User: | Open Learning Faculty Member:
For my initial data collection, I implemented stratified random sampling using Google Earth and QGIS. I created polygons based on Google Earth satellite images for each predictor zone, exported them as a KML file and used QGIS to generate random points within one of the polygons to collect sample replicates. I then exported these points as a GPX file, put them on my GPS and located them in the park to take samples. I used discrete classes to represent percent coverage as outlined in the sampling design tutorial, ranging from 1-6.
I think my method for generating worked fairly well, but I fear that my areas my be too small to justify stratification. I’m also unsure if statification is the best approach, given that the basis of the stratification is also the predictor variable (dominant tree species as an indirect measure of soil moisture). The zones are fairly distinct in the park, but there are some interspersed wetter/drier sites, leading me to think that perhaps I should use a non-stratified approach and just record the predictor variable with each individual sample. The number of random points which land in one of the smaller zones (arbutus/ garry oak, alder) may be smaller, but since frequency of predictor variable is not a measure of concern it may be ok if I have more samples from the doug-fir/grand fir zone.
Yes, you could measure tree species composition at each site as predictor variable though you would have to come up with how you would categorize the variable or maybe a way it could be continuous? Keep in mind your plots for tree species will likely need to be larger than quadrats for ground vegetation. You could also measure soil moisture at each site as a predictor variable and also record tree species not as a main predictor variable, but extra data to help explain the patterns. Just some options to consider!