Biology Reference
In-Depth Information
southwestern Idaho, evaluating their classified surface using Incremental Cover
Evaluation. With this method, accuracy was calculated for different levels of cover
(e.g., greater than 10%) to determine what percent cover of hoary cress was
required for it to be accurately detected. They were able to detect it with 0-10%
cover, but only correctly identified 5 of 19 locations. Thirty percent cover was
required for the accuracy to be useful to management applications (82% accu-
racy). Similarly, Glenn et al. (2005) detected small infestations of leafy spurge
(
Eupohorbia esula
) in the Swan Valley in Idaho using 3.5-m resolution hyperspec-
tral data, with the species detectable at 10% cover but for repeatability the dis-
crimination threshold was around 40% cover. So, even in this relatively small area,
detecting new infestations while they are small is still problematic. The costs asso-
ciated with high resolution data over areas greater than a few hectares currently
makes it prohibitive to use remote sensing for early detection (Shaw 2005), even
if it could be used to detect species when they have very low cover.
6.3.5 Statistics with Remote Sensing
Several statistical methods can incorporate remotely-sensed imagery layers or
derived layers like Normalized Difference Vegetation Index (NDVI) as parameters
in models. These types of models could be used to determine areas most likely to
be invaded by a particular species. These models could then direct field surveys to
areas with a high probability of invasion for early detection efforts or to areas with
high levels of uncertainty.
Remotely-sensed imagery products may also capture temporal variability impor-
tant in predicting invasions. For example, NASA researchers created three derived
products from MODIS satellite data that captured annual variability in NDVI,
including the range in greenness throughout a year, the mean NDVI value, and the
average date of green up for a pixel (Morisette et al. 2006). These products
improved models for habitat suitability.
6.3.6 The Future
None of the techniques we found addressed all of the considerations we dis-
cussed in the previous section (Table 6.2). Each technique has it pros and cons,
and may be useful to answer different questions. Many of the techniques are
time and labor intensive, and often require extensive previous knowledge of a
species. We need to become better at making this information easily available to
other researchers and at making our models more sophisticated, incorporating
considerations of time since invasion, invasion stage, issues related to scale, and
other important factors.
