|EVERITT, JAMES - Retired ARS Employee|
Submitted to: Geocarto International
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/29/2010
Publication Date: 6/1/2011
Citation: Fletcher, R.S., Everitt, J.H., Yang, C. 2011. Identifying saltcedar with hyperspectral data and support vector machines. Geocarto International. 26(3):195-209.
Interpretive Summary: Saltcedar, a dense group of phreatophytic shrubs and trees native to Eurasia, invades riparian regions in the United States and uses excessive amounts of water in these areas. ARS scientists employed ground-based hyperspectral data collected in December 2008 and April 2009 to train a computer learning algorithm (support vector machines) to differentiate saltcedar from vegetative and non-vegetative surfaces. Saltcedar was identified with accuracies ranging from 95% to 100%. The findings support further exploration of airborne hyperspectral data and machine learning algorithms to separate saltcedar from other cover types and endorse collecting data at other times of year and growth stages of saltcedar for plant identification.
Technical Abstract: Saltcedar (Tamarix spp.) are a group of dense phreatophytic shrubs and trees that are invasive to riparian areas throughout the United States. This study determined the feasibility of using hyperspectral data and a support vector machine (SVM) classifier to discriminate saltcedar from other cover types in west Texas. Spectral measurements were collected with a ground-based hyperspectral spectroradiometer (350-2500 nm spectral range) in December 2008 and April 2009. The optimum classification model was developed using a linear SVM kernel. It identified saltcedar with accuracies ranging from 95% to 100%. Findings of this study support further exploration of hyperspectral remote sensing technology and SVM classifiers for differentiating saltcedar from other cover types.