2.5 On the Prospects for Improved Tropical Cyclone Forecasts

Monday, 13 January 2020: 11:30 AM
104C (Boston Convention and Exhibition Center)
Feifan Zhou, IAP, Beijing, China; and Z. Toth

It is well established that due to the chaotic nature of the atmosphere, even with a perfect model, the lead time range of skillful weather forecasts (i.e., predictability) is limited and controlled by the level of initial (analysis) error. A recent BAMS essay (Landsea and Cangialosi 2018) noted a diminishing trend in the reduction of perceived positional errors (PPE) in National Hurricane Center Tropical Cyclone (TC) forecasts (i.e., difference between official forecast and observed positions). Hence the authors contemplate whether “the approaching limit of predictability for tropical cyclone track prediction is near or has already been reached”.

Here we consider an alternative interpretation of the PPE data. First we note that PPE is different from true positional error (TPE, difference between forecast and true position) as it is influenced by the error in the observed (i.e., “best track”) position of TCs (i.e., best track error - BTE), which imparts a bias into PPE. PPE is still customarily used as a substitute for TPE since the latter is not directly measurable. As an alternative, TPE is estimated here with an inverse method, using 36-120 hr PPE measurements over the 2001-2017 seasons, and a theoretically based assumption that TPE grows exponentially with lead time.

89% variance in the behavior of seasonally averaged PPE measurements are found to be explained with an error model using just four parameters: Initial error in 2001 (53 nm); the exponential error growth rate (1.9 days error doubling time); the ratio of BTE to initial TPE (0.3); and the rate of analysis error reduction over the years, which is also found to behave exponentially with a 4.4% reduction from one year to the next. The exponential growth of TPE implies a continual tapering off of PPE error reduction over the years, which, along with sampling fluctuations, may explain the flattening of PPE and skill curves observed in recent years. Such tapering of PPE reduction, however, is not necessarily indicative of nearing impending limit in predictability. Assuming that the level of investments in, and the pace of improvements to the observing, modeling, and data assimilation systems continue unabated, the 4-parameter error model indicates that the limit of predictability at the 167 nm error level, reached at day 5 in 2017, will be extended beyond day 6 / 8 in 10 / 30 years time.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner