Predicting the Performance of Tropical Cyclone Intensity Forecasts Using Environmental Parameters

Although operational tropical cyclone track forecasts have improved dramatically in the past 20 years, intensity forecasts have shown little improvement or even regressed. It could be helpful to both forecasters and end users to know whether certain synoptic environments are inherently more difficult for forecasting hurricane intensity. At the time of a forecast, certain initial conditions (“predictors”) can provide useful estimates of the expected error in different models' intensity forecasts. These skill predictions can guide scientists on how to improve model performance and inform forecasters about which model's forecasts will achieve higher accuracy on any given storm.

Four hurricane intensity models that were operational for the duration of the five hurricane seasons between 2006 and 2010, as well as the National Hurricane Center official (OFCL) forecast, were evaluated for Atlantic Ocean basin cases. The four models include the Logistic Growth Equation Model (LGEM), inland decay version of the Statistical Hurricane Intensity Prediction Scheme (DSHP), SHIFOR5 model, and Geophysical Fluid Dynamics Laboratory (GFDL) hurricane model. Each model's performance was assessed by computing the mean absolute error, bias, and skill relative to SHIFOR5 from 2006 to 2010 for multiple forecast times. The intensity forecasts and 0 hour verifications for GFDL, OFCL, and SHIFOR5 were obtained from NOAA's ATCF database. The predictor values as well as the DSHP and LGEM forecasts and verification data were obtained from the SHIPS database maintained by NOAA. Several predictors were tested individually and in combinations to demonstrate different regimes that were conducive to higher or lower skill forecasts. Examples of these predictors include initial shear, initial intensity, storm speed, mid-level atmospheric relative humidity, and maximum potential intensity. The results address conventional wisdom about which environmental conditions lead to better forecasts of hurricane intensity and highlight the different strengths of each model.