Invasive Timber Studies

Where is the loblolly pine?
The Savannah River Site in South Carolina is managed by the U. S. Forest Service. Defined geographic units are planted with a single timber species such as southern yellow pine and managed till maturity when the timber is harvested and a new generation of trees is planted. Some species, such as loblolly pine, don't always remain confined to where they are planted. Can you distinguish loblolly pine from other timber species in Landsat Thematic Mapper imagery?

IMAGINE Subpixel Classifier did!

Includes material © Space Imaging L.P.

Objective
Loblolly pine is an invasive species and the USFS wanted to determine where the species was and how much was present.

Approach
IMAGINE Subpixel Classifier was used to develop a signature for loblolly pine from Landsat TM imagery. The results were used in conjunction with a USFS stand map to quantify the amount of loblolly in each stand.

Challenges
· Ground survey and aerial photography were very time consuming
· Spectrally similar tree species were difficult to distinguish
· Land cover classifiers did not provide sufficient discrimination

Solution
A signature developed using IMAGINE Subpixel Classifier was used to process the Landsat imagery to look for occurrences of loblolly pine. These detections were aggregated according to the geographic units specified on USFS Savannah River stand maps as shown on this map. Each stand is color-coded to indicate what specie the unit is being managed for, and to indicate how much loblolly is present.

For example, all light yellow areas are managed for longleaf pine and small or non-existent amounts of loblolly were detected. The gold areas are also managed for longleaf pine, but at least 20% of the stand appears to be loblolly pine instead. The black dots within each stand indicate the location of the loblolly detections. They are frequently located along drainage areas within the stand.

This map can be used by resource managers to determine the location of the densest loblolly stands and to see how much loblolly has infiltrated non-loblolly stands. To measure the accuracy of the classification, 170 pixels were randomly selected for field verification to independently evaluate errors of omission and commission. The total classification accuracy was 88% (91% omission accuracy and 85% commission accuracy).