1. Non-metric Multi-dimensional Scaling (NMDS): How do the communities re-surveyed in 2007 compare to their original composition?
In this case, NMDS was used to visualize the change in composition from the first survey period (70/80's) to the second (2007). This analysis followed the same protocol as the initial NMDS described in the previous page (using PCA starting co-ordinates in 3-dimensions and Bray-Curtis distance measure).
It appears that the two sites surveyed in the "02-poor" series shifted downwards on Axis 3, whereas the sites in the "03-mesic" series shifted in opposite directions along Axis 1. It is unclear what the shift downwards in Axis 3 is driven by, but it does appear that a uniform shift in "02" sites towards the "03" sites has occurred. Increased sampling effort will hopefully clarify the trends for both site types.
2. Discriminant analysis: Classify the 4 sites re-surveyed in 2007 according to the composition of sites ("01" "02" "03) from the initial survey period
Discriminant analysis is a useful test to classify new observations based on previously classified data (McCune and Grace 2002). In this case, I used discriminant analysis as another technique to look at the change in community composition over time. Using the groups defined by site series ('01" "02" "03") from the first survey period (1970/80's) the 4 sites re-surveyed sites from "02-poor" and "03-mesic" was classified based on species composition and abundance in relation to the assigned classification of original sites. In other words, the repeated measures data was compared to the first time period community composition with the question: What site series should the 4 new data points be classified as based on this species data? It appears that one of the sites re-surveyed in 2007 in the "02-poor" would be re-classified as "03-mesic". This site is the "lower" point on the above NMDS of the two "02" sites that were surveyed in 2007.
Support for a new hypothesis:
This series of analyses may indicate that a fourth hypothesis exists where "02-poor" sites are shifting towards "03-mesic" sites in species composition, while "03-mesic" sites are remaining relatively unchanged. This would imply a reduction in landscape Beta-diversity with the "02" and "03" sites becoming more uniform, but still dissimilar to the wetter "01-rich" sites. However, the small sample size of re-surveyed sites may not allow further speculation. For instance, the response of "03" sites to disturbance is unclear due to the opposite response of the two sites, so the direction of the "03" arrow in the hypothesis diagram to the right may be moving in a direction that is not detectable to date. Further sampling will hopefully shed light on these responses.
3. Multiple Regression Tree (MRT) - Can any of the environmental parameters help explain composition patterns based on the initial survey period?
Multiple regression trees perform a similar function as the CART analysis described in the previous page, but in this case, the species-environment relationship explored is based on community response data (Breiman et al 1984, De'ath 2002). In this case, I examined if environmental parameters could explain the variation in communities in the 1970/80's dataset.
The results indicate that elevation was the first criteria for differentiating communities, with those at elevations above 1000m on the left of the tree, explaining 11 sites. Sites above 1000m in elevation were further separated by mineral soil, those with exposed mineral soil covering more the 0.5% of the ground going to the left (3 sites) and those with less then 0.5% mineral soil going to the right (5 sites).
MRT was not effective in predicting communities, with only 18% of the total variation explained by this tree (see table left). However, this is likely due to the factors included and not due to an error in the analysis or this technique. This tree was built with data from the first survey period in order to use the 2007 data as 'test data', but due to the lack of explanatory power in the overall MRT, this next step was not informative.
4. Indicator species analysis (ISA): What species characterize each terminal node in the MRT above?
Indicator species analysis (McCune and Grace 2002) as described in the previous page is effective at identifying species that are common and "faithful" to specified groups. Here, I examined which species characterize the terminal nodes of the above MRT to make sense of which species are common in sites defined by these environmental parameters.
Elevation indicator species
Mineral soil indicator species
Elevation:
Group 1 in the elevation indicator species analysis refers to the sites in the far-left leaf on the tree - defined by elevations of greater than 1000m. This group is characterized by the moss pleurozium schreberi, Pinus contorta and Empetrum nigrum. These species cover a large range of moisture conditions in the northern ESSF, from dry to wet, but they all tend towards slightly poor nutrient conditions (pleurozium is an exception as it may be present across a wide range of site conditions).
Mineral Soil:
The sites occurring at less than 1000m in elevation are split by mineral soil. The leaf on the far right is defined by mineral soil covering less than 0.5% of a plot. These sites are characterized by Abies lasciocarpa, which also occur across a wide range of moisture and nutrient regimes. The sites with mineral soil covering greater than 0.5% of plots are defined by Pinus albicaulis, Cassiope mertensiana, Empetrum nigrum, Racomitrium canescens and Cladina/Cladonia sp. lichens. These species are largely found on dry/nutrient poor sites, with some being characteristic of only dry sites or only nutrient poor sites exclusively (Beaudry et al 1999).
Overall the indicator species analysis provides insight into what species commonly occur at each of the terminal nodes. However, given the lack of fit of the tree model to the species data, the indicator species analysis may not provide much additional information in understanding how the environmental parameters are driving community composition at the different sites.
References:
Breiman, Friedman, Olshen, and Stone, 1984. Classification and Regression Trees. Wadsworth.
Beaudry, L., Coupe, R., Delong, C. and Pojar, J. 1999. Plant indicator guide for northern British Columbia: Boreal, Sub-boreal and Subalpine biogeoclimatic zones. Province of British Columbia. Crown publications, Victoria, BC
De'ath G., 2002. Multivariate Regression Trees : A New Technique for Constrained Classification Analysis. Ecology 83, 1103-1117.
McCune, B. and Grace, J.B. 2002. Analysis of ecological communities. MJM Software Design, Gleneden Beach, Oregon
McCune, B. and Mefford., M. J., 1999. PC-ORD. Multivariate Analysis of Ecological Data., Version 5.0, MjM Software, Gleneden Beach, Oregon, U.S.A.