Comparing Three Sampling Strategies in the Snyder-Middleswarth Natural Area

This was a highly relevant exercise to a self-paced online ecology class, and I am thankful that Bucknell University produced it. Note that I sampled 25 plots, not 24 as the instructions stated, as the systematic sampling along the topographical gradient actually produced 25 samples. Ultimately, the questions posed for this blog post do not have simple answers; it is up to the designer of the experiment to select the best sampling method given time constraints and relative abundance of selected species.

Question 1: Which method is the most time-efficient?

According to the Results Summary (see Verona Virtual Forest Data Summary), systematic was estimated to require 12:35:00, random 13:17:00, and haphazard 13:05:00. While somewhat useful, these estimates should only be used as a guide; actual field sampling will likely have vastly different sampling times, as discussed below.

Had sampling occurred in the field, the systematic method would be the most expedient with one caveat: it would require drawing out the transect on the ground over a long physical distance with topographical features. This could be laid out using logger’s tape, a compass bearing, GPS unit, and brightly-colored metal stakes. Once established, though, it would not require hiking across the unit to find random plots via GPS coordinates, or select haphazard ones. A sampler would simply “bump” along the transect, plot-to-plot. Systematic would also allow for quick re-sampling for a seasonally-dependent hypothesis/prediction. The haphazard plots could be particularly time-consuming for the first (or only) iteration as the sampler would need to hike (i.e., bush-whack) the entire unit and subjectively determine which plots would be “better” than others, returning to those sites after selection. For the purposes of this exercise, haphazard plots were chosen based on higher perceived tree densities with an aerial view.

Question 2: Compare errors of each sampling method for two most common and rarest species. Which was the most accurate for common species? For rare species? Did the accuracy decline for rare species?

For percentage error calculations, see Verona Virtual Forest Error Calcs (note that the Excel formula will not be visible, but is the same as the one listed in the tutorial). Species relative abundance (for selection of common or rare) was determined using the actual relative densities. Species density was chosen for percentage error calculations, because  it is a direct function of the total number of a given species over the 7.5 ha project area. It was also selected because I will likely use density and relative density for sampling fern species in my cedar grove research project. The percentage errors are summarized in Table 1.

Table 1: Error percentages of two most common (C1 & C2) and two rarest (R1 & R2) species, based on sampling method (systematic, random, or haphazard). Note that errors are listed as the absolute value of a percent of relative error between true and observed values.

Clearly, systematic sampling along a transect is the most accurate for common species. Additionally, systematic had an error of 8.6% for striped maple, one of the two rarest species. While this was much lower than the random and haphazard errors for striped maple, it may not indicate preference for systematic sampling, as the start of the transect was selected randomly. The transect could have been placed through an area of high maple density instead, for example. Depending on the spacing of maple throughout the entire area (i.e., its preference to grow in clusters or as single trees), systematic may be ideal for this rare species – another two or three transects may offer better data for using systematic sampling over random or haphazard. If a sampler was focused on striped maple, 25 plots would likely be too low a plot count; he or she may consider three transects totaling 75 plots.

For the rarest species, white pine, the only method to record any trees was haphazard, and even it had a sizeable error of 138.1%. While this was technically higher than systematic and random errors (both 100%), their values should not be considered as they recorded zero white pines for the entire project area. If systematic or random sampling were the only data in the study, one may have concluded that white pine was not present. If a researcher was concerned primarily with white pine, of which there were 63 individuals in this 7.5 ha area, he or she might consider a haphazard approach for selecting “micro-plots” of areas with greater white pine density, wherein a more systematic approach could be used. Based on the fact that this sampler selected areas of perceived greatest stem density for haphazard sampling, perhaps white pine are only present in more environmentally-favorable areas, exhibit a preference for the shade of surrounding trees, or grow in population clusters. White pine were not the largest trees (actual dominance value: 0.9).

Question 3: Were 25 plots adequate to capture the number of species and their abundance?

The systematic transect method returned fairly low errors across the board, but the fact that it recorded zero white pines fails the method for accurately reporting the number of species. In this regard, 25 plots would be too few to accurately sample the 7.5 ha project area. As previously stated, three transects would provide a better view of the amount of species and their relative abundance – this would require 75 plots and about 40 hours of work according to the program’s estimate, which could be ideal for a typical work week in the field.