GFDL-Type Typhoon Initialization in MM5

Transcription

1 2966 MONTHLY WEATHER REVIEW GFDL-Type Typhoon Initialization in MM5 H. JOE KWON AND SEONG-HEE WON Department of Atmospheric Science, Kongju National University, Kongju, Chungnam, South Korea MYUNG-HWAN AHN, AE-SOOK SUH, AND HYO-SANG CHUNG Meteorological Research Institute, Korea Meteorological Administration, Seoul, South Korea (Manuscript received 9 April 2001, in final form 6 May 2002) ABSTRACT The Geophysical Fluid Dynamics Laboratory (GFDL) hurricane initialization algorithm is implemented in the community fifth-generation Pennsylvania State University National Center for Atmospheric Research Mesoscale Model (MM5). This work is applied to the MM5-based Regional Data Assimilation and Prediction System model (RDAPS), the Korea Meteorological Administration s regional forecast model. The bogus procedure starts by initializing the winds within the bogus area. The main difficulty lies in the generation of other variables, such as humidity, temperature, geopotential, etc., which are dynamically consistent with the prescribed wind. However, it was found that there is a simple and practical way of tropical cyclone (TC) initialization. It is achieved by the use of the built-in function of MM5, the four-dimensional data assimilation (FDDA). In order to do so, a miniature RDAPS is constructed. After the initialization of wind within the filter region, all other variables are generated by the model through a strong 24-h nudging to the prescribed wind. It is found after careful analyses that there is an improvement over the no-bogus model. Failures are mostly due to the fake vortex or the spurious deepening of the vortex, which have been problems of the original RDAPS model. The bogus RDAPS never cures the failure of the original model. 1. Introduction The community fifth-generation Pennsylvania State University National Center for Atmospheric Research Mesoscale Model (MM5) has contributed greatly to a very wide range of meteorological fields including tropical cyclone (TC) modeling/prediction (Liu et al. 1999; Bao et al. 2000; Xiao et al. 2000). In TC modeling, there are two requirements: a fine numerical model and a proper TC bogus technique. Regional Data Assimilation and Prediction System (RDAPS), the operational model of the Korea Meteorological Administration (KMA) has gone through many upgrades since it was initially in operation in The current version of RDAPS based on PSU NCAR MM5, version 2 (Grell et al. 1995) runs twice daily on the supercomputer (NEC SX5/12A). RDAPS uses a grid distance of 30 km in a grid system so that the domain is wide enough to cover the KMA tropical cyclone watch area (west of 140 E and north of 20 N). The model has fine physics on a very wide domain so that it may serve not only as a regional forecast model but also as a typhoon Corresponding author address: Dr. H. Joe Kwon, Department of Atmospheric Science, Kongju National University, Kongju, Chungnam South Korea. hjkwon@kongju.ac.kr model upon successful implementation of the TC initialization. The GFDL hurricane prediction model has a sophisticated state-of-the-art TC initialization procedure (Kurihara et al. 1995, hereafter KBTR). The analysis fields are decomposed into basic and disturbance fields by scale consideration. With the use of the 850-hPa disturbance wind, the bogus region is carefully determined. A target wind is then constructed by including the nonhurricane component and observations as well as the beta gyre. The target wind becomes a forcing to generate the axisymmetric component of all the other variables within the context of the axisymmetric version of the original model. The GFDL model has also gone through some evolutions, such as three-dimensional bogus TC generation (Kurihara et al. 1997) and the inclusion of hurricane ocean interaction (Bender and Ginis 2000). In the current work, an attempt is made to apply the GFDL TC bogus algorithm to the community model MM5 completely for the first time. As in KBTR, winds are initialized in a straightforward manner within the filter region surrounded by 24 boundary points. The main difficulty lies in the generation of other variables, such as humidity, temperature, geopotential, etc., which are dynamically consistent with the prescribed wind. In KBTR an axisymmetric version of the original model 2002 American Meteorological Society

2 DECEMBER 2002 KWON ET AL is run to generate the other synoptic variables. If we follow the similar procedure, we should write a computer code for the axisymmetric version of MM5, which may require a tremendous amount of additional work. However, we found that there is a simple and very practical way of TC initialization. It can be handled by the use of the built-in function of MM5, the four-dimensional data assimilation. This method is comparable to the way that a three-dimensional TC bogus is generated in the GFDL hurricane model (Kurihara et al. 1997). Section 2 describes the details of the bogus algorithm. This TC bogus method is applied to some of the TC predictions of east Asia in The forecast cases are described in section 3. The conclusions and discussion follow in section Procedure a. Spatial filter The RDAPS model follows the pre-processing procedure of MM5, version 2 and uses the following modules 1) TERRAIN, which sets up the model domain, grid size and produces the terrain elevation, sea surface temperature, etc., 2) DATAGRID, which interpolates the global analysis forecast field into the model grids, 3) RAWINS, which blends the first guess field with observational data, 4) INTERP, which interpolates data into the model sigma level and 5) MM5, the main model. After the DATAGRID is finished, we take the latitude and longitude information of the model grids and the horizontal winds from the DATAGRID output. The first step of GFDL TC initialization is to find the filter domain by examining the disturbance wind at 850 hpa. The filter domain is an area where the vortex from the global analysis will be replaced by the bogus vortex. The disturbance wind is obtained by subtracting the basic wind that consists of the large-scale motions from the total wind. We apply the spatial smoother to the 850-mb wind to obtain the basic wind. We elaborate here only on the GFDL wind initialization that differs from that of KBTR. When separating the wind into the basic and the disturbance wind, a spatial smoother is applied. The KBTR spatial smoother is not directly applicable in our case because the coefficients in the smoother are for a grid distance of 1, while the grid distance of our data is 30 km. Therefore, we modify the smoother as follows: L h i,j hi,j K(hi 4,j hi 4,j 2h i,j ). (1) The application of (1) with the same K in KBTR has a similar effect to 1 space, resulting in filtering out smallscale features except those shorter than 4 waves. And then, in order to remove waves shorter than 4 wavelength, L L L L h i,j h i,j K(h i 1,j h i 1,j 2h i,j) (2) is applied by varying m 2, 3, 4, 2. Upon the application of the above two smoothers, more than 95% of the features with less than 1000-km wavelength are filtered and the amplitudes of those with 2000-, 3000-, and 4000-km wavelength are reduced by 62%, 34%, and 24%, respectively. The spatial smoothing is completed by applying the above two-step operator also in the meridional direction. We choose a case for Tropical Storm Bolaven at 1200 UTC 28 July According to the analysis of Regional Specialized Meteorological Center (RSMC) Tokyo, Bolaven s central pressure is 980 hpa and the maximum wind is 55 kt. Figure 1a shows the basic component of the 850-hPa wind resulting from the application of the spatial smoother. Anticyclonic large-scale circulation on the east side of the storm and a largescale cyclonic circulation (monsoon gyre) surrounding Bolaven are clearly retained even after the smoothing. The disturbance component (Fig. 1b) is obtained by subtracting the basic component from the total wind. b. Determination of the filter domain The next step is to determine the TC bogus domain (or filter domain). We follow the similar procedure of KBTR. First we determine the filter center by examining the azimuthally averaged tangential wind profile. Next, we scan radially outward for 24 azimuthal directions to search for the boundary that separates the TC region and the environment. KBTR suggests two conditions in finding the filter radius. Radial scanning is done at the point 1) where the condition tan 6ms 1 and tan / r s 1 is met for the second time, and 2) tan 3ms 1 is met. The global analysis of Global Data Assimilation and Prediction System (GDAPS) includes the TC bogus in the routine data assimilation and prediction cycle so that the use of 6 m s 1 in the GDAPS analysis data tends to overestimate the filter radius. Therefore, for our case we change the value of the first condition from 6 m s 1 to 7.5 m s 1. The filter domain for the previous Tropical Storm Bolaven case is shown in Fig. 1b. Clear separations are seen between the storm and the surrounding environment including the cyclonic circulation northwest of the storm and the anticyclonic circulations southeast and to the northeast side of the storm. One must note that searching for the filter radius for one direction is completely independent of another, so that there may be a situation where one of the filter boundary points may exhibit a significant protrusion. We see such a case in the northeast and the south part of the filter boundary in Fig. 1b. This may adversely affect obtaining the non- TC component by the optimum interpolation, which will be discussed later. In such a case, we adjust the point slightly inside. Specifically, the following treatments are used: If a certain filter radius exceeds 1.3 times the average of the two adjacent ones, it is reduced to 1.1 times. The modified filter boundary is also shown in

3 2968 MONTHLY WEATHER REVIEW FIG. 1. The 850-hPa (a) basic component and (b) disturbance winds associated with Tropical Storm Bolaven (0006) at 1200 UTC 28 Jul (b) Filter boundary is shown. Fig. 1b. The rest of the procedure determining the boundary is identical to that of KBTR. c. Little RDAPS 1) SYNOPSIS The TC initialization generates the complete threedimensional bogus vortex within the filter domain. It includes the initialization of all synoptic variables as well as specifying the bogus wind as in KBTR. The GFDL-type wind initialization is not simple, but it can be readily done through the straightforward procedure. The difficulty is the initialization of other variables such as humidity, temperature, geopotential, etc. that are dynamically consistent with the already prescribed wind. In KBTR it is done by integrating the axisymmetric version of the original model. During the time of integration, the tangential component of wind is gradually forced toward the target wind profile based on the storm information provided by the National Hurricane Center. If we follow a similar procedure, we should write a computer code for the axisymmetric version of MM5, which may require a tremendous amount of additional work. Meanwhile, the next version of the GFDL hurricane prediction model (Kurihara et al. 1997) uses three-dimensional vortex generation. For a given threedimensional target state that slowly varies from the calm state, the other variables are also slowly generated by forcing the model. MM5 has a built-in function that can do the above forcing, namely, four-dimensional data assimilation (FDDA). The idea of FDDA is to combine current and past data in an explicit dynamical model such that the model s prognostic equations provide time continuity and dynamic coupling among the various fields (Grell et al. 1995). FDDA is accomplished though Newtonian relaxation or nudging (Hoke and Anthes 1976). The nudging relaxes the model state toward the target state (usually data blended with observation). In actual application of FDDA in MM5, a user just turns on the FDDA switch and prescribes the target state to be nudged onto. Since we need the TC structure only within the bogus area through the built-in function of MM5, we construct a miniature of the original model that is identical to the original except for a smaller horizontal domain. This is done simply for cost-effectiveness. We will call it little RDAPS (Fig. 2). The little RDAPS is constructed with grid system, which corresponds to a 3000 km 3000 km domain. The center of the little RDAPS domain is determined by the center of the filter domain in the previous step. The little RDAPS is located between the DATAGRID and RAWINS step and runs for every TC that exists over the KMA s TC watch area. In order to do so, TERRAIN is run to generate data regarding the model domain for each TC. The locations of the model grids should be different from those of the original model. DATAGRID is also run again to put the initial data into the little RDAPS grid. We take DA- TAGRID output to construct the three-dimensional bogus wind. We separate the basic and the disturbance wind and obtain the non-tc wind with the use of information on the bogus boundary that is transferred from the previous stage. We then construct the axisymmetric wind with the use of the empirical formula (Hol-

4 DECEMBER 2002 KWON ET AL FIG. 2. Flow chart showing the procedure of tropical cyclone bogus. The little RDAPS runs for every TC. land 1980) and the TC information provided by the typhoon center. Detailed wind initialization will be given in section 2c(2). All the other variables are generated through the nudging to the target wind, which will be discussed in detail in section 2c(3). After that, data are put back into the original RDAPS domain, which requires a special treatment because the grid mesh in the little RDAPS is different from that in RDAPS. Not only an interpolation, but also an adjustment of the wind are needed because of the different axes. Then, the TC initialization is completed. 2) CONSTRUCTION OF WIND The disturbance wind inside the bogus domain consists of non-tc components as well as TC components. Assuming that winds at the filter boundary are 100% non-tc wind, we obtain the non-tc winds inside the filter domain by optimum interpolation using the wind values at the 24 boundary points as in KBTR. Figure 3 shows the results. We see the smooth inflow and outflow crossing the boundary associated with the environmental circulation, which could have been discarded without considering the non-tc component in this way. The rest are the TC winds analyzed in the global model. We then replace it with the synthetic observation utilizing the empirical formula (Holland 1980) and the TC information reported by RSMC Tokyo, the regional typhoon center. Specifically, information such as the maximum wind, central pressure, 50-kt wind radius (if it exists) and 30-kt wind radius are used when constructing the axisymmetric wind profile. When RSMC reports more than one number in the 50-kt or 30-kt wind radii for an asymmetric TC, we use the average of the two. Asymmetry of the TC should be taken into account by the basic and the non-tc component winds. We elaborate further the procedure of constructing the axisymmetric wind because there is something that needs to be described specifically. From the empirical surface pressure profile (Holland 1980), we construct the cyclostrophic wind and the gradient wind profiles. We need to determine two coefficients to do so. The radius of the maximum wind (RMW) and one of the coefficients are solved by vanishing the radial gradient of the cyclostrophic wind. For the other coefficient, the 30-kt wind radius reported by the typhoon center is used. We find the coefficient, which yields the gradient wind of 30 kt at the reported 30-kt wind radius. In doing so, the equation cannot be directly solved so that we get solution by iteration. During this procedure, the iteration sometimes does not converge to the solution. This happens when the previously determined radius of maximum wind and the corresponding coefficient are not consistent with the outer wind structure given by the gradient wind relationship. When this occur, we arbitrarily prescribe the radius of maximum wind as reasonably as possible depending upon the magnitude of the maximum wind. The next task is to combine the cyclostrophic wind and the gradient wind. We introduce a weighting function w that decreases linearly from 1

5 2970 MONTHLY WEATHER REVIEW weights are 0.95, 1.00, 0.97, 0.88, 0.82, 0.65, 0.40, and 0.35 for p 1000, 850, 700, 500, 400, 300, 250, and 200, hpa, respectively, and 0 above p 150 hpa. The basic wind is added to the sum of the axisymmetric wind and the non-tc wind. This is the wind of the bogus TC, which will serve as the target state in nudging. 3) FOUR-DIMENSIONAL DATA ASSIMILATION FIG. 3. Non-TC wind component at the 850-hPa level for the case shown in Fig. 1. at 1.2 times the RMW to 0 at 2.5 times the RMW. Therefore, the axisymmetric wind is given as tan cw gr (1 w), (5) where tan is the axisymmetric tangential wind, c and gr are the cyclostrophic and the gradient wind, respectively. When constructing the three-dimensional bogus wind, the axisymmetric wind is vertically weighted. The The DATAGRID output is modified with the new bogus wind. At this stage, the wind and the other variables are not in dynamical balance. We expect that the dynamical inconsistency will disappear at the end of the four-dimensional data assimilation. We assume that the target state does not change in time during the nudging. For practical purposes, exactly the same dataset is put in order for every 12 h during all the nudging periods. Then, we turn on the FDDA switch in MM5 and perform the analysis nudging to the wind. Since RDAPS is an operational model, we need to consider the timeliness for the operational schedule, meaning that there should be a compromise between perfection and practicality. Tests have shown that 24-h nudging suffices for practical purposes. After 24 h, all the variables except the target wind do vary in time, but only to the extent of the nonlinear quasi equilibrium (Kwon and Williams 2000). Figure 4 shows the mean sea level pressure (MSLP) field before (Fig. 4a) and after (Fig. 4b) the 24-h FDDA period. It is evident that the original smooth and loose vortex becomes sharp and tight. The MSLP at the vortex center drops from 990 to 983 hpa and the geopotential height at the 700-hPa level drops from 2972 to 2921 m. Monsoon gyrelike circulation surrounding Tropical Storm Bolaven becomes clearer than before the bogus FIG. 4. The mean sea level pressure field in the little RDAPS domain (a) before and (b) after 24-h FDDA is completed.

6 DECEMBER 2002 KWON ET AL FIG. 5. The mean sea level pressure (a) before and (b) after bogus for the case shown in Fig. 1. procedure, so that one may expect that the TC will move toward the north. d. Back to the original domain After we obtain the complete bogus TC in the little RDAPS, we need to restore all the data to the bogus region of the original RDAPS domain. In doing so, three factors have to be considered. First, when performing the interpolation, special care is needed for each variable since some variables are at cross points and some other variables are at dot points in the staggered grid system. Second, if we literally restore the bogus field of the little RDAPS to the original RDAPS bogus region, there always occurs a discontinuity across the filter boundary. Third, wind definitions of the two systems are different because of the different axes so that there must be an adjustment of the wind. Let be the angle between the two coordinate systems. Specifically, is defined as positive if the little RDAPS domain is on the right side of the original domain. Let (u O, O ) and (u L, L ) be the wind components of the original and the little RDAPS domains, respectively. Then, the winds in the original model domain can be obtained from the following coordinate transform rule: O L uo cos sin ul. (6) sin cos All other variables are just put back into the original RDAPS grids, only inside of the filter domain. When the 24-h nudging is finished, all variables except the wind in the little RDAPS go through a great change. The largest change occurs near the storm center. If we restore data to the original grids, there will be a discontinuity across the boundary. Therefore, for any field variables, we let the following bogus parameter where w (1 w), b n (7) 1, r r o/2 w 2 r r /2 (8) o exp 4 r r o/2 r o/2 [ ] and r o is the mean value of the 24 filter radii; the sub- scripts b and e refer to the bogus parameter and the environment, respectively. Figures 5a and 5b show the mean sea level pressure fields before and after the bogus, respectively. Even though the sharp and tight vortex is put into the original domain, the discontinuity between the bogus TC and the environment disappears so that one cannot figure out where the bogus boundary is. e. Multiple tropical cyclones Often, there may be a time when there are more than one TC in the model domain. In this case, the previous procedure of generating the bogus vortex is applied to each TC. The procedure includes separating the basic state and the disturbance state, determining the filter area, obtaining the non-tc wind component within the bogus region as well as generating all the other variables except winds by using the little RDAPS individually (Fig. 2). 3. Forecasts Several forecasts are made using the previous bogus algorithm for two TC s in the year Nine cases of Tropical Storm Bolaven (0006) and ten cases of Tropical Storm Jelawat (0008) are taken. The computational procedure is exactly the same as in operations, such as the

7 2972 MONTHLY WEATHER REVIEW TABLE 1. Mean forecast track error in km for the cases when the original RDAPS produces reasonable forecasts. The numbers in parentheses at the top are the number of cases compared and those in the BOGUS row are the number of cases where TC bogus actually produces better forecasts. NO BOGUS BOGUS 12 h (9) 24 h (9) 36 h (8) 48 h (5) 60 h (4) 72 h (2) (5) (6) (4) (3) (3) (1) forecast fields of the global model supplied for the lateral boundary and 12-h FDDA with blending with realtime observations through the global telecommunication system. Although a tremendous amount of work is done in expectation of a great improvement of the model performance for the TC prediction, the results are not quite so. The trends found in the original model are retained also in the bogus version. If the model TC moves to a certain direction in the forecast of the original RDAPS, the bogus version also shows similar behavior. The bogus RDAPS never cures a correction of the failure of the original model. However, we have found after careful examination that if the original RDAPS produces a reasonable forecast, the TC bogus helps to produce a better forecast. We have separated the 19 forecast cases, into two groups: good forecasts (9 cases) and bad forecasts (10 cases). The two groups are separated by a 48-h track prediction error. If the 48- h prediction error exceeds 400 km, we take it as poor forecast. Tables 1 and 2 show the mean forecast track error for the 9 cases when the original RDAPS produces reasonable forecasts and for the 10 cases when original RDAPS produces erroneous forecasts, respectively. The numbers in parentheses at the top are the number of cases compared. In Table 1 the numbers in parentheses in the BOGUS row shows the number of cases where TC bogus actually produces better forecasts, which implies that the bogus improves most of the reasonable RDAPS forecast. Likewise, it may be said that the bogus procedure is not helpful for the failure of the original model (Table 2). The tropical cyclone motion is mainly influenced by the environmental flow. The typhoon initialization modifies the initial state in a very small area near the typhoon center. The environmental flow does not change much. Therefore one cannot expect that the typhoon track improves very significantly if the original model does not predict well. This suggests that we may also need a better assimilation with additional information outside the typhoon area as well as the proper TC initialization in order to improve the typhoon track prediction. In order to demonstrate that bad forecasts are due to an inherent problem of the original model that seems to be related to a spurious convective activity, we show the observed track of Typhoon Bolaven and the forecast tracks by the original RDAPS and by the bogus version starting at 1200 UTC 27 July 2000 (Fig. 6). The initial and the 36-h-forecast mean sea level pressure fields are also shown for this purpose (Fig. 7). Evidently the forecast track shows a profound difference from the observed track. At this time, Bolaven is located in the western periphery of the subtropical high pressure region, which is elongated in a north south direction. In addition, the storm is surrounded by the very large-scale cyclone, Monsoon Gyre (Lander 1994). According to the systematic approach (Carr et al. 1997), the current synoptic situation is in the poleward (P) region of gyre (G) pattern. Gyre is about to diminish so that the synoptic environment for Bolaven is about to change to poleward pattern. For a synoptic forecaster it is not difficult to expect that Bolaven will move northward even without referring to the numerical model guidance. On the contrary, the model storm penetrates deeply into the center of the subtropical high. Figure 7 shows the model results for this case. The mean sea level pressure at initial model time (1200 UTC 27 July 2000) and 36 h later are shown in Figs. 7a and 7b, respectively. What happens is that a spurious vortex ( boguscane, artificial hurricane created by numerical model) begins to grow faster after 12 hours or so from very weak trough located in the southwest side of the center of Bolaven. At the same time the strong cyclonic circulation of Bolaven rotates the boguscane. At about hour 36, the boguscane grows to about the same intensity of the real Tropical Storm Bolaven and begins to interact directly with the real storm (Fujiwhara 1921; Brand 1970). This is the reason why the forecast results in this absurd outcome. This spurious vortex is also seen in the original model with a slightly different magnitude, which results in a similar absurd forecast track (Fig. 6). The modelers of KMA have been observing many boguscanes over the sea since they reduced the model grid from 40 to 30 km and expanded the model domain wide enough to cover the KMA TC watch area. It may have TABLE 2. Mean forecast track error in km for the cases when the original RDAPS produces extremely erroneous forecasts. The numbers in parentheses at the top are the number of cases compared and those in the BOGUS row are the number of cases for which TC bogus actually produces worse forecasts. NO BOGUS BOGUS 12 h (10) 24 h (10) 36 h (9) 48 h (9) 60 h (7) 72 h (5) (4) (7) (6) (6) (5) (4)

8 DECEMBER 2002 KWON ET AL FIG. 6. Observed 6-hourly track of Tropical Storm Bolaven ( ), forecast 6-hourly tracks by the original RDAPS ( ), and by the bogus version ( ) starting at 1200 UTC 27 Jul something to do with the improper choice of cumulus physics at least for the prediction of tropical cyclones. RDAPS currently uses the Kain Fritsch scheme (Kain and Fritsch 1992) for cumulus parameterization. Since RDAPS is the regional forecast model, in that all physical options had been already tuned toward the best performance for meteorological phenomena including tropical cyclones, there is not much to be done to resolve this problem at this moment. Most of the bad-forecast cases that we have observed seem to be related to problems associated with boguscanes or the spurious deepening of the vortex. In this sense the performance of the current RDAPS in terms of TC prediction is not satisfactory. This does not mean that TC prediction by RDAPS is useless when the model produces an extremely bad forecast (e.g., as in Fig. 6), because model failure like boguscane can be easily discerned by the forecaster on duty by examining the forecast synoptic field (e.g., as in Fig. 7). In this sense, the current bogus work certainly proves to be of help in improving the forecast. 4. Summary and discussion The GFDL TC bogus algorithm is successfully implemented to the KMA s regional forecast model RDAPS, which is based upon Penn State NCAR MM5, version 2. The bogus is put between the DATAGRID and the RAWINS steps. Using the 850-hPa wind taken from the DATAGRID output, the filter domain is determined from the disturbance part of the wind. The disturbance wind is obtained by subtracting the basic wind from the total wind. In order to obtain the basic FIG. 7. (a) Mean sea level pressure at initial model time (1200 UTC 27 Jul 2000) and (b) that at 36 h later. Tropical Storm Bolaven at 36 h is located at E, 27.4 N. The vortex at about 137 E, 29 N is the boguscane. wind, we apply a similar smoother slightly modified from KBTR s. Then, we scan outward from the filter center to determine the filter radius for every 24 azimuthal direction. Next, we construct the miniature RDAPS (or, little RDAPS) in order to generate the three-dimensional bogus TC that is dynamically consistent with the prescribed bogus wind. The little RDAPS is constructed with grids, which corresponds to a 3000 km 3000 km domain. The center of the little RDAPS domain is determined by the center of the filter domain in the previous step. The little RDAPS is run for every TC that exists over the KMA s TC watch area. The wind within the filter domain of the little RDAPS is replaced

9 2974 MONTHLY WEATHER REVIEW by the sum of the basic wind, non-tc wind component, and the axisymmetric wind compiled by the empirical formula (Holland 1980). At this stage, the wind and the other variables are not in dynamical balance. We assume that the bogus target wind does not change in time during the nudging. For practical purposes, exactly the same dataset is put in order for every 12 h during all the nudging periods. Then the FDDA switch in MM5 is turned on and the analysis nudging to the wind is performed. Tests have shown that 24-h nudging suffices for operational purposes. Forecasts are made using this current bogus algorithm for two tropical cyclones in the year Nine cases of Tropical Storm Bolaven and ten cases of Tropical Storm Jelawat are taken. The computational procedure is exactly the same as in operations such as the forecast fields of the global model supplied for the lateral boundary and 12-h FDDA with blending with real-time observations through the global telecommunication system. Although a tremendous amount of work is done in expectation of a great improvement of the model performance for the TC prediction, the results are not quite so. The trends found in the original model are retained also in the bogus version. The bogus RDAPS never cures the failure of the original model. However, we have found after careful examination that if the original RDAPS produces a reasonable forecast, the TC bogus helps to produce a better forecast. Most of the bad-forecast cases that we have observed are related to problems associated with boguscanes. In addition to the boguscanes, a sudden spurious deepening and intensifying of the target TC seems to result in bad forecasts. In this sense the performance of current RDAPS in terms of TC prediction is not satisfactory. This does not mean that TC prediction by RDAPS is useless because when the model produces an extremely bad forecast, model failure like boguscane can be discerned by the forecaster on duty by examining the forecast synoptic field. Besides, the current bogus work certainly proves to be of help in improving the forecast when the original RDAPS produces normal forecasts. Perhaps the main purport of the work is that GFDL TC initialization scheme is implemented in the community model MM5 for the first time. The current work is done over the KMA regional forecast model, which uses a single nest of 30-km grid distance. If we apply the current algorithm to the multiple nests with a proper physics, the performance of the model relative to the TC prediction should be improved. This prompts further investigation. Acknowledgments. This work is supported by the project A study on improving weather forecast skill using a supercomputer of Meteorological Research Institute KMA, One of the authors (HJK) thanks Mr. Christian Olson for reviewing the draft of the manuscript. We also wish to express our sincerest thanks to all the anonymous reviewers for their effort in improving the manuscript. REFERENCES Bao, J.-W., J. M. Wilczak, J.-K. Chio, and L. H. Kantha, 2000: Numerical simulation of air sea interaction under high wind conditions using a coupled model: A study of hurricane development. Mon. Wea. Rev., 128, Bender, M., and I. Ginis, 2000: Real-case simulations of hurricane ocean interaction using a high-resolution coupled model: Effects on hurricane intensity. Mon. Wea. Rev., 128, Brand, S., 1970: Interaction of binary cyclones of the western North Pacific Ocean. J. Appl. Meteor., 9, Carr, L. E., R. L. Elsberry, and M. Boothe, 1997: Condensed and updated version of the systematic approach meteorological knowledge base western North Pacific. Tech. Rep. NPS-MR , Naval Postgraduate School, Monterey, CA, 169 pp. Fujiwhara, S., 1921: The natural tendency towards symmetry of motion and its application as a principle of motion. Quart. J. Roy. Meteor. Soc., 47, Grell, G. A., J. Dudhia, and D. R. Stauffer, 1995: A description of the fifth-generation Penn State/NCAR mesoscale model (MM5). NCAR Tech. Note NCAR/TN-398 STR, 122 pp. Hoke, J. E., and R. A. Anthes, 1976: The initialization of numerical models by a dynamical initialization technique. Mon. Wea. Rev., 104, Holland, G. J., 1980: An analytic model of the wind and pressure profiles in hurricanes. Mon. Wea. Rev., 108, Kain, J. S., and J. M. Fritsch, 1992: Convective parametrization for mesoscale models: The Kain Fritsch scheme. The Representation of Cumulus in Numerical Models of the Atmosphere, Meteor. Monogr, No. 46, Amer. Meteor. Soc., Kurihara, Y., M. A. Bender, R. E. Tuleya, and R. J. Ross, 1995: Improvements in the GFDL hurricane prediction system. Mon. Wea. Rev., 123, ,, and, 1997: For hurricane intensity forecast: Formulation of a new initialization method for the GFDL hurricane prediction model. Preprints, 22d Conf. on Hurricane and Tropical Meteorology, Fort Collins, CO, Amer. Meteor. Soc., Kwon, H. J., and R. T. Williams, 2000: Nonlinear equilibration of barotropic waves in a zonally nonuniform basic current. J. Atmos. Sci., 57, Lander, M. A., 1994: Description of a monsoon gyre circulation and its effects on the tropical cyclones in the western North Pacific during August Wea. Forecasting, 9, Liu, Y., D.-L. Zhang and M. K. Yau, 1999: A multiscale numerical study of Hurricane Andrew (1992). Part II: Kinematics and innercore structures. Mon. Wea. Rev., 127, Xiao, Q., X. Zou, and B. Wang, 2000: Initialization and simulation of a landfalling hurricane using a variational bogus data assimilation scheme. Mon. Wea. Rev., 128,