I am dealing with a species distribution. From aerial imagery, the species' presence has been assessed at specific locations by identifying the presence/absence of the species of interest under 9500 points placed in a grid over the image. The procedure has been repeated over images captured in 5 different years over the same area, and the points (forming a grid) were always placed in the exact same location (having the points in the same place was necessary for another analysis). I would like to use this dataset to check/quantify changes in the species' coverage between the 5 years, but the data isn't really independent since my points are not placed at random, and they're always at the same place.

I think one option would be to randomly sub-sample a part of my data in each year, and use that for analysis. In that case, I don't know how much of the data should be used (25% of data from each year? or 30%? or 50%?... is there some sort of guideline, or should I just pick a lower proportion of data that would minimize the possibility of randomly selecting the same point each year?). Are there any other ways of analyzing this dataset that I should look into as well?

  • 1
    $\begingroup$ I don't see why having your locations fixed in advance makes the data non-independent. However, repeatedly measuring the same location would mean the same points at different times wouldn't be independent. In addition, nearer points should be more similar that further points--another kind of non-independence. $\endgroup$ – gung - Reinstate Monica Apr 28 '16 at 1:36
  • $\begingroup$ modeling data that varies over both space and time is the purview of spatio-temporal data analysis. amazon.com/Statistics-Spatio-Temporal-Data-Noel-Cressie/dp/… $\endgroup$ – Sycorax Jun 20 '19 at 17:23