Very good question. It is not based just on the IRI/CPC forecast plume. There are 11 human forecasters involved in the development of the probabilities. These forecasters inspect the plume, but also look at a large set of other inputs, some of which are current observations and their recent trends, and some of which are details of the forecasts of some of the models on the plume. These details include, for example, the spread of the model's ensemble members, which is related to the predictive uncertainty. The plume shows the mean forecasts of each model, but does not show the spread of the individual dynamical models' forecasts. Human judgement and experience also comes into play in each forecaster's vote on the probabilities. It is an interesting issue whether this human element improves the value of the forecasts, or if the set of the models' objective forecasts would be just as good, or even better. Some of the 11 forecasters have been around for decades and have a ton of experience with ENSO, while others are earlier in their career. Each person undoubtedly weights things a bit differently, and it is hoped that 11 forecasters is enough for some of these biases to offset one another. The probabilities based on the plume alone can be found in one of the graphs on an IRI web site: http://iri.columbia.edu/our-expertise/climate/forecasts/enso/current/  The plume-based probabilities are formed not just by counting the proportion of models forecasting each category (La Nina, neutral, El Nino), but are based on the average numerical forecast among all models, combined with an uncertainty distribution (Gaussian) derived from an estimate of the historical skill of the set of models over a a 30 or greater year period.