probabilities threshold iri global

[Tinkaa]

Computed global return periods of reaching the 42.5% threshold on IRI forecasts. Return period seems very high (i.e. not common) so would be great if you could have a look if you think this approach makes sense or I am overlooking something.

[Tinkaa]

Actually one thing might be that blobs of raster cellls with less extreme probabilities might generally be larger and are thus dominating this analysis when looking at per raster cell? 🤔

[caldwellst]

Ha, step to use config wasn't that hard!

What a tough problem. Some questions I have, also based on what you've asked:

• Should we restrict to similar longitudes, and maybe even just look at the Sahel for instance?
• To answer the question about hitting the area thresholds, could we look at whenever a cell hits the threshold, percent of other areas that have? Or even expand that to a look at how many cells in a radius of X hit the threshold? Basically testing if it's often that 42.5 is the highest probability we see, and we are only capturing peaks in probability that are not met in neighbouring cells that fall below it, or if the peaks are much higher? Judging by the histogram, I hope it's the latter, but unclear.
• On top of that, if we group them together, we could classify these periods and cells as 1 event, which might make it very clear how low the return period is. It could be that the 5% above the threshold actually are just a large number of cells in a very few number of years/geographic areas.

[caldwellst]

Sorry I don't have time to work more on this! Thanks for your work and hopefully these comments stay relevant for us when we work on this after you leave!