Determining 70 Percent Probability-Circle Radii of Tropical Cyclone Track Forecasts with Multiple Ensembles

Abstract

The effectiveness of multiple ensembles to determine 70% probability-circle radii of operational tropical cyclone (TC) track forecasts in the Japan Meteorological Agency (JMA) is investigated. The ensembles used in this study are global ensembles from JMA, the European Centre for Medium-Range Weather Forecasts (ECMWF), the National Centers for Environmental Prediction (NCEP) and the Met Office in the United Kingdom (UKMO). The verification for all TCs from 2016 to 2018 reveals that the multiple ensemble-based method has stronger correlation with operational TC track forecast errors and the clearer degree of separation among confidence levels derived from the ensemble spreads than the conventional statistical and single ensemble-based methods. It indicates that the multiple ensemble method provides situation-dependent forecast uncertainty most appropriately. As the effectiveness of the multiple ensemble method has been confirmed, JMA started to operate the multiple ensemble-based 70% probability-circle radii for its operational TC track forecasts for all forecast times up to 120 hours in June 2019. The radii are based solely on confidence levels derived from cumulative ensemble spreads of the multiple ensemble from the 4 centers. This is a good example of successful research to operation transfer of The International Grand Global Ensemble (TIGGE) project.

Probability Ellipse for Tropical Cyclone Track Forecasts with Multiple Ensembles

“…Meanwhile, the mean spread is larger than the mean error in the bin with large spreads, indicating that the ensemble forecasts are overdispersive. The correlation coefficients (r) are positive, ranging from 0.29 to 0.62, and comparable to those for the spread-error relationship (r = 0.42 and 0.32 at FT of 48 and 96 h, respectively) obtained without decomposition into the AT and CT components (Fukuda and Yamaguchi 2019). The correlation coefficients tend to be larger when the relationship in the AT direction is evaluated than when it is evaluated in the CT direction, implying that the ensembles are more skillful in predicting uncertainties in the AT direction.…”

“…cumulative ensemble spreads of the multiple ensemble (Fukuda and Yamaguchi 2019). The probability circle implicitly assumes that position errors are distributed isotropically, but the introduction of the probability ellipse makes it possible to provide appropriate information about uncertainty in both the direction and speed of TC movement.…”

“…Therefore, in 2019, JMA implemented probability circles with a 70 % probability radius obtained by the multiple-ensemble technique for operational TC track forecasts. Fukuda and Yamaguchi (2019) also found, however, that the mean probability circle radius and, therefore, the mean area differed little, whether determined by the conventional statistical, single-ensemble, or multiple-ensemble method, indicating that the multiple-ensemble method does not produce a significant reduction in probability circle area compared with the statistical and single-ensemble methods. The use of a probability circle involves the implicit assumption of an isotropic error distribution, whereas the introduction of the probability ellipse would make it possible to provide information as to which is more uncertain; the direction or speed of TC movement.…”

“…Figure 6 shows the relationships between the ensemble spreads in the AT (CT) direction and the position errors of the ensemble mean predictions in the AT (CT) direction at FT of 48, 72, and 96 h. Errors are binned according to the spreads. The number of the bins is set to three in line with the number of categories of TC track forecast uncertainties used in RSMC Tokyo (Fukuda and Yamaguchi 2019). In a well-calibrated ensemble, the resulting line is close to the diagonal (Haiden et al 2019).…”

“…However, the differing methods used to determine probability circle radii between the 3-to 72-h and 96-to 120-h forecasts resulted in discontinuities in the rate of increase of the radius. To address this issue, JMA has explored a new method of determining radii uniformly for forecast lead times of up to 5 days that uses multiple ensembles, from JMA, the European Centre for Medium-Range Weather Forecasts (ECMWF), the U.S. National Centers for Environmental Prediction (NCEP), and the Met Office in the United Kingdom (UKMO) (Fukuda and Yamaguchi 2019). Fukuda and Yamaguchi (2019) found a stronger relationship between TC position errors and ensemble spreads for multiple ensembles than for a single ensemble and showed that the multiple-ensemble method provided the most appropriate probability circles with the highest correlation between radii of the circle and TC track forecast errors among the statistical, single, and multiple-ensemble methods.…”