Discussion

The original Bavarian classification tree identified five ecological factors important in the distribution of P. alni. Where these factors occur together in the environment, the likelihood of infection is increased. Specifically, where silt minimum values are less than 20 percent, and sand mean values are less than 5 percent, the probability of a P. alni infection is high. When silt minimum values are less than 20 percent, and sand means are greater than 5 percent, the site is more likely to have P. alni infections if slopes are less than 2.97 degrees and have warmer aspects. Sites with a landform index measure of less than 6.6 (concave, flat or slightly convex) also have an increased probability of a P. alni infestation. These results make biological sense. Areas with poor drainage and warmer aspects provide an optimal environment for the pathogen to flourish, as will sites with fairly flat or concave physical structure. Conversely, areas with less clay and more silt or sand will drain better, as will sites with steeper slopes and convex landform. These types of sites will not provide a wet environment for this waterborne pathogen to form sporangia and release zoospores that are essential for the spread and infection of P. alni.

Not all of the five ecological factors identified as being important by the first Bavarian model for predicting the distribution of P. alni were available for the global model. A second model for Bavaria, which utilized only the three datasets that were available globally, demonstrated the limitations of modeling invasive species at a global scale without appropriate data. The limitation most notable was the soil texture data because it was selected by the first Bavarian model as the most important variable for predicting the presence and absence of the soil-bourn pathogen P. alni. In addition, data was not available for: (1) forest species type, (i.e., distribution of the individual alder species), and (2) susceptibility of North- and South American and Asian alder species to P. alni. We addressed forest species type by keeping our analysis near and around streams and floodplains where most alders tend to grow. We also looked at the temperature range, eliminating areas with temperatures that were too cold or hot for alder and P. alni survival. Although we made compromises to work within the data limitations, this work emphasizes the need for quality spatial environmental data at the global scale.

Since planting infected nursery stock is one of the primary pathways by which P. alni has been spread, we were careful to consider the social or cultural habits in association with outplanting alder trees. Of particular concern was the outplanting of infected alder trees in respect to elevation. At higher elevations, alder trees are planted much less frequently than at lower elevations. Yet, it has been observed by the Jung that where P. alni-infected alder was planted at higher elevations, those sites have become infested and further contribute to infections downhill and downstream. Because alder was rarely planted at higher elevations, P. alni was much less prevalent on higher elevation sites. We therefore assumed that the model would be biased toward selecting elevation as an important variable for predicting presence and absence. Consequently, elevation was not used in the model.

A higher accuracy was attained for predicting the P. alni-infested sites than for predicting healthy sites. This is likely an outcome of having three times more infested than healthy sample locations. Had we sampled a greater number of locations for the healthy condition, it is likely that the accuracy for predicting healthy sites would improve.

Of the 127 healthy alder tree locations collected between 2003 and 2006, some sites may have changed in status. Some of the sites that were not infested by 2006 may become infested in the future. These are problems one would expect in attempting to model a species that is unlikely to have been in existence before the 1980s (Gibbs and others 2003, Jung and Blaschke 2004) and has not yet completely expanded into its potential range. With no complete range map for P. alni, the Bavarian model provides managers worldwide with useful decision rules and a data mining tool for estimating the susceptibility of their resources to P. alni.

Because all of the applicable variables from the first Bavarian model are available in datasets for the United States, the extrapolation of the Bavarian model to forests in the United States should demonstrate the specific improvement that can be gained by applying the appropriate datasets identified by the Bavarian model.