3. Impact of spectral nudging on real-time TC forecast
3.3. Case Study: typhoon Neoguri and Vongfong
26
Table 3.2 Forecast Details and Initial Times for the 18 Typhoons which Occurred during 2013 and 2014. The Interval of Forecast Initial Time is 24 h.
TC number TC name Forecast initial time (interval of 24 h) Number of cases
1307 SOULIK 2013/07/08 00 UTC 1
1311 UTOR 2013/08/10 00 UTC 1
1319 USAGI 2013/09/17 00 UTC ~ 2013/09/18 00 UTC 2
1323 FITOW 2013/10/01 00 UTC ~ 2013/10/02 00 UTC 2
1325 NARI 2013/10/10 00 UTC 1
1326 WIPHA 2013/10/11 00 UTC ~ 2013/10/13 00 UTC 3
1327 FRANCISCO 2013/10/17 00 UTC ~ 2013/10/21 00 UTC 5
1329 KROSA 2013/10/30 00 UTC 1
1330 HAIYAN 2013/11/04 00 UTC ~ 2013/11/06 00 UTC 3
1408 NEOGURI 2014/07/04 00 UTC ~ 2014/07/06 00 UTC 3
1409 RAMMASUN 2014/07/13 00 UTC ~ 2014/07/15 00 UTC 3
1410 MATMO 2014/07/18 00 UTC ~ 2014/07/20 00 UTC 3
1411 HALONG 2014/07/30 00 UTC ~ 2014/08/06 00 UTC 8
1416 FUNG-WONG 2014/09/18 00 UTC ~ 2014/09/19 00 UTC 2
1417 KAMMURI 2014/09/25 00 UTC 1
1418 PHANFONE 2014/09/30 00 UTC ~ 2014/10/01 00 UTC 2
1419 VONGFONG 2014/10/04 00 UTC ~ 2014/10/08 00 UTC 5
1422 HAGUPIT 2014/12/02 00 UTC ~ 2014/12/06 00 UTC 5
total 51
27
compared to GFS; however, the difference is not large. The NOSN track forecast is better than that of GFS for forecast lead times shorter than 78 h; however, the error in NOSN increases rapidly for longer forecasts and has the largest TPE (approximately 500 km) at 120 h. On the other hand, SN maintains an error similar to GFS after 78 h, and has a 120 h TPE of approximately 350 km. Therefore, due to spectral nudging, the mean track error at 120 h is reduced by about 30%. Conversely, it is apparent that error in the SN run is reduced by nudging from the GFS forecast data. This result indicates that spectral nudging is effective for TC forecasts over 3 days because its impact increases over time.
Figure 3.1 Mean track position error of the NOSN, SN, and GFS runs for 16 forecasts of Typhoon Neoguri and Vongfong (6 and 10 forecasts, respectively). The sample sizes are the same at each forecast hour.
Despite the improved track forecast, spectral nudging also has an adverse effect on the forecast intensity, as indicated by previous studies. Figure 3.2 shows the intensity error of SN, NOSN and GFS forecast runs as compared to JTWC best-track data. Each experiment has a large initial intensity error because initialization methods for TC forecasting are not applied. The surface maximum wind speed errors for NOSN are less than 5 m s-1 for the entire forecast duration (except for 30 h), and the 5-day mean error is approximately 3 m s-1. This indicates that the TC intensity is realistically predicted by the mesoscale WRF model. Conversely, the
28
intensity error at 30 h is about 20 m s-1 for GFS and the 5-day mean intensity error is approximately 15 m s-1, which is five times larger than that of NOSN. Due to the coarse model resolution, the intensification processes of the strong typhoon are not resolved by GFS. For SN, the intensity error notably increases relative to NOSN because spectral nudging is configured too strongly. The 5-day mean intensity error of SN is 6 m s-1, which is almost double that of NOSN. The intensity error of SN increases markedly between 30 and 96 h, when most intensification processes occur. In particular, the intensity error of SN at 66 h is five times that of NOSN. As found by Cha et al. (2011), the intensity of TCs can be underestimated due to spectral nudging because the mesoscale features generated by the high-resolution model are suppressed by spectral nudging from low- resolution global model results. However, previous research (e.g., Feser & Barcikowska, 2012) showed that spectral nudging did not affect TC intensity forecast. The different results between Cha et al. (2011) and Feser and Barcikowska (2012) indicate that the effect of spectral nudging depends on model configurations such as model resolution, forcing data, and nudging options. Therefore, the configuration of spectral nudging should be optimized to improve TC forecast.
Figure 3.2 Same as Figure 3.1 but for mean absolute maximum surface wind speed error.
To verify the effect of spectral nudging on each case, we analyze the characteristics of the track and intensity forecasts initialized at 0000 UTC 7 Oct 2014 in Figure 3.3. Typhoon Vongfong moves approximately westward
29
after initiation and shifts to the north after 0000 UTC 9 Oct 2014 when it reaches 130ºE (Figure 3.3a). In NOSN, the simulated track deviates slightly to the south of the JTWC best-track data, changes direction before reaching 130ºE, and shifts to the northeast thereafter. Thus, NOSN has a relatively large track error after the shift. In SN, the simulated track also deviates southward prior to turning. Similar to NOSN, SN shifts prior to reaching 130ºE; however, after the shift the northwestward track is more accurately represented. Thus, SN considerably reduces the track error after 48 h as compared to NOSN, and the TPE at 120 h in SN is approximately 54 % smaller than that of NOSN. It is notable that the SN track closely corresponds to GFS, because SN is profoundly influenced by the GFS forecast data through spectral nudging. Similarly, tracks of all other SN runs follow the GFS run (not shown). In the case of typhoon Vongfong, GFS simulates a more accurate track as compared to the WRF model without spectral nudging. Therefore, the track forecast in SN is prominently improved by spectral nudging from GFS forecast data. This indicates that spectral nudging can play a role in improving track forecasts when the GFS global forecast simulates TCs accurately.
Figure 3.3 (a) Simulated track and (b) surface maximum wind speed of NOSN, SN, and GFS runs for Typhoon Vongfong initialized at 0000 UTC 7 Oct 2014. The JTWC best-track data is given as a reference.
Conversely, spectral nudging tends to decrease TC intensity (Figure 3.3b). From the best-track data, the maximum intensity of Typhoon Vongfong is 80 m s-1 at 0000 UTC 8 Oct 2014. In NOSN, the initial intensity error is large due to the absence of initialization, and the simulated intensity then gradually improves and becomes realistic after 66 h. The TC intensity simulated by GFS is approximately 40 m s-1 for most of the forecast period, which is markedly weaker than observations. This can be attributed to GFS being unable to capture the intensification processes of typhoon Vongfong. The simulated TC intensity of SN tends to be
(a) (b)
30
weaker than that of NOSN, although it is slightly improved relative to GFS; SN maximum surface wind speed is approximately 50 m s-1, and in NOSN it is 60 m s-1. This is because spectral nudging suppresses the development of mesoscale features, such as TC intensification processes, when the GFS forecast data including a weak TC vortex is nudged into the WRF model.
To examine the reasons for the improvement in SN track forecast, we analyzed the synoptic fields for the three experiments. We focus on 72 h when spectral nudging has effectively reduced the track error. In comparison, 500 hPa winds that are similar to steering wind and a geopotential height of 500 hPa are used (Xie et al. 2010). Figure 3.4 shows the 500 hPa winds and geopotential height for the GFS analysis and the three simulations. The GFS analysis data shows that the western North Pacific subtropical high (WNPSH) is developed at the southeast of the model domain, and southwesterly wind induced by anticyclonic circulation blows along the 5880 gpm line. In NOSN, the WNPSH is overestimated and the 5880 gpm line is closer to the center of the TC. The distance between the TC center and subtropical high center of NOSN is 2097 km while that of GFS analysis is 2531 km. Thus, strong southwesterly wind blows to the east of TC center, which results in northeastward TC movement. In GFS, southwesterly wind near the 5880 gpm line is realistically simulated, because the WNPSH is not expanded to the west. Therefore, GFS properly simulates the track to the northwest due to the weak impact of anticyclonic flows related to the WNPSH. In SN, the distance between the TC center and subtropical high center also increases to 2220 km by spectral nudging from the GFS forecast, which presents a distance of 2234 km. Since the WNPSH associated circulations are weakened, the southwesterly winds causing northeastward movement of the TC are weakened. Therefore, the steering wind is improved by spectral nudging, leading to reduced track error in SN.
31
Figure 3.4 A geopotential height (contour) and wind (vector) of 500 hPa in the NOSN, SN, and GFS run, and the GFS global analysis 72 h after model integration. The forecast initial time is 0000 UTC 6 Oct 2014. The contour ranges from 5870 gpm to 5910 gpm by 10 gpm.