Experiments with a Simple Tropical Cyclone Intensity Consensus

Abstract

Abstract Consensus forecasts (forecasts created by combining output from individual forecasts) have become an integral part of operational tropical cyclone track forecasting. Consensus aids, which generally have lower average errors than individual models, benefit from the skill and independence of the consensus members, both of which are present in track forecasting, but are limited in intensity forecasting. This study conducts experiments with intensity forecast aids on 4 yr of data (2003–06). First, the skill of the models is assessed; then simple consensus computations are constructed for the Atlantic, eastern North Pacific, and western North Pacific basins. A simple (i.e., equally weighted) consensus of three top-performing intensity forecast models is found to generally outperform the individual members in both the Atlantic and eastern North Pacific, and a simple consensus of two top-performing intensity forecast models is found to generally outperform the individual members in the western North Pacific. An experiment using an ensemble of dynamical model track forecasts and a selection of model fields as input in a statistical–dynamical intensity forecast model to produce intensity consensus members is conducted for the western North Pacific only. Consensus member skill at 72 h is low (−0.4% to 14.2%), and there is little independence among the members. This experiment demonstrates that a consensus of these highly dependent members yields an aid that performs as well as the most skillful member. Finally, adding a less skillful, but more independent, dynamical model-based forecast aid to the consensus yields an 11-member consensus with mixed yet promising performance compared with the 10-model consensus. Based on these findings, the simple three-member consensus model could be used as a standard of comparison for other deterministic ensemble methods for the Atlantic and eastern North Pacific. Both the two- and three-member consensus forecasts may also provide useful guidance for operational forecasters. Likewise, in the western North Pacific, the 10- and 11-member consensus could be used as operational forecast aids and standards of comparison for other ensemble intensity forecast methods.