12 Figure 2.4 Average track position error for all cases (black line), top 25% of error cases (red line) and bottom 25% of error cases (blue line) with distribution of top and bottom error cases (red and blue shading, respectively) . 14 Figure 2.6 Time series of the difference between the optimal model control vector and the analysis data for TC group 1 (red) and TC group 2 (blue). 19 Figure 2.10 Time series of the difference between the optimal control vector and the actual TC motion (a) observation and (b) model for northward TCs (red) and westward TCs (blue).
Introduction
Research background
To derive the general result for improved prognosis, statistical approach with a large number of cases is required. To obtain the general result with a large number of cases, the study of TC forecast data of several years is necessary. For the study with a large number of cases with consistency, the Weather Research and Forecast (WRF) model is used in this study in terms of computational efficiency and availability.
Research objectives
However, the results of the case study had the limitation that the improvement cannot be applied to other TC cases. However, the performance models were continuously upgraded by changing the dynamical kernel, grid resolution, physical parameterization, and data assimilation method. A total of 125 forecasts of 37 TCs are considered in this study to verify and improve TC forecasts by the WRF model.
Verification of WRF forecast skill for the western North Pacific TCs
- background
- Model configuration and method
- Cluster analysis
- Optimal steering vector analysis
Therefore, the optimal steering vector of the model and analysis data can identify the environmental winds that best match the actual TC motion (Galarneau Jr and Davis 2013). The difference in the steering vector between model and analysis data is related to the error in the environmental winds in the model simulation. To simulate Phanfone, the simulated WNPSH is initially extended north of the TC.
The transverse (longitudinal) component is defined as the component of the residual vector that is perpendicular (parallel) to the TC passage direction. A negative value of the transverse component means that the actual motion is shifted to the left of the control vector.
Impact of spectral nudging on real-time TC forecast
- background
- Model configuration and experiments
- Case Study: typhoon Neoguri and Vongfong
- Optimization of Spectral nudging on TC forecast
- General effect of spectral nudging on TC forecast
The first was a case study to examine the effects of spectral drift for typhoons Neoguri and Vongfong. Therefore, the threshold wavelength of spectral stimulation should be optimized to maximize the positive effects of spectral stimulation. In the spectral shift method, the difference between the large-scale fields of the global model and the WRF model is shifted to the model field by multiplying by the nudging coefficient.
In the case of Typhoon Vongfong, the GFS simulates a more accurate track compared to the WRF model without spectral forcing. Therefore, the track forecast in the SN is significantly improved by the spectral forcing from the GFS forecast data. The influence of spectral thrust can be controlled by three thrust options (i.e., cutoff wavelength, thrust coefficient, and thrust interval).
It should be noted that the weaker impact of spectral nudging leads to stronger, more accurate TC intensity (Figure 3.5c and d). Therefore, the improved prediction of synoptic fields by spectral nudging causes a positive effect on the trajectory prediction. Although the effect of spectral nudging appears not to be significant up to 96 hours (Figure 3.7), the effect can vary significantly between cases.
Therefore, spectral drift tends to improve track prediction when the forecast time is longer than 3 days. The spectral shift has a positive impact on the forecast for these systems due to improved synoptic fields. Therefore, the positive effects of spectral pushing can be enhanced by applying it only to TCs with curved tracks.
Increasing model resolution with moving nesting domain
- background
- Model configuration and experiments
- Impact on TC intensity forecast
- Impact on TC track forecast
The track error was decomposed into along-track and cross-track deviations with respect to the actual motion of the TC. The cross (longitudinal) component was defined as the component of the residual vector perpendicular (parallel) to the transition direction of the TC. The 1DM run tends to underestimate the MWS, while the MWS bias of the 2DM run is significantly reduced.
In particular, the 2DM bias is minimized in the middle of the forecast time, when TCs have intensified for many cases. It appears that the decay process is not properly reproduced in the 2DM run because the cooling effect from the TC–ocean interaction is not taken into account due to the lack of an ocean model. There is a sudden increase in wind speed at the initial time of the 2DM run.
For MWS as well as MSLP, the root mean square errors (RMSE) of the 2DM run are generally smaller than those of the 1DM run. However, for weak TCs with LMI less than 60 m s-1, higher model resolution may increase the intensity error due to wind speed overestimation. Since the two experiments differ only in the resolution of the TC core region, there is no significant difference in the environmental field.
As a result, the decrease in track error of the 2DM run is primarily associated with the reduced cross-track bias.
Impact of TC-ocean interaction on TC intensity forecast
Model configuration and experiments
The tropical cyclone heat potential (TCHP) is calculated from the following equation, and figure 5.7 shows the TC translation speed and TCHP of the underlying ocean. The GFS forecast and 1DM experiments generally underestimate the maximum wind speed throughout the forecast time. The intensity of 2DM experiments is relatively increased and the bias is significantly reduced with 2DM experiments. Since the western North Pacific has a natural climate system (e.g., the subtropical high or midlatitude of the western North Pacific), different features for TC prediction by location were used to create an integrated model of TC forecast.
Doyle, 2014: An evaluation of the impact of horizontal resolution on tropical cyclone predictions using COAMPS-TC. A., 1996: Specific tropical cyclone track types and unusual tropical cyclone motions associated with a reverse oriented monsoon trough in the western North Pacific. Xie, 2012: A scale-selective data assimilation approach for improving tropical cyclone track and intensity forecasts in a limited-area model: A case study of Hurricane Felix (2007).
Pelissier, 1981: Models for predicting tropical cyclone motion over the North Atlantic: An operational evaluation. Sun, Y., and Coauthors, 2017: Impact of ocean warming on the tropical cyclone track over the western North Pacific: A numerical investigation based on two case studies. Leslie, 1991: A basic relationship between tropical cyclone intensity and environmental convection layer depth in the Australian region.
Satoh, 2016: The role of the vertical structure of a simulated tropical cyclone in its motion: A case study of Typhoon Fengshen (2008).
TC-ocean interaction
Verification of integrated TC forecast model
Integrated TC forecast model for real-time forecast
The integrated TC forecast model is developed from previous results, and a real-time TC forecast system is constructed. The ocean coupling model is not included in the integrated TC forecast model since the data are not available in real time. After the GFS 120-hour forecast is completed, the system downloads the GFS data for use as initial and boundary data for the integrated TC forecast model.
The simulated track with the GFS forecast is obtained by tracking the GFS forecast data. After calculating the distance between the GFS forecast track and the center point of each cluster as listed in Table 6.1, the cluster whose distance is the smallest is determined as the target TC cluster. If a C1, C2, or C5 cluster is specified, spectral shift is applied during model integration.
The five-day integrated TC forecast is performed with the moving grid domain, and then the simulated track and intensity data are produced by the post-processing system.
Data and experiments
Verification of track and intensity error of integrated TC forecast model
The correlation coefficients of the experiments are similar between all experiments, but the standard deviations increase in the 2DM experiments. However, the standard deviation is still lower with the moving nest domain compared to HWRF. This result indicates that the intensification and decay process needs to be improved for realistic intensity prediction.
The SOSN_2DM developed through this study are verified with NOSN_1DM, GFS prediction and HWRF. The track errors and maximum wind speed RMSEs for three-day, five-day and five-day averages are compared with each model error (Table 4.4 and Table 4.5). The NOSN_1DM is the first version of this study and SOSN_2DM shows the significant improvement in terms of track and intensity prediction.
However, the five-day maximum RMSE wind speed is slightly increased due to the overestimated intensity in the late forecast time. This shows the advantage of the higher resolution domain of the regional model in intensity prediction. SOSN_2DM's track forecast is more accurate than GFS and its intensity forecast is more accurate than HWRF.
With spectral shift, track error is reduced by 8%, with higher resolution, track error is reduced by 1%, and intensity error is reduced by 16.
Discussion and conclusion
Discussion
Conclusion
Emerton, 2015: ECMWF 앙상블 및 결정론적 예측 시스템을 통해 북반구 열대 저기압 예측으로 수명 주기가 연장되었습니다. Mohanty, 2011: WRF 모델을 사용하여 벵골만의 열대 저기압 Nargis 예측에 대한 물리적 매개변수화의 민감도. 이 글을 통해 제가 학업을 마치는 동안 많은 도움을 주신 분들께 감사의 마음을 전하고 싶습니다.
먼저, 학부 2학년부터 박사학위까지 지도해주신 차동현 교수님께 감사의 말씀을 전하고 싶습니다. 여러분의 아낌없는 조언 덕분에 제가 박사 학위를 마칠 수 있었고, 차근차근 연구를 계속할 수 있었습니다. 연구뿐만 아니라 생활에서도 많은 것을 배웠습니다.
선생님께서 가르쳐주신 교훈을 결코 잊지 않고 앞으로도 자랑스러운 학생이 되도록 노력하겠습니다. 또한, 저와 수많은 시간을 함께 해주시고 항상 연구실 생활을 즐겁게 해주신 연구실 구성원들에게도 감사의 말씀을 전하고 싶습니다. 그리고 연구실 선배가 없을 때 저를 위해 연구실 선배 역할을 맡아주신 은교님과 대현님에게도 감사드립니다.
앞으로는 더 자랑스러운 아들이 될게요.