17:15 〜 19:15
[AAS02-P14] 台風カーヌン(2023)の予測可能性
キーワード：台風、カーヌン(2023)、アンサンブル予報、予測可能性
This study investigated the predictability of Typhoon KHANUN, which struck the Okinawa main island twice in the summer of 2023, causing significant damage. KHANUN formed east of the Philippines on 28th July at 0000 UTC, rapidly intensified, and moved northwards. It then experienced a sharp decrease in both intensity and speed before making a sudden northeastward turn off the coast of Taiwan on 3rd August at 1200 UTC. Thereafter, the typhoon turned east before making another sharp turn to the north off the southern coast of Kyushu on 7th August at 0000 UTC, before weakening into a remnant low over the Korean Peninsula on 10th August at 0600 UTC.
This study used medium-range ensemble forecasts from ECMWF, JMA, and NCEP to examine the predictability of KHANUN’s two turns. Ensemble members that successfully predicted the turn were defined as those whose predicted typhoon position errors consistently remained within 200 km from the forecast initial time to 48 hours after the turn time. Those that successfully predicted both the turn and intensity were defined as members whose position errors consistently remained within 200 km and whose central pressure errors remained below 20 hPa for the same period.
Regarding the first turn, ECMWF and NCEP showed gradual improvements in prediction accuracy as the forecast initial time became more recent. At 24 hours before the first turn began, 70% of the members from these two centres successfully predicted the turn. For both the turn and intensity to be predicted by over 70% of the members, only ECMWF's forecasts initialised 12 hours before the first turn achieved this.
For the second turn, the forecast accuracy rapidly increased after 1200 UTC on 3rd August, corresponding to the time of the first turn. ECMWF forecasts initialised 36 hours before the second turn successfully predicted it with over 70% of the members, while NCEP and JMA forecasts initialised 24 and 12 hours prior, respectively, also reached this threshold. The prediction of both the second turn and its intensity, with over 70% success, was also achieved by the respective centres using the same initial times. Moreover, a member from ECMWF initialised at 0000 UTC on 2nd August was the first to accurately predict both the first and second turns, including their intensities.
Comparing low- and high-accuracy forecasts, it was found that the low-accuracy forecasts had lower prediction accuracy for the shape of the high-pressure system around KHANUN, whereas high-accuracy forecasts were better at predicting the high-pressure pattern. These results suggest that the predictability of KHANUN's turns was significantly influenced by the accuracy of the background field predictions.
This study used medium-range ensemble forecasts from ECMWF, JMA, and NCEP to examine the predictability of KHANUN’s two turns. Ensemble members that successfully predicted the turn were defined as those whose predicted typhoon position errors consistently remained within 200 km from the forecast initial time to 48 hours after the turn time. Those that successfully predicted both the turn and intensity were defined as members whose position errors consistently remained within 200 km and whose central pressure errors remained below 20 hPa for the same period.
Regarding the first turn, ECMWF and NCEP showed gradual improvements in prediction accuracy as the forecast initial time became more recent. At 24 hours before the first turn began, 70% of the members from these two centres successfully predicted the turn. For both the turn and intensity to be predicted by over 70% of the members, only ECMWF's forecasts initialised 12 hours before the first turn achieved this.
For the second turn, the forecast accuracy rapidly increased after 1200 UTC on 3rd August, corresponding to the time of the first turn. ECMWF forecasts initialised 36 hours before the second turn successfully predicted it with over 70% of the members, while NCEP and JMA forecasts initialised 24 and 12 hours prior, respectively, also reached this threshold. The prediction of both the second turn and its intensity, with over 70% success, was also achieved by the respective centres using the same initial times. Moreover, a member from ECMWF initialised at 0000 UTC on 2nd August was the first to accurately predict both the first and second turns, including their intensities.
Comparing low- and high-accuracy forecasts, it was found that the low-accuracy forecasts had lower prediction accuracy for the shape of the high-pressure system around KHANUN, whereas high-accuracy forecasts were better at predicting the high-pressure pattern. These results suggest that the predictability of KHANUN's turns was significantly influenced by the accuracy of the background field predictions.