Improve probabilities computation
Hi @vagudets,
I'm opening this issue to write down what we have been discussing in today's SUNSET meeting.
For creating the multi-model ensemble using the "mean" approach, we need to load the probabilities saved in the individual model outputs. However, I found the following problems:
- The probabilities are not computed for the observations, which is needed for the "mean" approach. For this, I suggest to use s2dv::GetProbs functions, as it is able to compute the probabilities for both predictions and observations. Besides, it allows to compute the probabilities using weights, which will be very useful to implement the multi-model "weighted" approach.
- The probabilities are actually saved (sorry that I said in the meeting that they are not saved) in the Skill folder. I suggest to create another module "Probabilities" for their computation and saving (in another folder). The reason is that probabilities and skill computation are two different things (Probabilities creates the probabilistic forecast, while Skill evaluates its quality).
- Skill module should accept probabilities. In this way, the probabilities are computed once (with the Probabilities module) and not for every probabilistic metric. Also, this is required to use the multi-model "mean" approach.
Please let me know if you want to discuss anything.
Carlos