Verification of deterministic tropical cyclone intensity forecasts: Moving beyond mean absolute error

Deterministic forecasts of tropical cyclone (TC) intensity are typically verified by calculating a summary accuracy measure, such as mean absolute error or mean squared error. These summary measures quantify, with a scalar value, the quality of the relationship between a set of forecasts and the corresponding set of observations (i.e. "forecast quality"). While a useful simplification, representation of forecast quality with a summary accuracy measure does not respect the complexity of the relationship between forecasts and observations. To obtain a complete representation of forecast quality, the joint probability distribution of forecasts and observations must be estimated. Here, such joint distributions are estimated for operational forecasts of TC intensity (produced by the NHC, decay-SHIPS, SHIFOR and GFDL), using 5 recent years of Atlantic basin forecasts and the corresponding best track observations. Analysis of the joint distributions shows that for all forecast systems, intensity predictions tend to asymptote with lead time toward the central tendency of the climatological intensity distribution (assuming dissipation of the TC is not predicted). It is argued that this behavior is a response to the demand that operational TC intensity forecast systems minimize the mean absolute error of their predictions.