Accuracy assessment for detection of leafy spurge with hyperspectral imagery

Abstract

When flowering, leafy spurge (Euphorbia esula L.) has conspicuous yellow-green bracts that are spectrally distinct from other vegetation and may be distinguished with hyperspectral remote sensing. In July 1999, Airborne Visible Infrared Imaging Spectrometer (AVIRIS) data were acquired in northeastern Wyoming, near Devils Tower National Monument. Using the reflectance spectrum of flowering leafy spurge, leafy spurge occurrence was determined using a new method of spectral mixture analysis, Mixture Tuned Matched Filtering (MTMF). Ground reference data (146 sites) were obtained 2 weeks before and after AVIRIS overflight to test the classification accuracy of leafy spurge. For 3 land cover types: mixed prairie, riparian, and coniferous woodlands, the presence or absence leafy spurge was detected with an overall accuracy of 95% using a 0.10 threshold for detection. Differences in classification thresholds resulted in a trade-off between false positives, pixels that were mapped as leafy spurge but did not contain leafy spurge on the ground, and false negatives, areas that had leafy spurge on the ground but were not mapped as leafy spurge. Detection of leafy spurge occurrence was best for mixed prairie and riparian cover types, and somewhat less successful for conifer woodlands because of interference from tree crowns and their shadows. The advantage of the MTMF technique is it allows automated processing of hyperspectral imagery to generate accurate maps of leafy spurge occurrence.