Project Number: WVA00074
CRIS Number: 0168879
INCORPORATING LANDSCAPE PATTERNS AT VARYING SCALES INTO LONG-TERM FOREST MANAGEMENT PLANS
Investigators: Gribko, L. S.
Performing Department: Forestry -- 1240
Start Date: 08/04/1995
Termination Date: 06/30/2001
Reporting period: 08/04/1995 to 06/30/2001
Termination Report:
Important accomplishments during the duration of this project included: 1) an analysis of the sources of error in computer-assisted classification of Appalachian forest cover types based on large-scale aerial photographs, 2) determination of the optimal spatial resolution for the acquisition of aerial images of Appalachian hardwood forests, and 3) a methodology for the classification of species composition of mixed hardwood forests using remotely-sensed digital images. First, we determined that shadowing caused by topography and low angles of illumination was a significant source of error in cover type classifications made using photographs taken at the peak of autumn color change. In addition, subtle differences in shades of some leaf colors could not be discerned using supervised classification of the scanned images. In particular, oaks and species such as black gum could not be differentiated. Because the presence or absence of oak is an important classification characteristic, the fact that this species could not be reliably identified was found to be a significant impediment to the use of aerial photographs in cover type classification. Using a more advanced digital image acquisition system that allowed us collect images at very high resolution, we were able to avoid many of the sources of error revealed in the initial classification. Ground-truthing revealed greater than 70% accuracy in the classification of individual tree species. Identification of species of economic importance, especially black cherry, was particularly reliable using the digital imaging system. We further found, using autocorrelation techniques, that images should be obtained at a resolution of approximately 2.0 meters or about one-third the average width of the tree crowns. Images of higher resolution did not provide enough additional accuracy to offset the significantly greater computational time required. Several poster presentations have been made describing the optimal scale analyses. A manuscript describing the classification methodology is still in preparation. Two master's theses also resulted from this work.
Publications:
Pacurari, D., Gribko, L. S., Warner, T. A., and Yuill, C. B. 2001. Determination of an optimal spatial resolution of remotely sensed images for mixed Appalachian forests. 3rd International Conference on Geospatial Information in Agriculture and Forestry. November 5-7, 2001. Denver, CO. (poster presentation, paper published in proceedings)
Pacurari, Doru I. 2000. Evaluation of the use of remotely sensed images to speciate mixed Appalachian forests. Masters thesis. West Virginia University, Division of Forestry.
Pacurari, D., Gribko, L. S., Warner, T. A., Wilson, T. H., and Yuill, C. B. 1999. Optimal spatial resolution for Appalachian hardwood forest images. Fourth International Airborne Remote Sensing Conference and Exhibition. Ottawa, Ontario, Canada. June 21-24, 1999. (poster presentation; brief paper published in proceedings)
Neese, C. P., Jr. 1998. Error in computer-assisted classification of Appalachian forest cover types. Masters thesis. West Virginia University, Division of Forestry.
Neese, C. P., Jr. and Gribko, L. S. 1997. Assessing Variability within Forest Cover Types Delineated Using Color-infrared Aerial Photographs. p. 400 in: Pallardy, S. G., Cecich, R. A., Garrett, H. G., and Johnson, P. S., eds., Proc: Eleventh Central Hardwood Forest Conference. March 23-26, 1997. Columbia, MO. USDA For. Serv. Gen. Tech. Rep. NC-188. 401 pp. (poster presentation; abstract published in proceedings)
Impact:
Currently, forest cover type delineation is accomplished in central Appalachian hardwoods through manual interpretation of relatively low-resolution aerial photographs. This method is extremely expensive and labor intensive and the accuracy of the results depends upon the ability of the interpreter to differentiate among multiple shades of color and slight textural differences. The development of techniques for automated interpretation of high resolution digital images using geographic information systems (GIS) technology significantly improves our ability to accurately describe landscapes dominated by deciduous tree species.