Using remote sensing to map rhododendron: results from Held and lab trials

摘要

Compared with traditional ground surveys, remote sensing has the potential to map the extent of non-native invasive species in the UK rapidly and reliably. This paper assesses the potential of radiospectrometry to distinguish and characterise the status of invasive non-native rhododendron (Rhododendron ponticum). Absolute reflectance of target plant material was measured under standardised lab conditions and in the field to characterise spectral signature in the winter, during leaf-off conditions of woodland overstory, and in the summer when mature rhododendrons are flowering. Logistic regression was used to determine success of discriminating rhododendron from other target species (evergreen laurel, holly and ivy, and dead retained beech leaves) likely to be encountered in woodlands during the winter, and distinguish flowers from leaves. The species logistic regression model was highly significant (p0.0001), with 90.3% of target plant material correctly identified as rhododendron or non-rhododendron. Rescaling the data to emulate airborne and satellite acquired data only decreased the total success rate of correctly identifying rhododendron by 1.4%. At a wavelength of 450nm, the reproductive logistic regression model had an 80% success rate of distinguishing flowers from leaves (p<0.05). This demonstrates the potential to map both bush presence and reproductive status using airborne hyperspectral data. Such information is critical to the development of successful strategic management plans to effectively eradicate rhododendron from a site. The next step will be to map rhododendron using hyperspectral data captured in March 2010 during leaf-off conditions in a study test site at Coed y Brenin, Wales.