Thresholding Strategies in Ensemble Predictions: Pre-Processing vs. Post-Processing Mean and Variance Calculations

[mrcreamio]

I noticed that in the code, the program is thresholding the ensemble predictions before calculating the mean and variance:

I'm curious about the rationale behind applying the thresholding before the mean and variance calculations. Could anyone elaborate on the benefits and potential drawbacks of this approach?

For instance, how would the results and their interpretation change if we were to calculate the mean and variance directly on the raw predictions and then apply the thresholding afterward? Specifically, what impact would this have on the variance map and the final mean prediction?

Moreover, from your experience or in the literature, is there a preferred or best practice for such thresholding in ensemble prediction scenarios? Any insights on scenarios where one method might be more advantageous than the other would be greatly appreciated.

Looking forward to your thoughts and a fruitful discussion on this topic.

@JuliaWolleb @Warvito @vfdev-5 @drbeh