I am performing some species distribution model comparisons under different sampling strategies. I have two sampling constraints with five levels (e.g. presence sample sizes), for a total of 25 sampling strategies. Each strategy has a total of 100 replicates, with a model fit to each replicate.
Each of the models regress a binary response variable (0,1) and has two predicted species distribution maps. One is continuous between zero and one and the other is a binary map.
I want to look at the variability of map outputs within each sampling strategy. I am looking at Fleiss Kappa to quantify agreement between the binary maps, but I also want to estimate the similarity between the regression maps. Initially I was considering calculating a pairwise Pearson's correlation coefficient between each output and taking an average. This could become computationally taxing as there would be 4950 pair combinations for each of the 25 strategies for 4 modelling approaches and 2 species (990,000). I am aware of the R package spatialEco, which samples each raster to reduce computation time, which could mitigate this issue.
However, is there another more appropriate method that can be used to assess the variability of within the replicate predictions?