Computing and Understanding Forecast Errors in Tropical Cyclone Motion

Previous work has shown that the environment flow is the leading factor that influences TC motion. What exactly is meant by “environment flow” varies greatly among the refereed papers on the subject, but in general it has been suggested that the motion of well-developed TCs can be best approximated by the deep-layer mean (DLM) wind. How the environment flow is defined may vary greatly from case to case. Typically, the computation of environment flow involves removing the wind field associated with the TC vortex. Since TCs vary in horizontal scale, the computation of the environment wind is sensitive to the radius of TC removal in addition to the vertical depth. How well might the environment flow agree with the actual TC motion when computing the environment flow over a fixed depth and TC removal radius versus a varying depth and radius? Are forecast errors in TC motion primarily due to error in environment wind, or other factors such as errors in vertical depth and/or horizontal structure of the TC? These questions will be addressed in this presentation.

The aim of this presentation is three-fold. First, we will discuss the overall characteristics of TC motion vector errors in the 2011 version of the Advanced Hurricane Weather Research and Forecasting (AHW) model run in real-time, and retrospectively for 2008–2010, at NCAR. Second, we will examine the computation of steering flow and how it is sensitive to the vertical depth and the radius of TC removal. Third, we will examine the TC track forecast busts in AHW for the 2008–2010 retrospective period. Specifically, we will diagnose the sources of error quantitatively in the 24-h AHW forecasts that contributed to large errors in TC motion. We will examine the motion vector errors at 24-h in the AHW forecast because it is (i) relatively certain that the actual position error is small enough that the diagnosis makes sense, and (ii) expected that large errors in TC motion early in the forecast will lead to huge position errors at longer lead times.

Emerging results indicate that the standard DLM flow is not always the best discriminator for TC motion, in agreement with previous work on the subject. New findings that build on previous work include i) the DLM flow provides a good match with the actual storm motion only 48% of the time, and is not always the best match for TC motion even for strong well-developed TCs, and ii) there is a better match between steering flow and TC motion when the vertical depth and TC removal radius are allowed to vary. Specifically for AHW, westward and northwestward-moving TCs most commonly have opposite- (systems move too slow) and right-of-motion errors. These errors are primarily due to errors in synoptic-scale features that contribute to large environment wind biases, most notably the westerly wind bias in the main development region that is related to persistent errors in the western flank of the subtropical ridge over the central and eastern North Atlantic.