Inventory and Survey Methods for Nonindigenous Plant Species (PDF)

Chapter 6 • Stratified Random Sampling Methodlogging), and elevation (an automatic field in the GPS), aswell as other fields recommended by the North AmericanWeed Mapping Association (NAWMA 2003) (see Chapter 2,Table 1). Many of the NAWMA fields can be entered later inthe office.Data are then downloaded from the GPS unit to a maincomputer and imported into a GIS. By collecting NIS presenceand patch length, and changes in habitat and dominantnative species in the field, but no absence data, time spenton any one transect is minimized. Because the start, end,and trajectory of each transect is known, we can use thedata to generate continuous NIS location data (presence andabsence) using extensions created in ArcView (version 3.2;ESRI, Inc.) and macros in Excel (Microsoft Corp.).Predicting NIS Occurrence UsingStratified Random Sampling DataWhile it is good to have a map of where a particularspecies is along each transect sampled (Figure 1), these datacan be used to provide much more information (Figure 4),as well as predictions for the whole management area(Figure 5). The initial approach, and the simplest, is tocreate graphs from the field and GIS data to visualize howtarget NIS populations are correlated with specific variablesmeasured in the field (Figure 4). If it is not possible to developpredictive models, the type of graphs depicted in Figure 4could be used to direct and prioritize further survey activitiesbecause they demonstrate the environments in which targetspecies are more likely or less likely to be observed.However, we can also use the field and GIS data togenerate probability of occurrence maps for individualNIS (Figure 5). Such maps are for the entire managementarea and not just the area sampled. This is important forlarge management areas where it will be impossible to eversample the area entirely but where NIS management needsto address the entire area.Using the data we collected in the field and processedin the office we have presence/absence data for each targetNIS along every transect. We use 10-m by 10-m data cellsbecause that fits with the GIS data layers, but other cell sizescan be used. For each 10-m cell of NIS occurrence we alsohave information on factors such as distance to nearest roadand trail, aspect, elevation, and slope, which we obtain fromthe GIS data layers. These data can then be analyzed using alogistic regression model to predict the occurrence of targetNIS for the entire area. The model takes the form:where y = target NIS, and xj = the variables that the targetspecies could be correlated with, including slope, elevation,distance from road, distance from trail, and vegetationhabitat type. The best model—that is, the one whichbest predicts the distribution of the target NIS—can bedetermined using a formula called the Akaike InformationCriterion. The results or coefficients from the model canthen be used to generate a probability of occurrence map ofthe target species (see Rew et al. 2005 for more detail). Theonly limitation is that the factors put in the model must beones for which there is a GIS layer of the whole area. Forexample, if there is no GIS layer which depicts habitat for themanagement area as a whole, the habitat information wouldbe used descriptively as discussed above (e.g., Figure 4), but▲ Figure 4. Proportional rose diagrams demonstrate that different target species have higher occurrence on some aspectsthan others. Similar graphs could be produced to demonstrate different slope, elevation, soil type, etc., preferences. Data arefrom three years of sampling in the Northern Range of Yellowstone National Park. Red, Dalmatian toadflax (Linaria dalmatica);green, timothy (Phleum pratense); blue, Canada thistle (Cirsium arvense).53Inventory and Survey Methods for Nonindigenous Plant Species