Category: Post 5: Design Reflections

While implementing my sampling strategy, some difficulties arose when deciding how to ensure randomization when selecting which sample units to measure. I needed to find a sampling method that allowed me to randomly choose a plant by assigning numerical values to put into a random number generator. I originally wanted to divide each strata into a grid of coordinates, but found out quickly that this was not feasible. I found it difficult to make a grid small enough that each box contained only one stem, therefore, I ended up using the distance-based method. I then found that as I was measuring the distances to the selected point, the plants were not fixed in space and so it was difficult to accurately select the correct plant. I then decided to place the measuring tape on the ground and measure to the bottom of the stem to ensure accuracy. 

I found some of the data collected to be surprising, especially in the low degree of sunlight exposure strata. What surprised me was the degree of variation within these strata, especially that of the low degree strata. In this area, most of the plants were under 90 cm tall but there were a few that reached almost 110 cm. Because of this, I decided that five replicates was not enough data to accurately represent the population in each strata. I plan to continue to use the distance-based sampling method with the random number generator, but will generate more replicates so that I have a total of 10 sample unit measurements in total for each strata. I think that increasing the sample size will better represent the sample and decreases the variance within each strata. 

Edit:

I decided to revise my research study, and focus on density of the Canada goldenrod as my response variable. I believe density is a better representation for plant growth success, as it eliminates other factors. Height would not be an accurate representation of plant success, unless I measured plant growth over a period of time. Due to the time constraints of my study, I would not be able to record meaningful data for differential plant growth among goldenrod plants in the different strata. As such, I have revised my study design and sampling methods to measure goldenrod density at different levels of sunlight exposure. To calculate density, I will use a stratified random quadrat sampling method, using the same strata that I initially identified. I will divide my study area into three 10 m by 10 m strata, representing low, moderate, and high levels of sunlight exposure. I will randomly place 10 quadrats in each strata by using a random number generator to generate two coordinates. I will continue generating coordinates until I have 10 points in each stratum. I will then locate these points on my study plot by placing a measuring tape on the ground and measuring the appropriate point, which will represent the bottom left-hand corner of the quadrat.

When implementing my sampling strategy, I found that the process took much longer than expected. Based on the experimental design I chose I utilized stratified random sampling because I needed to have a plot from each “location” along my gradient. I wanted to use an adequate sized plot so I decided to divide my plots into 25 smaller quadrants in an attempt to be more accurate and ease the process of counting the clovers. Unfortunately, the act of splitting the plots took a long time. However, I do think it was beneficial to use the 12 inches by 12 inches quadrants.

The data collected was not overly surprising, I was correct in expecting the highest abundance to be found in the no shade area. However, I was surprised that the clovers observed in the partial shade location were noticeably larger than the clovers from all other locations. They did also seem to be healthier and thriving more compared to the clovers found in other locations.

My only difficulty in implementing my sampling strategy was purely the duration. Although it was not difficult both the counting of the clovers (which were more numerous than I had thought) and the plotting and measuring of the quadrantswere very time consuming.

I will continue to use the same technique because I believe it yielded the most accurate results and feel confident with how my data collection went. If I was to use larger quadrants it would increase my chances of error.

The largest issue found with my sampling strategy would be time constraint. Due to the large number of samples which I wanted to collect in order to create as accurate of a representation as possible, I had to be sure to schedule a full afternoon and evening to be out in the field.

Additionally, occasional thick vegetation growth slowed down data collection due to the method used; pacing at a set compass bearing meant that I had to go through the thick vegetation (being careful to cause little damage/disturbance) rather than around it in order to have the correct number of randomized paces.

The data collected was not very surprising. The only piece that surprised me was how little of the Knapweed grew in the canopied forest; while visual observations suggested this, the samples had virtually no occurrence of the Knapweed.

I do not plan on modifying my approach to collecting data. While significant amount of time is required for my approach, when considering the number of samples taken it is fairly efficient. I think that my data collection methods align well with the hypothesis that I am trying to test.

EDIT: The sampling design was changed to a randomized transect method. Ten transects were randomly chosen using a randomizer app on my cell phone. Transects were sampled every 10 metres for the presence/absence of Knapweed and the cover type at the sample point. This method took roughly 1/3 the time previous method required and provided more sound data to analyze. Similar difficulties with thick bush areas were encountered with this method, however they were not insolvable instances. Similar patterns were noticed with collecting data where Knapweed had a low frequency in the canopied area and a higher frequency in the open grassland. Further statistical analysis is required for the data.

For the initial collection of data, there were no issues encountered. However, in some areas close to the industrial site, I observed a much denser canopy of the nodding onion plant than expected. In the future, I plan to carry out soil and air quality testing of the presence of heavy metals in these areas at proximity to the industrial site. This might explain the higher than expected amount of the nodding onion plant in some patches of land close to the industrial site.  Areas at proximity to the industrial site might have a higher penetrance of sunlight than areas farthest from the industrial site and this might be a plausible explanation for the denser than expected canopy of nodding onions closest to the industrial site. Additionally, wind direction might be of importance and has to be taken into consideration when performing the sampling of plant species in this area. One change that I will perform in the future will be to determine the prevalence of more plant species other than the nodding onion species described above. This will permit to have a more general overview of all species that might be affected by the presence of heavy metals in the soil.