I am simulating virtual species on a 100x100 grid (the size for now). Each grid layer represents one environmental variable. The "suitability function" defines the probability of a presence at a given point based on the value of each variable (grid layer). Actual probabilities are used (using Bernoulli trials), rather than the threshold approach (where all probabilities above a given level are set to true). Stratified random sampling is used. The sampling and determination of whether a point is a presence or absence is done in the same step. Random forests are used to create a species distribution model (SDM).
Is there some way to do a power analysis which tells you what sample size you need to achieve some measurable metric, such as AUC-ROC or Brier score (or other type of metric)?