Skip to main content

CPC/IRI started jointly putting together probabilistic ENSO forecasts in early 2012, so unfortunately the record is too short to quantify our reliability scores, but we plan to verify in the future. It is easier to develop robust statistics for forecasts made more often (i.e. every day), but in our outlooks, we only obtain data points once monthly, which requires time to verify.

However, our probabilities do not stray too widely from more objective probabilistic tools, which the forecasters view and take into account.  For example, there are results using 30 years of CFSv2 reforecasts (or hindcasts) that show that model tends to be a touch overconfident at longer forecast leads (i.e. the model tends to forecast slightly higher probabilities than what were actually observed).

You can also view archives of IRI’s probabilistic forecasts based on the ENSO model plume, which are available here:

http://iri.columbia.edu/our-expertise/climate/forecasts/enso/ (click on the month/year on the top right and click "plume based probabilistic ENSO forecast') 

And there is an archive of CPC ENSO diagnostics discussions going back to 2001, if you want a more qualitative estimate of how good the forecasts are: 

http://www.cpc.ncep.noaa.gov/products/expert_assessment/ENSO_DD_archive…