Evaluating undeveloped urban forest resources using color infrared imagery

Abstract

The last decade has witnessed unprecedented urban expansion. The Edwards Plateau region of Texas is a prime example of this rapid growth. The populations of Austin and San Antonio, Texas increased 47.7% and 20.2% respectively from 1990 to 2000, thereby putting increased pressure on undeveloped rural areas. With urban development expanding into rural areas (mostly non-industrial private forest land), resource managers need to examine existing forest resources and recommend methods of protection to guide future development and conserve greenspace. Current tree inventory methods involve expensive and time-consuming ground survey techniques. The objective of this study was to develop an alternative method of assessing undeveloped urban forest resources using color infrared imagery. Specifically, the aim was to develop a mathematical relationship between crown diameter (CD) and diameter at breast height (DBH) for selected tree species in the study area and use that relationship to estimate DBH of trees from CD measurements derived from aerial photography. Using aerial photograph interpretation and photogrammetric techniques, a strong linear relationship between CD and DBH for the selected tree species was found. When this relationship was applied to the CD measurements derived from aerial photography, observed and predicted values of DBH were statistically the same at α=0.05. However, multi-stemmed trees or mottes misidentified as single trees raised questions about the utility of this method. Based on this study, it would be more effective to measure total crown area rather than individual trees. Another objective of this study was to determine the level of species classification that can be achieved with 1-m color infrared, digital orthophoto quadrangles (DOQ). Using digital image processing techniques, it was determined that DOQs alone are not suitable for the identification of single trees or the classification of individual species. However, using unsupervised classification techniques, the following classes can be identified- forest, range, impervious surfaces, and shadows - with an overall classification accuracy of 86%.