Skip to main content
ARS Home » Research » Publications at this Location » Publication #199624

Title: Hyperspectral remote sensing and geospatial modeling for monitoring invasive plant species

item Hunt, Earle - Ray
item Gillham, John

Submitted to: Meeting Abstract
Publication Type: Abstract Only
Publication Acceptance Date: 11/14/2006
Publication Date: 2/12/2007
Citation: Hunt, E.R., Gilham, J.H. 2007. Hyperspectral remote sensing and geospatial modeling for montoring invasive plant species [abstract]. Proceedings of the 60th Annual Meeting of the Society for Range Management. 2007 CDROM.

Interpretive Summary:

Technical Abstract: Remote sensing is used to show the actual distribution of distinctive invasive weeds such as leafy spurge (Euphorbia esula L.), whereas landscape modeling can show the potential distribution over an area. Geographic information system data and hyperspectral imagery [NASA JPL’s Airborne Visible Infrared Imaging Spectrometer (AVIRIS)] were collected for Devils Tower National Monument in northeastern Wyoming, USA. Leafy spurge was detected in the AVIRIS imagery using the Spectral Angle Mapper with a 74% overall accuracy. The areas of leafy spurge presence and absence were compared to the predictions of the Weed Invasion Susceptibility Prediction (WISP) model. Over the area of the AVIRIS imagery, about 8% of the landscape was covered by leafy spurge, whereas 23% of the landscape has the potential to be invaded. Using kappa analysis, the agreement between remote sensing and landscape modeling was 30%, which was significantly less than expected by chance, indicating model errors. Detailed analysis of individual data layers showed that only a few of the predictor variables were required. Elimination of non-significant predictor variables reduced the area predicted to be susceptible to 13%, and increased the accuracy of the predictions to 81%. Remote sensing was a powerful addition to landscape modeling because the entire landscape was used for the analysis, increasing its statistical power, whereas field data collection would be limited in scope and would be more costly.