Last week, I decided to check out the Army Corps website to see if they have updated their wetland plant list for 2018. There, I found a link to a presentation that sets forth an approach for a major revamping of wetland indicator statuses. http://wetland-plants.usace.army.mil/nwpl_static/home/home.html
They are proposing using annotated herbarium data to get a more objective idea of species affinities for wetland versus non-wetlands sites. Huzzah! I like plants. I like numbers. I like the idea.
However, as I watched the presentation, a HUGE problem became obvious. To explain, I have to take a step back and talk about wetland indicator statuses. For each species, these are supposed to describe the fidelity to wetland conditions. In the past, these were associated with numerical values, but they have been more recently been revised to be more qualitative.
It turns out that the Army Corps is proposing to rely heavily on a study that focused on blue spruce in Colorado. In that study, they determined the percentage occurrence of blue spruce in the field, compared it to the “old school” indicator numerical values, and drew a conclusion.
In that study , wetland delineation indicators are used around a plot-based rather than a plant-based approach. FAC plants are roughly equally likely (34-66%) to occur on wetland sample plots vs. non-wetland plots (they have no affinity for wet vs. not). If this concept is detached from the plots and applied to individual plants on a landscape, it is biased by the proportion of land that is wetland vs. upland. In other words, the percentage value you determine by investigating individual plants and whether or not they are growing in a wetland (like the spruce) becomes a direct function of the percentage of the landscape that is wetland, which disconnects the observation from the biology of the plant.
I use pooled data (numbers of trees in wetlands and non-wetlands) from the study (paywalled, so I had to request a copy) and tested a null hypothesis that blue spruce is borderline facultative (34% of occurrences expected in wetland sample plots), given several possible values for landscape wetland coverage. I had to guess at these, because I don’t have a value for landscape wetland percentage for your study area.
Figure 1 illustrates how the percentage of plants observed in wetland for a borderline FAC species varies depending on the percentage of the landscape that is wetland. Looking at this, keep in mind that Colorado wetland extent is ~2% (the study was mostly in Colorado) and Lower 48 wetland extent is ~5%. If half of the landscape is wetland, you expect 34% of occurrences to be in wetland and the rest to be in non-wetland. However, as the proportion of wetland on the landscape gets smaller and smaller, you expect to encounter a smaller and smaller proportion of species occurrences in wetland. This is independent of biology and occurs as an artifact of landscape wetland/non-wetland composition. This is because the number of plants distributed relatively sparsely in uplands overwhelms the numbers occurring relatively densely in wetlands as wetlands become rare. Even as the percentage/proportion of plants in wetlands decreases with the wetlands on the landscape, the chance of finding the species in any single upland plot and any single wetland plot (standard effort) does not change! So surely our spatial interpretation of indicator statuses must be tied plots or units of standardized effort, if we want them to have any biological/ecological meaning.
There is a way to estimate a landscape percentage or proportion of species occurrences and roll those into a useful analysis. However, if that is the most practical approach, we need to take one more step to determine whether or not we’re seeing a species in wetland more or less than is expected. We can test our observed proportion against expected proportions, which can be calculated to take into account the landscape cover of wetlands (Table 1). This is the simplest case of such a test of whether observed values equal expected values, so an appropriate test would be a binomial exact test (a chi-square test is also appropriate, if expected values aren’t too small, but in this case they are, so I went with the binomial test…inference is the same).
Figure 2 is a simple bar graph with the estimated proportion of spruce in wetlands with 95% confidence intervals based on the binomial exact test. Horizontal lines from bottom to top are expected proportions in wetlands for a borderline FAC (34%) species for 2% landscape coverage of wetlands (Colorado) on up 30% (in case the study area is 6 times as wet as the lower 48 as a whole). As you can see, unless the landscape coverage of wetlands is something more than 30%, the data aren’t significantly different from what you would expect from a borderline FAC species, and if the wetland coverage is more in line with Colorado or the lower 48, the study actually observed many times more trees in wetlands than would have expected by chance. In other words, if you went out in the field and established a wetland sample plot (vs. hunting for trees), you would be much more likely to see blue spruce than you would if you established an upland sample plot. Or to put it another way, there is a significant affinity between blue spruce and wetland site conditions.
If wetland covers 30% of the landscape, then the observed proportion of blue spruce in wetland is not significantly different from what would be expected from a borderline FAC (34%) species.
If you want to take this to the extreme, all you have to do imagine a landscape like that of the lower 48 (5%) wetland. A species that occurs in 100% of wetland samples and in only 10% of upland samples would be expected to be encountered on the whole landscape 66% of the time in uplands and 34% of the time in wetlands! It’s worth reiterating, we must either compare percentages when effort is equal between wetland and uplands or adjust our expected values based on landscape wetland coverage.
Botanists can do math. Right?