The algorithm for probability maps begins by pooling forecast data into 2° x 2° bins, centered at each grid point, over a 7-day time window associated with the second week of the forecast. This process is repeated for the entire ensemble member set, which is then combined and reshaped into a single large array, forming an empirical cumulative distribution function. A predefined percentile, tuned to 87, is used to subsample the extreme tail of the distribution, from which probabilities associated with given thresholds are computed. Following group meetings with forecasters from OPC/NWS and considering the Beaufort scale and Saffir-Simpson Hurricane scale, the following levels have been defined: wind speeds of 28, 34, 41, 48, 56, and 64 knots; significant wave heights of 4, 6, 9, and 14 meters; and peak periods of 17, 20, and 22 seconds.
The probability maps have been generated on a daily basis since June, 2023, and the program has been run for historical conditions related to significant case studies to validate the method across different meteorological events. Moreover, the GEFS forecasts have been evaluated against buoy and altimeter data using ensemble-specific metrics such as Continuous Ranked Probability Score (CRPS) and rank histograms. Fuzzy verification has been employed to validate long-term forecasts, and the resulting probabilities have been analyzed using Reliability Diagrams and the Brier score. We present and discuss both successful and unsuccessful cases, as well as the conditions under which the probabilistic forecast proves to be most beneficial.

