Statistical sampling to determine habitat improvement

We are trying to improve the habitat in a defined area of a park or natural area. A specific example would be increasing spring wildflowers and ephemerals. One method of taking a baseline and subsequent census would be to create a plan to take observations in every square foot of the defined area.

Does anyone have a scientifically valid process to take a sample or use randomly collected observations instead? As an example, is there a calculation that would be statistically valid like, “X number of observations within one-half acre (21,780 sq. ft.)” would yield a sample of wildflowers that has a 95% confidence lever +/- 3%.

I think the details of how to effectively sample an area are probably species specific – but it is sure to be a topic that is covered in the ecological literature.

The main thing that leaps out to me is that you don’t mention a control. A lot of methodological challenges become less severe if you subdivide an area and look at how ‘improved’ vs ‘unimproved’ regions compare/contrast in terms of species diversity etc. As long as your control and experimental regions are analyzed the same way, you should be able to measure the impact of your intervention even if the precise relationship of your sample to reality isn’t known. (Ofc then the challenge is dividing up the control and experimental habitats appropriately, but this too is sure to be in the literature.)

I’ll say to begin with I am not a statistician.

I think the question of how many random samples you need depends on how variable the vegetation is to begin with and how small a difference you want to detect. Suppose you take 40 random samples and you divide them into two batches of 20, then you analyse each 20 independently. If the results for each batch are similar enough that you accept they give you the same answer, then I think 20 samples is enough. If the vegetation is so variable that the two batches of 20 look like they came from different plant communities, then you need to have a sample size bigger than 20. And for 20 to be enough to show the effects of management, the difference caused by the management needs to be greater than the difference between your initial two batches of 20 samples.

1 Like

Not a real statistician, but:

  1. I agree with jhbratton’s point about looking at how variable your plant community is. You can use statistical power calculators to figure out how many individuals you’d need to sample to reliably detect a difference of x, depending on how variable the individuals you’re sampling are. (Although these are usually geared toward clinical trials/detecting a change in one variable, so I don’t know whether there’d be differences in power if you were trying to detect changes in multiple plant species.)

  2. You’ll also need control plots or a control area, like schizoform said.

  3. You may also want to read up on plot/quadrat layouts. A lot of people use structured sampling, where there are larger plots that then get sampled in multiple square foot or square meter plots.