• Barker, D. M., , W. Huang, , Y-R. Guo, , A. Bourgeois, , and Q. Xiao, 2004: A three-dimensional variational data assimilation system for MM5: Implementation and initial results. Mon. Wea. Rev., 132 , 897914.

    • Search Google Scholar
    • Export Citation
  • Dudhia, J., 1989: Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46 , 30773107.

    • Search Google Scholar
    • Export Citation
  • Hong, S-Y., , and H-L. Pan, 1996: Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev., 124 , 23222339.

    • Search Google Scholar
    • Export Citation
  • Hong, S-Y., , J. Dudhia, , and S-H. Chen, 2004: A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation. Mon. Wea. Rev., 132 , 103120.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., , and J. M. Fritsch, 1990: A one-dimensional entraining/detraining plume model and its application in convective parameterization. J. Atmos. Sci., 47 , 27842802.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., , and J. M. Fritsch, 1993: Convective parameterization for mesoscale model: The Kain–Fritsch scheme. The Presentation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 165–170.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., , M. A. Bender, , and R. J. Ross, 1993: An initialization scheme of hurricane models by vortex specification. Mon. Wea. Rev., 121 , 20302045.

    • Search Google Scholar
    • Export Citation
  • Lim, J-O. J., , and S-Y. Hong, 2005: Effects of bulk ice microphysics on the simulated monsoonal precipitation over east Asia. J. Geophys. Res., 110 , D24201. doi:10.1029/2005JD006166.

    • Search Google Scholar
    • Export Citation
  • Park, K., , and X. Zou, 2004: Toward developing an objective 4DVAR BDA scheme for hurricane initialization based on TPC observed parameters. Mon. Wea. Rev., 132 , 20542069.

    • Search Google Scholar
    • Export Citation
  • Parrish, D. F., , and J. C. Derber, 1992: The National Meteorological Center’s spectral statistical-interpolation analysis system. Mon. Wea. Rev., 120 , 17471763.

    • Search Google Scholar
    • Export Citation
  • Pu, Z-X., , and S. A. Braun, 2001: Evaluation of bogus vortex techniques with four-dimensional variational data assimilation. Mon. Wea. Rev., 129 , 20232039.

    • Search Google Scholar
    • Export Citation
  • Skamarock, W. C., , J. B. Klemp, , J. Dudhia, , D. O. Gill, , D. M. Barker, , W. Wang, , and J. G. Powers, 2005: A description of the Advanced Research WRF version 2. NCAR Tech. Note TN-468+STR, 88 pp.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., , and J. Sun, 2007: Multiple-radar data assimilation and short-range quantitative precipitation forecasting of a squall line observed during IHOP_2002. Mon. Wea. Rev., 135 , 33813404.

    • Search Google Scholar
    • Export Citation
  • 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 , 22522269.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., , Y-H. Kuo, , Y. Zhang, , D. M. Barker, , and D-J. Won, 2006: A tropical cyclone bogus data assimilation scheme in the MM5 3D-Var system and numerical experiments with Typhoon Rusa (2002) near landfall. J. Meteor. Soc. Japan, 84 , 671689.

    • Search Google Scholar
    • Export Citation
  • Zhang, X., , Q. Xiao, , and P. J. Fitzpatrick, 2007: The impact of multisatellite data on the initialization and simulation of Hurricane Lili’s (2002) rapid weakening phase. Mon. Wea. Rev., 135 , 526548.

    • Search Google Scholar
    • Export Citation
  • Zou, X., , and Q. Xiao, 2000: Studies on the initialization and simulation of a mature hurricane using a variational bogus data assimilation scheme. J. Atmos. Sci., 57 , 836860.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    The best tracks of seven hurricanes (Charley, Frances, Ivan, Jeanne, Katrina, Rita, and Wilma) in the 2004 and 2005 seasons. The hurricane positions are shown with filled circles at 0000 UTC, open circles at 1200 UTC, and plus signs at 0600 and 1800 UTC. The shading in the map is terrain height, with the scale on the right.

  • View in gallery

    The mean absolute errors of the (top) hurricane track, (middle) maximum surface wind, and (bottom) central sea level pressure for the forecasts from statistics of 21 cases listed in Table 1. The light and dark bars are the statistical results for the control experiment and BDA experiment, respectively. The errors of 24-, 48-, and 72-h forecasts with their averages are shown.

  • View in gallery

    Hurricane Katrina at 0000 UTC 26 Aug 2005 for the (a) CT and (b) GB experiments. The solid lines are SLP, with a contour interval of 2.5 hPa, and the barbs show the 10-m wind field (a full barb represents 5 m s−1).

  • View in gallery

    Analyses of Hurricane Humberto at 1200 UTC 12 Sep 2007 showing (top) SLP (thick isolines), surface temperature (thin isolines), surface wind barbs, and surface wind speed (shading) for the (a) CT and (b) GB experiments and (bottom) 500-hPa geopotential height (isolines), wind barbs, and absolute vorticity (shading) for the (c) CT and (d) GB experiments.

  • View in gallery

    The 72-h track forecast for Hurricane Humberto from 1200 UTC 12 to 1200 UTC 15 Sep 2007. The dashed line with the open circles is the best track, the solid line with asterisks is the forecast from the CT experiment, and the gray line with the open triangles is the forecast from the GB experiment. The date/hour are shown in boxes.

  • View in gallery

    The 72-h intensity forecast for Hurricane Humberto from 1200 UTC 12 to 1200 UTC 15 Sep 2007: (a) CSLP and (b) MSW. The solid line is the forecast from the GB experiment, and the dashed line is from the best-track observation.

  • View in gallery

    The 36-h forecasts of Hurricane Humberto at 0000 UTC 14 Sep 2007 by (a) GB and (b) GB with increased surface friction in the simulation. The thin solid isolines are for SLP (2-hPa interval). The surface winds are shown with wind barbs (a full barb represents 5 m s−1) and thick gray isolines (2 m s−1 interval).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 65 65 14
PDF Downloads 25 25 2

Evaluations of BDA Scheme Using the Advanced Research WRF (ARW) Model

View More View Less
  • 1 Mesoscale and Microscale Meteorology Division, Earth and Sun Systems Laboratory, National Center for Atmospheric Research, * Boulder, Colorado
  • | 2 Mesoscale and Microscale Meteorology Division, Earth and Sun Systems Laboratory, National Center for Atmospheric Research, * Boulder, Colorado, and Institute of Atmospheric Environment, China Meteorological Administration, Shenyang, China
  • | 3 Mesoscale and Microscale Meteorology Division, Earth and Sun Systems Laboratory, National Center for Atmospheric Research, * Boulder, Colorado
© Get Permissions
Full access

Abstract

A tropical cyclone bogus data assimilation (BDA) scheme is built in the Weather Research and Forecasting three-dimensional variational data assimilation system (WRF 3D-VAR). Experiments were conducted (21 experiments with BDA in parallel with another 21 without BDA) to assess its impacts on the predictions of seven Atlantic Ocean basin hurricanes observed in 2004 (Charley, Frances, Ivan, and Jeanne) and in 2005 (Katrina, Rita, and Wilma). In addition, its performance was compared with the Geophysical Fluid Dynamics Laboratory (GFDL) hurricane initialization scheme in a case study of Hurricane Humberto (2007). It is indicated that hurricane initialization with the BDA technique can improve the forecast skills of track and intensity in the Advanced Research WRF (ARW). Among the three hurricane verification parameters [track, central sea level pressure (CSLP), and maximum surface wind (MSW)], BDA improves CSLP the most. The improvement of MSW is also considerable. The track has the smallest, but still noticeable, improvement. With WRF 3D-VAR, the initial vortex produced by BDA is balanced with the dynamical and statistical balance in the 3D-VAR system. It has great potential for improving the hurricane intensity forecast. The case study on Hurricane Humberto (2007) shows that BDA performs better than the GFDL bogus scheme in the ARW forecast for the case. Better definition of the initial vortex is the main reason for the advanced skill in hurricane track and intensity forecasting in this case.

Corresponding author address: Dr. Qingnong Xiao, NCAR, MMM, P.O. Box 3000, Boulder, CO 80307-3000. Email: hsiao@ucar.edu

Abstract

A tropical cyclone bogus data assimilation (BDA) scheme is built in the Weather Research and Forecasting three-dimensional variational data assimilation system (WRF 3D-VAR). Experiments were conducted (21 experiments with BDA in parallel with another 21 without BDA) to assess its impacts on the predictions of seven Atlantic Ocean basin hurricanes observed in 2004 (Charley, Frances, Ivan, and Jeanne) and in 2005 (Katrina, Rita, and Wilma). In addition, its performance was compared with the Geophysical Fluid Dynamics Laboratory (GFDL) hurricane initialization scheme in a case study of Hurricane Humberto (2007). It is indicated that hurricane initialization with the BDA technique can improve the forecast skills of track and intensity in the Advanced Research WRF (ARW). Among the three hurricane verification parameters [track, central sea level pressure (CSLP), and maximum surface wind (MSW)], BDA improves CSLP the most. The improvement of MSW is also considerable. The track has the smallest, but still noticeable, improvement. With WRF 3D-VAR, the initial vortex produced by BDA is balanced with the dynamical and statistical balance in the 3D-VAR system. It has great potential for improving the hurricane intensity forecast. The case study on Hurricane Humberto (2007) shows that BDA performs better than the GFDL bogus scheme in the ARW forecast for the case. Better definition of the initial vortex is the main reason for the advanced skill in hurricane track and intensity forecasting in this case.

Corresponding author address: Dr. Qingnong Xiao, NCAR, MMM, P.O. Box 3000, Boulder, CO 80307-3000. Email: hsiao@ucar.edu

1. Introduction

Bogus data assimilation (BDA) is a technique proposed by Xiao et al. (2000) and Zou and Xiao (2000) that assimilates a synthetic vortex using variational data assimilation to initialize the hurricane structure. With the optimization (minimization) procedure, the synthetic vortex structures are gradually incorporated into the hurricane initial conditions. The generated initial vortex by BDA not only fits the bogus vortex structure but also is consistent with the model resolution and physics in four-dimensional variational (4D-Var) data assimilation or the dynamical and statistical balance constraints in three-dimensional variational (3D-Var) data assimilation (Xiao et al. 2006). Previous studies with the fifth-generational Pennsylvania State University–National Center for Atmospheric Research (NCAR) Mesoscale Model (MM5) indicated that BDA was able to improve the forecasts of both hurricane track and intensity (Pu and Braun 2001; Park and Zou 2004; Xiao et al. 2006; Zhang et al. 2007).

As a replacement for MM5, the Advanced Research Weather Research and Forecasting (WRF) model (ARW; Skamarock et al. 2005) has been developed. Developing an advanced hurricane initialization scheme for the ARW is imperative for the WRF hurricane research community. The WRF 3D-Var data assimilation system can incorporate many types of observations, including conventional data, bogus data, satellite data, and radar data (Xiao and Sun 2007). Its application to hurricane initialization is an interesting work. As the first step, the BDA technique is tested in this study.

From a computational perspective, BDA using 3D-Var is much faster than using 4D-Var. Because integrations of forward model and backward adjoint are involved in the iterative minimization procedure in 4D-Var, extensive experiments or real-time application of hurricane BDA with the 4D-Var framework have their difficulties. On the contrary, 3D-Var with BDA can be easily applied for large numbers of experiments. In addition, the WRF 3D-Var with BDA allows the hurricane structure to be generated by its balance transform and background error covariance structures. Because the background error covariance is produced using the ARW model forecasts, the hurricane initialization can still achieve balance to some extent with the ARW model.

Although BDA has been documented in the literature for several years, its application in WRF hurricane forecasts has not been seen. In recent years, we made several achievements in WRF 3D-Var to conduct BDA. Different from MM5 3D-Var (Barker et al. 2004), the preconditioned control variables in WRF 3D-Var are designed based on the characteristics of the WRF model (Skamarock et al. 2005). The control variables are streamfunction, unbalanced velocity potential, unbalanced temperature, pseudo–relative humidity, and unbalanced surface pressure. The National Meteorological Center (NMC) method (Parrish and Derber 1992) is used to construct correlations among the control variables for multivariate analysis. The BDA algorithm is improved relative to the MM5 3D-Var (Xiao et al. 2006). An asymmetric vortex perturbation is extracted from the previous 24-h forecast and relocated to the observed position. Summation of the symmetric vortex (Xiao et al. 2006) and the asymmetric perturbation produces a synthetic vortex structure. Assimilation of the synthetic sea level pressure (SLP) and wind profiles is performed through the WRF 3D-Var minimization procedure in the proposed BDA scheme.

In the BDA technique, the size of bogus vortex is a sensitive parameter to the analytical hurricane intensity and structure (Xiao et al. 2000). How to define the vortex size is an important issue for vortex relocation and bogusing (Kurihara et al. 1993). Numerical experiments indicate that the radius of 34-kt wind in the hurricane advisory can be used to calculate the bogus vortex size in BDA implementation (Xiao et al. 2006). In the experiments of this paper, we follow the same calculation to determine the vortex size.

In the next section, a brief description of the BDA scheme, the experimental design, and verification results of the WRF BDA scheme compared with the experiments without BDA in the forecasts of the hurricane track and intensity from 21 cases in the 2004 and 2005 hurricane seasons are provided. In addition, we carried out a case study (Hurricane Humberto, 2007) to compare the performance of WRF BDA with the Geophysical Fluid Dynamics Laboratory (GFDL) hurricane initialization, and the comparison results are presented in section 3. In section 4 we summarize our work and draw some conclusions.

2. Verifications in hurricane track and intensity

a. Cases

To verify the capability of BDA with WRF 3D-Var, we selected 21 cases from seven hurricanes in the 2004 and 2005 seasons to conduct parallel experiments (Table 1). They are Hurricanes Charley, Frances, Ivan, and Jeanne in 2004, and Katrina, Rita, and Wilma in 2005. These hurricanes’ activities are summarized in Fig. 1. We selected three cases for each hurricane before its landfall. The initial time for each case is provided in Table 1. Also listed in Table 1 are the hurricane categories, locations, and intensities [central sea level pressure (CSLP) and maximum surface wind (MSW)] at the initial time of each experiment. These cases are very famous because of their striking effects to Florida in 2004 and devastating impacts in the Gulf of Mexico in 2005.

b. BDA scheme and experimental design

The BDA scheme in WRF 3D-Var is similar to that in MM5 3D-Var (Xiao et al. 2006). However, because the WRF analysis variables are different from MM5 and this is the first implementation in ARW, we provide a brief description of the BDA scheme here. The preconditioning control variables in WRF 3D-Var are streamfunction, unbalanced velocity potential, unbalanced temperature, pseudo–relative humidity, and unbalanced surface pressure. The bogusing is constructed with the variables of SLP and wind profiles at seven levels (sea level and 1000, 925, 850, 700, 600, and 500 hPa) above hurricane vortex. There are two components (symmetric and asymmetric) in the bogusing. The asymmetric component is extracted from the 24-h model forecast field and relocated to the observed position. We simply take the difference between the forecast field and its azimuth average around the hurricane center, in the hurricane area, as the hurricane asymmetric component.

The initialization of the hurricane is done through minimization of a predefined cost function,
i1558-8432-48-3-680-e1
where Jb is the background term and Jo is the regular observation term; details of the first two terms in WRF 3D-Var can be found in Skamarock et al. (2005). To include BDA capability, two additional terms, Jp and JV, are added to the cost function. They are defined as
i1558-8432-48-3-680-e2
i1558-8432-48-3-680-e3
where P(r) and V(r, k) represent the SLP and wind fields (u and υ components) of the model atmosphere, Pbogus(r) and Vbogus(r, k) are the bogus SLP and wind fields, 𝗢P and 𝗢V are diagonal error variance matrices for the bogus SLPs and wind fields, RB is the radius of the bogus area, r is the radius from the hurricane center, and k denotes the vertical layers of the bogus wind profile.

The BDA and numerical forecasting experiments were conducted with a single domain. An example of the domain setup is shown in Fig. 1, which shows the experiments on Katrina (2005). The domain size and model configurations are the same for all experiments. The grid spacing is 12 km with 400 × 301 grid points. The domain center for each hurricane is different, but all three cases from the same hurricane have the same domain center (Table 1). There are 35 layers in the vertical direction, and the pressure at model top is 50 hPa. Physics options used in the ARW for all experiments include the Yonsei University YSU planetary boundary layer (PBL) scheme, which is the new generation of Medium-Range Forecast Model PBL scheme described by Hong and Pan (1996), the Kain–Fritsch cumulus scheme (Kain and Fritsch 1990, 1993), and the WRF single-moment three-class (WSM-3) microphysics scheme (Hong et al. 2004), which is a so-called simple-ice scheme wherein the cloud ice and cloud water are counted as the same category (Dudhia 1989). The ARW forecasts are executed for 72 h and are started from the time indicated in Table 1.

Two parallel sets of experiments are carried out. The first set of experiments (CT) for all cases uses Global Forecast System (GFS) analysis as background and assimilates only the conventional observations from the Global Telecommunications System (GTS) of the World Meteorological Organization. The second set of experiments (GB) is the same as CT but including BDA. Although the GTS data are assimilated in the National Centers for Environmental Prediction (NCEP)/GFS analysis, assimilating the data again with WRF 3D-Var enhances WRF initial conditions in comparison with simple interpolation of GFS analysis for WRF. The GFS analysis has much lower resolution than the WRF model we used in our experiments. For all WRF 3D-Var experiments, the same background error covariance is used, which is calculated from one-month statistics in September 2004 using the NMC method (Parrish and Derber 1992).

c. Results

Table 2 shows the number of cases with decreased or increased errors at 24-, 48-, and 72-h forecasts for the hurricane track and intensity (CSLP and MSW) in the 21 BDA experiments when compared with the 21 control experiments. Among the 21 cases at 24-h forecast, 12 cases have a reduction in track error and 9 cases increase track error. At 48- and 72-h forecasts, the numbers of cases with decreased/increased track error are 14/7 and 12/9, respectively. The errors of hurricane CSLP and MSW have similar characteristics. At the 24-h forecast, only one case produces an increased intensity error, but the number increases to six at 48-h and to eight at 72-h forecasts. With the model forecast time increasing, more BDA cases have increased intensity errors relative to CT, but the forecast skill with BDA is improved in general. The number of cases with decreased errors is much more than that with increased errors at all forecast times for all of the verification parameters.

Figure 2 shows the mean absolute errors of the forecasts (position, CSLP, and MSW) against the best-track observations at 24, 48, and 72 h. The errors from the BDA experiments are smaller than those from control experiments for all of the verification parameters (position, CSLP, and MSW). Over the ocean, there are sparse GTS observations, and assimilating only conventional GTS data is not enough to improve the forecast effectively, especially the hurricane intensity forecast. Bogus data are able to remedy the data-sparseness issue and improve the hurricane forecasting skill. We calculated the error reduction percentage by BDA from the statistics in Fig. 2. The largest reduction of average error is in the forecast of CSLP, with 26.4% reduction of average error by BDA. The improvement in hurricane MSW is also considerable; the average error is reduced by 24.0%. The track has the smallest improvement among the three verification parameters.

Because of a model spinup problem, both sets of experiments (CT and GB) present smaller CSLP and MSW errors at 48 and 72 h than at 24 h. However, the spinup problem in GB is much less than in CT. BDA alleviates the spinup problem and produces larger hurricane intensity improvement at 24 h than at 48 and 74 h. The benefit of BDA in hurricane intensity forecasts becomes less with the increase of the forecast time (Fig. 2). The initial intensity using BDA in GB experiments is much closer to the observed than it is in CT. With the model runs, the difference of intensity between GB and CT decreases, reflecting that the model forecast at longer times is less sensitive to initial conditions than at shorter times. The improvement of hurricane CSLP and MSW in GB relative to CT at 24 h is much more remarkable than that at 48 and 72 h. In the track forecasts, however, experiment GB has the biggest improvement at 72 h over the experiment CT.

Based on statistics, the improvement of hurricane intensity using BDA is larger than that of hurricane track. It is further verified that the large-scale environment influences hurricane track but that intensity is mainly affected by a hurricane’s internal, dynamical, and thermodynamical vortex structures. The BDA technique, which mainly improves the hurricane vortex structure according to the hurricane concept model, results in significant improvement of the hurricane intensity forecast. To support the assertion, we take Hurricane Katrina at 0000 UTC 26 August 2005 as an example to compare the vortex structures in CT and GB (Fig. 3). The CSLP and MSW errors are both reduced by BDA for Katrina initialized at 0000 UTC 26 August 2005. As shown in Fig. 3, the hurricane positions of CT and GB are both very close to the observation. However, GB produces a vortex with lower CSLP and larger MSW than those in CT. BDA enhances the cyclone circulation and makes the vortex much more compact. The area of the circular isobar with 1010 hPa in GB is much reduced when compared with that in CT. Note that the GFS analysis has its bogusing/relocation procedure, but it is apparently not sufficient for a good hurricane intensity forecast in the ARW. We verified that the initialization with BDA could improve the hurricane intensity forecast when compared with the forecast from a simple GFS analysis. In the next section, we will compare the scheme with the GFDL bogusing in a hurricane case study.

3. A comparison with the GFDL bogus scheme

The GFDL bogus scheme (Kurihara et al. 1993) has been applied in hurricane initialization for over a decade. It consists of three major steps: 1) interpolate the GFS analysis onto the GFDL model grid, 2) use a sophisticated filtering to remove the GFS vortex, and 3) add a GFDL-model-generated vortex to the GFDL analysis. As compared with the BDA scheme described in section 2, both GFDL and BDA hurricane initializations use GFS analysis as the first guess. However, the GFDL model does not have its own data assimilation, and the addition of the vortex to its final analysis is simply relocated, which usually leads to discontinuity problems around the vortex border. The vortex in the GFDL analysis is not balanced with the WRF. On the contrary, BDA uses a WRF 3D-Var technique to assimilate the vortex structure gradually back into its final analysis through a minimization procedure. There is no discontinuity problem around the vortex border. Because of the multivariate structure of the analysis, its analysis is balanced under the WRF 3D-Var background and balance constraint. Note that the transform between WRF 3D-Var control variables and model variables implies the geostrophic and hydrostatic balance constraints. These constraints are applied to increments only and not to the whole analysis. Although there are limitations in the constraints, WRF 3D-Var has a multivariate analysis for hurricane initialization.

First of all, two parallel experiments on Hurricane Humberto at 1200 UTC 12 September 2007 are compared to evaluate further the performance of the BDA scheme in WRF 3D-Var. The first experiment (CT) uses the WRF preprocessing system (WPS) to interpolate the GFDL analysis as initial conditions; the second experiment (GB) uses the aforementioned procedure to initialize the hurricane vortex. The GFDL analysis in the experiment CT includes the GFDL vortex bogusing at 9-km resolution. The experiment GB uses GFS analysis as the first guess in WRF 3D-Var and assimilates the bogusing vortex plus conventional data. The forecasts for both experiments (CT and GB) are executed on three domains, with the moving nested domains being 2 and 3. The grid spacings of the three domains are 12, 4, and 1.333 km, respectively. The physics options for domain 1 are the same as those of the experiments in section 2. We have nested domains for the experiments with Humberto (2007). The physics options in domains 2 and 3 include the YSU PBL scheme and WSM-5 (five classes) microphysics (Lim and Hong 2005) but no cumulus parameterization. In this study, we took advantage of an existing run (CT experiment) that uses WPS to interpolate the GFDL analysis as initial conditions. The domain configuration is different from that in section 2. BDA in the GT experiment is conducted in domain 1 only. The initial hurricane structures in domains 2 and 3 are interpolated from domain 1.

The initial SLP, surface temperature, and wind fields for CT and GB at 1200 UTC 12 September 2007 are shown in Figs. 4a and 4b, and the geopotential height, wind, and absolute vorticity fields at 500 hPa are shown in Figs. 4c and 4d. Experiment GB has a greater storm intensity when compared with experiment CT. The initial CSLP is 1011 hPa in CT, whereas it is 1007 hPa in GB and is closer to that observed (1006 hPa). The maximum surface wind is also stronger in GB (17.84 m s−1) than in CT (12.46 m s−1). The temperature in GB is much warmer around the storm center than in the environment, whereas it is flat in the experiment CT. Referring to the tropical storm stage of Humberto at 1200 UTC 12 September 2007, the SLP and surface wind initialized in GB indicate a compact and enhanced structure (Fig. 4b) relative to that in CT (Fig. 4a).

There are the same characteristics at 500 hPa (Figs. 4c,d), where the storm intensity is stronger in GB than CT. The geopotential height in the storm center is decreased and the cyclonic circulation becomes more intensified in GB (Fig. 4d) than in CT (Fig. 4c). The absolute vorticity field shows that CT has a ring of maximum vorticity around the storm center, whereas GB produces maximum vorticity in the vortex center. The cyclonic circulation in GB is more enhanced in rotation. From the above analysis, GB initializes the storm structure with details that mimic the real vortex structure. The experiment CT had a much weaker storm and an ill-defined vortex structure for Humberto at 1200 UTC September 2007. The vortex is not well organized in experiment CT, which affects its subsequent forecasts of both track and intensity.

Figure 5 shows the forecast tracks from experiments CT and GB as well as the best track for the period of 1200 UTC 12–1200 UTC 15 September 2007. The experiment CT, which has the vortex not well organized at the initial time, fails to predict the storm’s inland movement. It lingers along the coastal area of Texas and Louisiana for 2 days and then turns back to the Gulf of Mexico. When compared with the best-track observation, CT fails to predict the storm’s track. On the contrary, the experiment GB successfully predicts the storm’s landfall and inland movement. Its predicted track follows the best-track observation. Note that in Fig. 5 the best track from the National Hurricane Center extends to 2100 UTC 14 September, whereas the prediction extends to 1200 UTC 15 September.

For the intensity prediction, CT also fails to predict the storm’s intensification before landfall. Because it fails to predict the storm’s intensification before landfall and fails to predict the storm’s inland movement, its overall intensity forecast is not successful at all and therefore is omitted in the comparison. In Fig. 6, we thus only analyze the storm’s intensity in GB and compare it with observations. The trends of CSLP (Fig. 6a) and MSW (Fig. 6b) from 1200 UTC 12 to 1200 UTC 15 September show good agreement between the forecast and observation; GB successfully predicts Humberto’s intensification from a tropical storm (1200 UTC 12 September) to a category-1 hurricane (0600 UTC 13 September) before its landfall over the Texas coast. The observation indicates the maximum intensity of Humberto at 0915 UTC 13 September with a CSLP of 986 hPa and an MSW of 85 kt (44 m s−1); GB predicts the maximum intensity at 0945 UTC 13 September with a CSLP of 989 hPa and an MSW of 82 kt (43 m s−1). However, it overpredicts Humberto’s strength inland. At 0000 UTC 14 September, for example, GB predicts a CSLP of 997 hPa and an MSW of 37 kt (19 m s−1), whereas the observations are 1006 hPa and 25 kt (13 m s−1).

The intensity overprediction of the hurricane over land is related to surface friction setup in the ARW in our experiment. To examine this, we conducted a third experiment in which the surface roughness length over land is doubled in the WRF land use table to increase the surface friction while all other experimental setup is the same as in GB. Figure 7 shows the 36-h forecast of Hurricane Humberto at 0000 UTC 14 September, about 15 h after landfall. Whereas GB predicts a CSLP of 997 hPa with an MSW of 19 m s−1 (Fig. 7a), the new experiment with increased surface friction reduces the hurricane intensity inland and produces a CSLP of 999 hPa and an MSW of 12 m s−1 (Fig. 7b). The current WRF land use table seems to underestimate the surface friction for the prediction of Hurricane Humberto (2007).

In this case study, GB with WRF 3D-Var performs better than CT with the GFDL analysis in the tropical storm initialization and subsequent forecast from the ARW. To have a comprehensive evaluation of the BDA scheme for WRF, however, many case studies and much real-time testing are necessary. We will conduct such testing in the future. It is anticipated that BDA will have the potential to improve hurricane forecasts with the ARW model.

4. Summary and conclusions

The BDA technique is introduced to ARW hurricane forecasting. By assimilating vortex bogusing observations, WRF 3D-Var incorporates a vortex structure in its hurricane initialization. We tested 21 cases from seven hurricanes in the 2004 and 2005 seasons to assess the BDA scheme in WRF 3D-Var for forecasts of the hurricanes. We also conducted one case study for Hurricane Humberto (2007) and compared the results between BDA and GFDL analyses for the ARW runs. The major results of this study are summarized as follows:

  • ARW hurricane forecasting using the BDA technique shows an improved forecast skill in hurricane track and intensity when compared with initialization from just GFS analysis. Using the WRF 3D-Var system, the bogus SLP and wind profile data can be efficiently assimilated to recover the initial hurricane structure under 3D-Var statistical and physical balances. The forecasts of hurricane track and intensity are therefore improved.
  • The enhancement of the hurricane forecast skill using the BDA technique is reflected in all forecast periods. With BDA, the largest improvement is in hurricane central pressure. A considerable improvement in hurricane maximum surface wind is also produced. The track has the smallest improvement among the three verification parameters.
  • The improvement in hurricane intensity forecasts decreases with increases of forecast length. The improvement of hurricane CSLP and MSW at 24 h is greater than that at 48 and 72 h. In the track forecasts, however, the BDA results show the most improvement at 72 h.
  • A case study with Hurricane Humberto (2007) shows that using BDA with WRF 3D-Var for hurricane initialization outperforms an interpolation from the GFDL analysis in the hurricane forecast. More cases studies and real-time forecasts are necessary to verify its performance further. The preliminary result in this study indicates, however, that BDA with WRF 3D-Var for hurricane initialization has the potential to improve ARW hurricane forecasts.
  • From the simulation results of Hurricane Humberto (2007), the WRF land use table seems to underestimate the surface friction over land. More experiments should be conducted to address this issue. This work is beyond the scope of this paper.

One of the most challenging problems for the hurricane forecaster and researcher is to define the vortex structure in light of insufficient observations over the ocean. The BDA has shown promise in MM5 hurricane forecasts, and, in this study, we have shown its potential for ARW hurricane forecasts. We will conduct more studies on this scheme in the future.

Acknowledgments

The authors thank James Done (NCAR) for processing the GFDL analysis for ARW and conducting forecasts with the analysis for Hurricane Humberto (2007). Comments on the early manuscript by Jordan Powers, Zhiquan Liu, and Yongrun Guo (NCAR) are greatly appreciated. Bobbie Weaver helped to edit the manuscript. This study has been supported by NOAA Grant 05111076 and the National Natural Science Foundation of China (Grant 40828005).

REFERENCES

  • Barker, D. M., , W. Huang, , Y-R. Guo, , A. Bourgeois, , and Q. Xiao, 2004: A three-dimensional variational data assimilation system for MM5: Implementation and initial results. Mon. Wea. Rev., 132 , 897914.

    • Search Google Scholar
    • Export Citation
  • Dudhia, J., 1989: Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46 , 30773107.

    • Search Google Scholar
    • Export Citation
  • Hong, S-Y., , and H-L. Pan, 1996: Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev., 124 , 23222339.

    • Search Google Scholar
    • Export Citation
  • Hong, S-Y., , J. Dudhia, , and S-H. Chen, 2004: A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation. Mon. Wea. Rev., 132 , 103120.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., , and J. M. Fritsch, 1990: A one-dimensional entraining/detraining plume model and its application in convective parameterization. J. Atmos. Sci., 47 , 27842802.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., , and J. M. Fritsch, 1993: Convective parameterization for mesoscale model: The Kain–Fritsch scheme. The Presentation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 165–170.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., , M. A. Bender, , and R. J. Ross, 1993: An initialization scheme of hurricane models by vortex specification. Mon. Wea. Rev., 121 , 20302045.

    • Search Google Scholar
    • Export Citation
  • Lim, J-O. J., , and S-Y. Hong, 2005: Effects of bulk ice microphysics on the simulated monsoonal precipitation over east Asia. J. Geophys. Res., 110 , D24201. doi:10.1029/2005JD006166.

    • Search Google Scholar
    • Export Citation
  • Park, K., , and X. Zou, 2004: Toward developing an objective 4DVAR BDA scheme for hurricane initialization based on TPC observed parameters. Mon. Wea. Rev., 132 , 20542069.

    • Search Google Scholar
    • Export Citation
  • Parrish, D. F., , and J. C. Derber, 1992: The National Meteorological Center’s spectral statistical-interpolation analysis system. Mon. Wea. Rev., 120 , 17471763.

    • Search Google Scholar
    • Export Citation
  • Pu, Z-X., , and S. A. Braun, 2001: Evaluation of bogus vortex techniques with four-dimensional variational data assimilation. Mon. Wea. Rev., 129 , 20232039.

    • Search Google Scholar
    • Export Citation
  • Skamarock, W. C., , J. B. Klemp, , J. Dudhia, , D. O. Gill, , D. M. Barker, , W. Wang, , and J. G. Powers, 2005: A description of the Advanced Research WRF version 2. NCAR Tech. Note TN-468+STR, 88 pp.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., , and J. Sun, 2007: Multiple-radar data assimilation and short-range quantitative precipitation forecasting of a squall line observed during IHOP_2002. Mon. Wea. Rev., 135 , 33813404.

    • Search Google Scholar
    • Export Citation
  • 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 , 22522269.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., , Y-H. Kuo, , Y. Zhang, , D. M. Barker, , and D-J. Won, 2006: A tropical cyclone bogus data assimilation scheme in the MM5 3D-Var system and numerical experiments with Typhoon Rusa (2002) near landfall. J. Meteor. Soc. Japan, 84 , 671689.

    • Search Google Scholar
    • Export Citation
  • Zhang, X., , Q. Xiao, , and P. J. Fitzpatrick, 2007: The impact of multisatellite data on the initialization and simulation of Hurricane Lili’s (2002) rapid weakening phase. Mon. Wea. Rev., 135 , 526548.

    • Search Google Scholar
    • Export Citation
  • Zou, X., , and Q. Xiao, 2000: Studies on the initialization and simulation of a mature hurricane using a variational bogus data assimilation scheme. J. Atmos. Sci., 57 , 836860.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

The best tracks of seven hurricanes (Charley, Frances, Ivan, Jeanne, Katrina, Rita, and Wilma) in the 2004 and 2005 seasons. The hurricane positions are shown with filled circles at 0000 UTC, open circles at 1200 UTC, and plus signs at 0600 and 1800 UTC. The shading in the map is terrain height, with the scale on the right.

Citation: Journal of Applied Meteorology and Climatology 48, 3; 10.1175/2008JAMC1994.1

Fig. 2.
Fig. 2.

The mean absolute errors of the (top) hurricane track, (middle) maximum surface wind, and (bottom) central sea level pressure for the forecasts from statistics of 21 cases listed in Table 1. The light and dark bars are the statistical results for the control experiment and BDA experiment, respectively. The errors of 24-, 48-, and 72-h forecasts with their averages are shown.

Citation: Journal of Applied Meteorology and Climatology 48, 3; 10.1175/2008JAMC1994.1

Fig. 3.
Fig. 3.

Hurricane Katrina at 0000 UTC 26 Aug 2005 for the (a) CT and (b) GB experiments. The solid lines are SLP, with a contour interval of 2.5 hPa, and the barbs show the 10-m wind field (a full barb represents 5 m s−1).

Citation: Journal of Applied Meteorology and Climatology 48, 3; 10.1175/2008JAMC1994.1

Fig. 4.
Fig. 4.

Analyses of Hurricane Humberto at 1200 UTC 12 Sep 2007 showing (top) SLP (thick isolines), surface temperature (thin isolines), surface wind barbs, and surface wind speed (shading) for the (a) CT and (b) GB experiments and (bottom) 500-hPa geopotential height (isolines), wind barbs, and absolute vorticity (shading) for the (c) CT and (d) GB experiments.

Citation: Journal of Applied Meteorology and Climatology 48, 3; 10.1175/2008JAMC1994.1

Fig. 5.
Fig. 5.

The 72-h track forecast for Hurricane Humberto from 1200 UTC 12 to 1200 UTC 15 Sep 2007. The dashed line with the open circles is the best track, the solid line with asterisks is the forecast from the CT experiment, and the gray line with the open triangles is the forecast from the GB experiment. The date/hour are shown in boxes.

Citation: Journal of Applied Meteorology and Climatology 48, 3; 10.1175/2008JAMC1994.1

Fig. 6.
Fig. 6.

The 72-h intensity forecast for Hurricane Humberto from 1200 UTC 12 to 1200 UTC 15 Sep 2007: (a) CSLP and (b) MSW. The solid line is the forecast from the GB experiment, and the dashed line is from the best-track observation.

Citation: Journal of Applied Meteorology and Climatology 48, 3; 10.1175/2008JAMC1994.1

Fig. 7.
Fig. 7.

The 36-h forecasts of Hurricane Humberto at 0000 UTC 14 Sep 2007 by (a) GB and (b) GB with increased surface friction in the simulation. The thin solid isolines are for SLP (2-hPa interval). The surface winds are shown with wind barbs (a full barb represents 5 m s−1) and thick gray isolines (2 m s−1 interval).

Citation: Journal of Applied Meteorology and Climatology 48, 3; 10.1175/2008JAMC1994.1

Table 1.

Categories, locations, and intensities (minimum central SLP and maximum surface wind) of the selected 21 hurricane cases in the 2004 and 2005 seasons. (The initial time for each case and its experimental domain center are provided; category TS indicates tropical storm)

Table 1.
Table 2.

Numbers of cases with decreased (dcsd) or increased (icsd) errors at 24-, 48-, and 72-h forecasts for the hurricane track and intensity (CSLP and maximum surface wind) in the 21 BDA experiments when compared with control experiments.

Table 2.

* The National Center for Atmospheric Research is sponsored by the National Science Foundation.

Save