Statistical post-processing of ECMWF tropical cyclone track forecasts

Dynamical tropical cyclone (TC) track predictions have become the primary guidance at a number of forecast centers. The global model analyses and forecasts have become more useful as the horizontal resolution has increased and synthetic TC observations have been introduced. However, one of the difficulties with global models is the early portions of the tracks may be inconsistent with the recent motion of the TC, including beginning at an incorrect location. A second situation that leads to an inconsistent initial model position is a major relocation of the TC, which usually occurs during the early stages when the center is not well-defined. Another important timing aspect of the 00 UTC and 12 UTC model track predictions is that many global models, such as Navy Operational Global Atmospheric Prediction System (NOGAPS), are not received at forecast centers until 4-5 h after the synoptic time, and thus are not actually used until the 06 UTC or 18 UTC warning.

The objective of this post-processing technique for European Centre for Medium-Range Weather Forecasts (ECMWF) is to make use of the new knowledge about the TC position at the 00 UTC and 12 UTC synoptic times to improve the accuracy of the ECMWF track predictions at those times. Because of ECMWF’s later data cutoff time, forecast tracks are not generated until at least 16 h after the model’s initial time. These forecast tracks could not be used until the warning 18 h after the model’s initial time. A similar study on the statistical post-processing of NOGAPS tracks revealed the potential for this technique. Stepwise regression utilizing the 0-h through 36-h forecast displacements, initial position offset, and the offset of the backward-extrapolated 0-h through 36-h forecast from the known best track was applied to a database of five seasons of TCs in the western North Pacific to provide the coefficients necessary to make an adjusted NOGAPS forecast. This technique was tested on an independent data set of one western North Pacific TC season. The adjusted NOGAPS improved upon the original NOGAPS forecast by 5% at 72h, and that improvement increased to a remarkable 53% at 12 h.

The high resolution and late cutoff time of ECMWF make it an accurate model to begin with. An adjusted ECMWF TC track forecast is expected to have the advantage of the information from the TC best track for the last 18 h to make an improvement upon the model’s track forecast, will most likely produce an improvement even greater than made by the adjusted NOGAPS, which only had six additional hours. The improvement of the adjusted ECMWF forecast over the original ECMWF forecast will be presented in the preprint and presentation.