement disturbance in western Oregon forests with Landsat imagery. Ecosystems.
Laboratory for the Application of
Remote Sensing in Ecology
Forestry Sciences Lab
3200 SW Jefferson Way
Corvallis, OR 97331
Last updated: 24 May 2001
FSRN home /