Abstract
Abstract Statistical methods are used to develop the Prediction of Intensity Model Error (PRIME) for both the absolute error and bias of intensity forecasts of Atlantic basin tropical cyclones from 2007 to 2014. These forecasts of forecast error are formulated using a stepwise multiple linear regression framework and are applied to 12–120-h intensity forecasts for the Logistic Growth Equation Model (LGEM), Decay–Statistical Hurricane Intensity Prediction Scheme (DSHP), interpolated Hurricane Weather Research and Forecasting Model (HWFI), and interpolated Geophysical Fluid Dynamics Laboratory (GHMI) hurricane model. The predictors selected for the regression are a combination of storm-specific characteristics, synoptic features, and parameters representing initial condition error and atmospheric flow stability. The performance of PRIME is assessed by comparing the predictions of forecast absolute error and bias to the climatology of these quantities for each of the models. Using paired t tests, the errors in PRIME are found to be significantly smaller than climatology at the 99% level for bias and at the 95% level for absolute error. These percentages vary based on the model and forecast interval, with larger improvement observed for less accurate models and long-range forecasts. A second, more accurate version of PRIME is trained using retrospective forecasts generated by applying the 2014 version of each model to the 2008–13 hurricane seasons. Considering the positive results and use of predictors available prior to the National Hurricane Center official forecast deadline, PRIME forecasts could be valuable to the hurricane forecasting community.
