Diagnosing Conditions Associated with Large Intensity Forecast Errors in the Hurricane Weather Research and Forecasting (HWRF) Model

Understanding and forecasting tropical cyclone (TC) intensity change continues to be a paramount challenge for the research and operational communities, partly because of inherent systematic biases contained in model guidance, which can be difficult to diagnose. The purpose of this paper is to present a method to identify such systematic biases by comparing forecasts characterized by large intensity errors with analog forecasts that exhibit small intensity errors. The methodology is applied to the 2015 version of the Hurricane Weather Research and Forecasting (HWRF) Model retrospective forecasts in the North Atlantic (NATL) and eastern North Pacific (EPAC) basins during 2011–14. Forecasts with large 24-h intensity errors are defined to be in the top 15% of all cases in the distribution that underforecast intensity. These forecasts are compared to analog forecasts taken from the bottom 50% of the error distribution. Analog forecasts are identified by finding the case that has 0–24-h intensity and wind shear magnitude time series that are similar to the large intensity error forecasts. Composite differences of the large and small intensity error forecasts reveal that the EPAC large error forecasts have weaker reflectivity and vertical motion near the TC inner core from 3 h onward. Results over the NATL are less clear, with the significant differences between the large and small error forecasts occurring radially outward from the TC core. Though applied to TCs, this analog methodology could be useful for diagnosing systematic model biases in other applications