1. Univariate response: How has Pinus albicaulis (Whitebark Pine) abundance changed after multiple disturbance?
Univariate graphical exploratory analysis was used to determine some general trends in the dataset, especially in comparing the two time periods. However the following three figures (Pinus albicaulis, total bryophytes and total lichens) should not be considered as having statistically significant trends due to the small repeated sample size.
This first figure indicates that Whitebark pine (Pinus albicaulis) total abundance appears to be decreasing in both the "02-poor" and "03-mesic" sites. This graph illustrates "all the available data" (23 sites: '19'=1970/80's; '4'=2007) on the left, and just the sites in the "02" and "03" sites series that were repeated in time in 2007 on the right (8 sites, see methods). This provides a good overview of changes in abundance over time in the context of all the site data available (left) as well as a focused look at how the subset of sites that were re-surveyed changed over time (right). The decrease in P. albicaulis in the "02" and "03" sites may not be surprising as the disturbances impacting this ecosystem are targeted towards this species.
Support for hypothesis 1:
A decrease in Pinus albicaulis may indicate a shift in communities towards the more common "01-rich" site type.
2. Univariate response: How has total bryophyte abundance changed after multiple disturbances?
This figure also contains "all data" on the left to indicate how the changes over time relate to the overall dataset, and just the repeated sites (right) to show the response of the actual sites re-surveyed post-disturbance. The total mean abundance of bryophytes appears to be increasing in the "03-mesic" sites in 2007 and decreasing in the "02-poor" sites. However a small sample size, makes it difficult to find strong, significant trends.
Support for hypothesis 3
A decrease in total bryophyte abundance may amplify the difference between an "02-poor" and an "03-mesic" commuity, wiht "03-mesic" shifting towards the more common "01-rich". The data above may also support yet another alternative hypothesis where the "03-mesic" and "02-poor" site types are shifting away from each other, but neither one moving towards "01-rich".
3. Univariate response: How has total lichen abundance changed after multiple disturbances?
The total abundance of lichen appears to be following a similar, but opposite trend to bryophytes, with increasing abundance in the "02" and decreasing in "03". This would support hypothesis 3, with "03" moving towards "01" composition and "02" maintaining, if not amplifying the lichen cover that may distinguish it from other sites.
Support for hypothesis 3:
Increased lichen cover may amplify the difference between "02-poor" and "03-mesic" sites, maintaining an ecologically unique site type ("02") on the landscape.
4. CART (Classification and Regression Trees): What environmental parameters explain the abundance patterns seen in Pinus albicaulis, total bryophytes and total lichen from the original survey period (1970/80's)?
Pinus albicaulis regression tree
Total bryophyte regression tree
Total lichen regression tree
Regression tree analysis (Breiman et al 1984, Ripley 1996) is an effective tool for predictive modeling of species-environment relationships (De'ath 2002). Here, I present CART output analyzed in 'R' (ver 2.8.1) to examine the relationship between select summary data and environmental parameters.
All graphs are presented with the environmental criteria at each node and the number of sites of each site type (colour-coded - see methods) that occur at each terminal node. The number beneath the node indicates the mean abundance (of the species) present in that leaf.
Pinus albicaulis abundance was best described by Mineral soil and rooting depth. One "01-rich", 2 "03-mesic" and 3 "02-poor" sites had the highest mean P. albicaulis abundance. These sites are described as having mineral soil covering greater then 0.5% of the plot. This result appears to be contrary to the means described above as P.albicaulis was virtually absent from "01" sites. the Table below indicates that this regression tree explains 3% of the variation in the P. albicaulis dataset, which likely explains odd classifications. This tree would be ineffective as a predictor of P. albicaulis abundance based on these environmental parameters. The other side of the figure indicates that the second node is defined by rooting depth; sites with depth greater then 47.5cm having the least amount of P. albicaulis (mean=0.9%) and sites with depth less then 47.5cm (on the left) having a moderate amount of P. albicaulis (7.6%).
The next two figures include both environmental data and overstory species as predictive parameters as tree cover would greatly influence species such as lichens and bryophytes.
The second figure illustrates the regression tree for total bryophytes. The table below indicates that the variation explained by the chosen parameters is 22%, and this is being driven by Abies lasciocarpa (Subalpine fir). When fir abundance is less then 21%, there is the least amount of bryophytes present in the understory (mean=31%). When fir covers greater then 21% of the plot, but less than 51%, there is a moderate abundance of bryophytes (mean=51%). Finally if fir covers greater than 51% of the plot, there is the highest amount of bryophyte cover. This analysis indicates that the "01-rich" and "03-mesic" sites may have a variable amount of bryophytes, but in general seem to have higher bryophyte cover then the "02-poor sites. However, the low variation explained by this regression tree does not provide confidence in the predictive power of these results.
Finally, total lichen cover was related to environmental and overstory parameters resulting in the 3rd tree diagram to the left. These results indicate that the highest abundance of lichen occurs in sites that have greater than 12% P. albicaulis cover (mean abundance=38%). A moderate abundance of lichens occur at sites that have less than 12% P. albicaulis cover, but have upper or crest meso-slope positions (mean=8%) and the least amount of lichen cover occurs on sites that have less then 12% P. albicaulis cover and lower, mid and top meso-slope positions (mean=1%). Further, the "01-rich" sites occur exclusively in the left-most terminal node where lichen abundance is lowest, whereas the "02-poor" and "03-mesic" sites occur exclusively withmoderate to high mean lichen abundance as defined by greater P. albicaulis cover or exposed slope positions. Again, these parameters explain only 26% of the variation in the total lichen dataset, so may not be an effective tool for predicting lichen cover
5. Predictions from Regression trees: Can the regression trees adequately predict the abundance patterns of P. albicaulis, total bryophytes and total lichens seen in the 2nd survey period (2007)?
Overall, the univariate regression trees were not effective in explaining a significant amount of variation in Pinus albicaulis, total bryophyte and total lichen abundance. However, the regression tree analysis was also tested for predictive power by using the repeated measures sites as 'test data'. This did not result in accurate predictions, with sites in each tree being mis-classified based on the parameters identified by the regression tree analysis.
For now, it appears that important parameters defining the abundance of these species and groups of species have not been included in the analysis and therefore more work is needed to identify better predictors.
References:
Breiman L., Friedman J. H., Olshen R. A., and Stone, C. J. (1984) Classification and Regression Trees. Wadsworth.
De'ath, G. 2002. Multivariate regression trees: A new technique for modeling species-environment relationships. Ecology, 83: 1150-1117
R version 2.8.1 (2008-12-22). Copyright (C) 2008 The R Foundation for Statistical Computing
Ripley, B. D. (1996) Pattern Recognition and Neural Networks. Cambridge University Press, Cambridge. Chapter 7.