Using forest inventory plot data and satellite imagery from MODIS and Landsat-TM to model spatial distribution patterns of honeysuckle and privet
| Authors: | Dumitru Salajanu, Dennis M. Jacobs |
| Year: | 2009 |
| Type: | Scientific Journal |
| Station: | Southern Research Station |
| Source: | In: Proceedings of ASPRS 2009 annual conference—reflection of the past, vision for the future. 2009 March 9-13; Baltimore, MD. Bethesda, MD: American Society for Photogrammetry and Remote Sensing. |
Abstract
Forest inventory and analysis data monitor the presence and extent of certain non-native invasive species. Onforestland, non-native species are considered part of the understory vegetation and can be found near canopyopenings as well as and the forest edge. The objective of this study is to incorporate the presence of select nonnativespecies into forest classification modeling procedures and determine the accuracy for producing non-nativespecies spatial distribution classifications. A secondary objective is to compare classification accuracies of thedifferent spatial resolution data (Landsat-TM and MODIS), which suggests that an increase in resolution provides anincrease in overall accuracy. The classification results provide forest distribution combined with non-native speciesoccurrence for honeysuckle (Lonicera spp.) and privet (Ligustrum spp.). Subsets of the plot data are used in adecision tree modeling process (See5) applied to the satellite data (Landsat-TM, MODIS), and ancillary data toclassify the land cover and model privet/honeysuckle spatial distribution. Classification results show that overallclassification accuracy for the percent of pixels correctly classified (%PCC) increased from 67.5% to 72.5% (privet)and from 67.5% to 70.0% (honeysuckle) when privet and honeysuckle are coded as present in the plot.Comparisons between overall classification accuracy show a 5.0% increase for privet, and 2.5% for honeysuckle. Acomparison between MODIS and Landsat-TM classifications shows a 3.7% increase in accuracy when both privetand honeysuckle are coded in several categories based on their percent of participation on the plot. Classificationsfrom Landsat-TM and MODIS models show a higher confusion in the privet/honeysuckle categories (percent privetor honeysuckle on plot), however the Landsat-TM model performed slightly better.