Results: overview of analyses

I will take a three-pronged approach to further analysing this question:

1. Ordinations: I used ordinations to further explore patterns in my data. The ordination presented in the previous section appears to be largely influenced by a handful of outlier plots; I removed these to see if they are obscuring other trends in the data. I also used a constrained ordination to see if the predictor variable or covariables are having notable effects on species composition.

2. Continued investigation into species trends: It appears that looking at the distribution of individual species in relation to distance from the trail may not yield particularly interesting results. Instead of pursuing this route further, I have decided to look into whether different groups of species respond differently to trampling. These groups are biologically-relevant ones, such as family and life form. 

3. Regressions: Regressions were done to test the significance of interesting trends found in my exploratory analyses.
 

Results: vascular composition

Species composition along the distance gradient

After running the initial NMDS ordination, I looked into the species composition of eight plots which I identified as outliers to find out why they were so dissimilar from other plots in the ordination. I found that they were all dominated by species that were not common components of my target community (i.e.: Dryas meadows). These plots appeared to come a different community type, and, because I am focusing on the response of Dryas meadows in this study, I decided to remove them from the dataset and rerun the ordination. Here is the resulting plot.

FIgure: NMDS ordination

In order to interpret this ordination, I would like to draw the viewer's attention to two distinct areas of the graph: the tightly clumped area near the centre of the plot, and the scatter that surrounds it. A look at the species composition of the plots located in the tightly clumped area showed that they all share what I would consider to be the typical composition for plots in my dataset. In contrast, the composition of the plots that make up the scatter tend to have one or more species that are showing an atypical response (i.e.: they are either more or less abundant than usual). The same gradient that was present in the first NMDS I ran is still visible here: edge plots are more on the left side of the graph, while 50m plots are located more on the right side. However, there is not much spread between plots of various distances, and most of the spread in this graph is accounted for by the scattered plots. The fact that spread in the plot is driven by more differences in one or two species per plot than by distance from the trail suggests that human impact is not the major driver of species composition on the Divide.

Species group responses to trampling

Because there is some degree of separation between the plots at different distances from the trail, I decided to look more closely into univariate species responses to see if there are species or subsets of species that are responding to human use in the area. My initial analysis of my five most common species indicated that looking at individual species responses may not be the most efficient way of going about this. Instead, I have decided to divide species into ecologically relevant groups and look for responses of these groups to the distance gradient. 

I first divided my species into family groups. I also divided them into a number of morphological categories, because previous research has shown that plants' morphology strongly influences their ability to tolerate trampling (e.g.: Cole 1995). I divided all my species by growth form (tufted, matted, cushion, rosette or upright), by Raunkiaer life form (cryptophyte, hemicryptophyte, chamaephyte or therophyte) and by physiological group (shrub, form, or graminoid) and graphed the responses to see if any of these groups showed interesting trends. 

Only two of the groups I looked into showed any sort of clear response to the distance gradient: sedges and graminoids (Figure 13). Because the vast majority of graminoids in my off-trail plots are sedges, these two groups are essentially showing the same pattern. 

Figure 13: Regression of graminoids along the distance gradient

Figure 13 is a regression of graminoids against distance from trail. The best fit line for this relationship only accounted for 12% of the variation, which seems low even for ecological data. However, there was a clear decrease in graminoid cover with increasing distance from the trail. Given that graminoids have been found to be among the most trampling-tolerant life forms, this is not surprising.

Constrained ordination of species data

The final analysis I chose to do in this section is a constrained ordination, to look into the effects of my predictor variable and covariables. I used the distance-based redundancy analysis method (dbRDA; Legendre & Anderson 1999) for this analysis. dbRDA is well suited to analyse community data, because unlike regular constrained ordination methods it allows for the use of different types of distance measures. I chose to use the Bray-Curtis distance measure once again. Distance from trail, soil compaction and aspect were all included in the model as predictor variables. I did not include slope, because it was measured at the transect level and I believe that this might obscure differences between plots in a single transect (aspect, on the other hand, was generally consistent within transects). Only distance and compaction came out as significant variables; as a result, aspect was left out of the final biplot (Figure 14).

Figure 14: dbRDA ordination

This ordination shows some degree of separation between groups (distance categories). However, they appear to be separating more along the distance gradient than along the soil compaction gradient. This suggests that changes in species composition along the transects may be driven more by natural variations in the environment, such as changes in slope or topography, than they are by human impact. 

Results: non-vascular and abiotic composition

I used linear regressions to look into the distributions of the four non-vascular cover types that were presented in the preliminary analyses section.

Figure 15: regressions of non-vascular cover types

In these four regressions, the lichen is clearly responding most strongly. The other three cover types have fairly low r-squared values. In the cases of rock and soil, I believe that part of the reason for these low values is that these cover types are extremely variable in alpine tundra ecosystems. Frost disturbance leads to the formation of naturally gravelly areas that have a high proportion of exposed rock and soil. In addition, rocks of various sizes are scattered across the tundra. As a result there is a lot of "noise" in these two cover types, which may be obscuring an ecologically relevant "signal". 

It is also possible that the patterns shown by these cover types would be better described with a non-linear regression, but it is difficult to know this without having data in the 15-49m range. 

The extremely low r-squared value and the very high AIC value for the litter regression suggest that any trend that litter is showing is likely to be quite weak. Unlike rock and soil, I do not see any reason why litter would show a high amount of natural variability in the tundra. I do not think this regression is showing an ecologically important pattern.

Lichen, on the other hand, seem to be responding quite strongly to the distance gradient. 0.26 is a relatively high r-squared for vegetation cover data, and lichens were also found to be strongly correlated with soil compaction (rho=-0.21; p=0.001). Previous studies have found lichens to be particularly sensitive to trampling (e.g.: Bayfield, Urquhart & Cooper 1981). Given all this, I would conclude that lichens are indeed showing a response to human impact on Cardinal Divide.