• Aberson, S. D., 2002: Two years of hurricane synoptic surveillance. Wea. Forecasting, 17 , 11011110.

  • Aberson, S. D., 2003: Targeted observations to improve operational tropical cyclone track forecast guidance. Mon. Wea. Rev., 131 , 16131628.

    • Search Google Scholar
    • Export Citation
  • Aberson, S. D., , and M. DeMaria, 1994: Verification of a nested barotropic hurricane track forecast model (VICBAR). Mon. Wea. Rev., 122 , 28042815.

    • Search Google Scholar
    • Export Citation
  • Aberson, S. D., , and J. L. Franklin, 1999: Impact on hurricane track and intensity forecasts of GPS dropwindsonde observations from the first-season flights of the NOAA Gulfstream-IV jet aircraft. Bull. Amer. Meteor. Soc., 80 , 421427.

    • Search Google Scholar
    • Export Citation
  • Aberson, S. D., , and B. J. Etherton, 2006: Targeting and data assimilation studies during Hurricane Humberto (2001). J. Atmos. Sci., 63 , 175186.

    • Search Google Scholar
    • Export Citation
  • Barrett, B. S., , L. M. Leslie, , and B. H. Fiedler, 2006: An example of the value of strong climatological signals in tropical cyclone track forecasting: Hurricane Ivan (2004). Mon. Wea. Rev., 134 , 15681577.

    • Search Google Scholar
    • Export Citation
  • Burpee, R. W., , J. L. Franklin, , S. J. Lord, , R. E. Tuleya, , and S. D. Aberson, 1996: The impact of Omega dropwindsondes on operational hurricane track forecast models. Bull. Amer. Meteor. Soc., 77 , 925933.

    • Search Google Scholar
    • Export Citation
  • Hodyss, D., , and S. J. Majumdar, 2007: The contamination of ‘data impact’ in global models by rapidly growing mesoscale instabilities. Quart. J. Roy. Meteor. Soc., 133 , 18651875.

    • Search Google Scholar
    • Export Citation
  • Liu, Q., , T. Marchok, , H-L. Pan, , M. Bender, , and S. Lord, 2000: Improvements in hurricane initialization and forecasting at NCEP with global and regional (GFDL) models. NCEP Tech. Proc. Bull. 472, 7 pp.

  • Majumdar, S. J., , S. D. Aberson, , C. H. Bishop, , R. Buizza, , M. S. Peng, , and C. A. Reynolds, 2006: A comparison of adaptive observing guidance for Atlantic tropical cyclones. Mon. Wea. Rev., 134 , 23542372.

    • 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
  • Reynolds, C. A., , M. S. Peng, , S. J. Majumdar, , S. D. Aberson, , C. H. Bishop, , and R. Buizza, 2007: Interpretation of adaptive observing guidance for Atlantic tropical cyclones. Mon. Wea. Rev., 135 , 40064029.

    • Search Google Scholar
    • Export Citation
  • Woollen, J. R., 1991: New NMC operational OI quality control. Preprints, Ninth Conf. on Numerical Weather Prediction, Denver, CO, Amer. Meteor. Soc., 24–27.

  • Zhang, R., , C. Snyder, , and R. Rotunno, 2003: Effects of moist convection on mesoscale predictability. J. Atmos. Sci., 60 , 11731185.

  • View in gallery

    GFS track forecasts for Hurricane Frances initialized at 0000 UTC 2 Sep 2004. Positions are plotted every 6 h through 5 days and are marked by a number every 12 h. BEST is the best track.

  • View in gallery

    Dropwindsonde observations from the 1 Sep 2004 surveillance mission at (a) 925, (b) 500, (c) 400, and (d) 300 hPa, in standard format except that the moisture data are indicated by relative humidity (%).

  • View in gallery

    Skew T–logp diagram of the dropwindsonde observation taken near Grand Bahama Island at 2331 UTC 1 Sep 2004. Values near 275 hPa are from the aircraft flight values reported along with the dropwindsonde data.

  • View in gallery

    The 400-hPa geopotential heights for the (a)–(d) GFSO and (e)–(h) GFSN at 0, 12, 36, and 60 h into the forecast initialized at 0000 UTC 9 Sep 2005.

  • View in gallery

    (Continued)

  • View in gallery

    Trajectory of a dropwindsonde released in the eyewall of Hurricane Emily 16 Jul 2005. The “X” marks the location provided on the TEMPDROP message and represents the only location available for assimilation of the dropwindsonde data.

  • View in gallery

    GFS track forecasts for Hurricane Ophelia initialized at 0000 UTC 9 Sep 2005. Positions are plotted every 6 h and marked by a number every 12 h. BEST is the best track.

  • View in gallery

    (top) Initial condition and (bottom) 12-h forecast differences in the 850–200-hPa mean wind between the GFSN and GFSO for the 0000 UTC 9 Sep 2005 surveillance mission.

  • View in gallery

    The 0-, 12-, 36-, 60-, and 72-h forecasts of 400-hPa vorticity from the (a)–(e) GFSO and (f)–(i) GFSN initialized at 0000 UTC 9 Sep 2005.

  • View in gallery

    (Continued)

  • View in gallery

    Targeting guidance from (top) Navy total energy SVs, (middle) ETKF from the NCEP global ensemble, and (bottom) NCEP global ensemble DLM wind variance. The box and circle define the verification regions for the respective technique.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 55 55 7
PDF Downloads 19 19 2

Large Forecast Degradations due to Synoptic Surveillance during the 2004 and 2005 Hurricane Seasons

View More View Less
  • 1 NOAA/AOML/Hurricane Research Division, Miami, Florida
© Get Permissions
Full access

Abstract

Though operational tropical cyclone synoptic surveillance generally leads to smaller track forecast errors in the National Oceanic and Atmospheric Administration Global Forecasting System (GFS) than would occur otherwise, not every case is improved. Very large GFS forecast degradations due to surveillance are investigated. Small perturbations to model initial conditions may have a large impact locally or downstream in a short time. In these cases, the perturbations are due either to erroneous data assimilated into the models or to issues with the complex data assimilation system itself, and may have caused the forecast degradations. Investigation of forecast and observing system failures can lead to procedural changes that may eliminate some causes of future large forecast errors.

Corresponding author address: Sim D. Aberson, NOAA/AOML/Hurricane Research Division, Miami, FL 33149. Email: sim.aberson@noaa.gov

Abstract

Though operational tropical cyclone synoptic surveillance generally leads to smaller track forecast errors in the National Oceanic and Atmospheric Administration Global Forecasting System (GFS) than would occur otherwise, not every case is improved. Very large GFS forecast degradations due to surveillance are investigated. Small perturbations to model initial conditions may have a large impact locally or downstream in a short time. In these cases, the perturbations are due either to erroneous data assimilated into the models or to issues with the complex data assimilation system itself, and may have caused the forecast degradations. Investigation of forecast and observing system failures can lead to procedural changes that may eliminate some causes of future large forecast errors.

Corresponding author address: Sim D. Aberson, NOAA/AOML/Hurricane Research Division, Miami, FL 33149. Email: sim.aberson@noaa.gov

1. Introduction

Since 1997, the National Oceanic and Atmospheric Administration (NOAA) has conducted synoptic surveillance missions to improve tropical cyclone (TC) track forecasts (Aberson and Franklin 1999; Aberson 2002, 2003; Aberson and Etherton 2006). During the first 2 yr of these missions, dropwindsonde data from the NOAA Gulfstream-IV (G-IV) aircraft, sometimes supplemented by observations from NOAA P-3 or Air Force C-130 aircraft, improved the National Centers for Environmental Prediction (NCEP) Global Forecasting System (GFS—then called the Aviation Model) forecasts by a smaller-than-expected average of <10%; this was attributed to suboptimal sampling and data assimilation procedures (Aberson 2002). Average track forecast improvements increased to 18% by utilizing targeting and optimized sampling strategies in which areas of likely perturbation growth [i.e., large NCEP global ensemble forecast 850–200-hPa deep-layer-mean (DLM) wind velocity variance at the observing time] are identified and regularly spaced observations are obtained within and completely surrounding them. This targeted data subset led to larger forecast improvements than all the available data; this was found to be because of the imperfect data assimilation system spreading good information from well-sampled areas into neighboring data-sparse regions in which perturbation growth is expected to be large (i.e., other suboptimally sampled or unsampled target regions). Missions have since been designed utilizing these strategies, and average improvements have been 15%–20% throughout the 5-day forecast period. Sophisticated objective targeting strategies such as the ensemble transform Kalman filter (ETKF) and singular vectors (SVs) from various modeling systems are also being studied for TC surveillance (Majumdar et al. 2006; Reynolds et al. 2007).

During the very active 2004 and 2005 Atlantic hurricane seasons, 31 and 44 surveillance missions were conducted, respectively. Despite the substantial average track forecast error reduction because of these missions, not every individual forecast was improved. A few cases with large forecast degradations due to the assimilation of the dropwindsonde data were outliers in this large sample. These cases are investigated to learn the causes of these failures, with the ultimate goal of correcting any problems for future missions. Three forecast sets are examined: 1) a Hurricane Frances (2004) forecast in which the dropwindsonde data caused a forecast recurvature that did not occur either in the forecast without the data or in reality, 2) Hurricane Ivan (2004) forecasts in which the GFS had a large northward bias, and 3) two Hurricane Ophelia (2005) and two Hurricane Wilma (2005) forecasts in which, again, the model had a large, incorrect northward (recurving) bias. These are discussed in sections 3, 4, and 5, respectively, with a short description of the model runs in section 2, and conclusions and recommendations are presented in section 6.

2. Model description and parallel cycle

Model results are from the operational GFS or retrospective parallel cycles, as in Aberson and Etherton (2006). The cycles are identical except for the dropwindsonde data assimilated and the computer on which they were run. The GFS was run in real time on the NCEP central computer; parallel runs were completed on the NOAA developmental computer. Input data and model code were identical between the various cycles. The GFS version for each run was that current at the initial time of each run; because of frequent model updates, the versions in each run are not identical.

The GFS is composed of a quality control algorithm, a TC relocation procedure, an analysis procedure, and a global spectral model. The quality control involves optimal interpolation and hierarchical decision making to evaluate observations before input to the analysis (Woollen 1991). The TC relocation procedure (Liu et al. 2000) finds the vortex center in the background field (the previous 6-h forecast), separates it from the environmental field, and moves it to the specified location; if no vortex is seen in the first guess, a synthetic vortex is added. The analysis scheme is the spectral statistical interpolation (Parrish and Derber 1992): the background field is combined with observations with a three-dimensional variational multivariate formalism. During 2004, the model horizontal resolution was spectral triangular 254 (T254) with 64 sigma levels (L64) in the vertical through 84 h, and T170L42 for the remainder of each forecast; during 2005, the model resolution was T382L64 throughout the entire forecast.

3. Hurricane Frances (2004)

Because of the potential for Hurricane Frances to impact the southeastern United States, eight surveillance missions were conducted from 29 August to just before landfall on the Florida east coast early on 4 September. The first two missions (29 and 30 August) reduced GFS forecast track errors an average of 39%–57% through 48 h. The subsequent mission on 1 September resulted in a track forecast error increase of up to 600% during most of the 5-day forecast period (Table 1; Fig. 1).

a. Data

Figure 2 shows the mid- and lower-tropospheric mandatory-level dropwindsonde observations from the surveillance mission on 1 September. Frances is located just north of Hispaniola inside the ring of observations. The G-IV departed Barbados and released dropwindsondes from its flight level above 200 hPa in all environmental quadrants of Frances before recovering at MacDill Air Force Base. A C-130 departed Keesler Air Force Base and sampled the northeastern Gulf of Mexico and the region just east of northern Florida from a flight level of about 300 hPa before returning to Keesler. (A NOAA P-3 and another C-130 observed the core of Frances for research and reconnaissance, respectively; dropwindsonde observations from these missions are not shown.) Aberson (2002) showed that the forecast improvement was related to the amount of areal coverage of dropwindsonde data in the TC environment. Since two planes gathered data in the environment on 1 September, the expectation was for a substantial forecast improvement.

A skew T–logp diagram of the observation near Grand Bahama Island (Fig. 3) suggests that the dropwindsonde was released with the sensor cap intact causing a slow response to the atmospheric temperature and humidity. This observation is warmer than surrounding ones at 300 hPa and cooler below (about 1.5°C cooler at 400 hPa and 5°C cooler at other levels); this reflects a slow response from the warm, ambient temperature aboard the aircraft to the cool upper troposphere and then to the warming as the probe descended. The humidity measurements are dryer than those from surrounding regions, reflecting a slow response from the dry air aboard the aircraft to the moist, tropical air outside.1 The geopotential height is about 150 m lower than surrounding observations at 400 hPa, and this difference decreases to about 20 m at 925 hPa. The wind observations fit well with surrounding ones since the sensor cap does not affect the advection of the probe by the wind. Though most of the sounding was flagged by the NCEP quality control algorithms (Woollen 1991), because the temperature observations at the sounding top and the height observations at the bottom were commensurate with those nearby, these data were not flagged as erroneous, and they were assimilated into the model. Increments of up to −8 m in the geopotential height field below 800 hPa and −0.5°C around 350 hPa at the sounding location resulted. Because of the relationship between the dynamic and thermodynamic fields, a region of relatively low temperature, geopotential height, and enhanced vorticity through the deep layer over the northern Bahamas in the model initial condition is seen (Fig. 4). This could advect Frances northward at a faster rate than otherwise would have occurred and lead to an erroneous forecast. The GFS track forecast was already north of that without the dropwindsonde observations by 12 h into the forecast (Fig. 1). A substantial forecast degradation had already occurred by this time (Table 1), though short-term forecast errors were small.

b. Model fields

Figure 4 shows the 400-hPa initial and forecast geopotential height fields in both the operational (GFSO) and no dropwindsonde (GFSN) runs. In the initial condition, an inverted trough corresponding to the area of enhanced vorticity and low heights described above extends from the northern Bahamas across southern Florida and western Cuba in the GFSO but is absent in the GFSN. Majumdar et al. (2006) found that, for major hurricanes, the various targeting techniques nearly unanimously found that small perturbations in the TC near environment would grow fastest and most impact model forecasts. During the first 24 h of the forecast, the differences between the GFSO and the GFSN grow in this region. The feature moves northeastward in the GFSO, eroding the subtropical ridge to the north of Frances allowing the TC to move northwestward. In the GFSN, the subtropical ridge remains strong, and Frances moves to the west-northwest toward Florida. This analysis does not rule out the possibility that other factors, such as changes to quality control flagging of satellite data near the dropwindsonde observation on later iterations of the quality control, could have impacted the forecast. However, the assimilation of the erroneous observation into the GFS seems to have been a major factor in the poor forecast track in this case.2

4. Hurricane Ivan (2004)

Sixteen surveillance missions were conducted around Hurricane Ivan during a 10-day period in 2004. Table 2 shows that the missions caused average track forecast degradations of 12%–42%, most of which are statistically significant at the 90% level.3 The cause of these average forecast degradations was found to be the assimilation of eye, eyewall, and rainband (hereinafter “TC core”) dropwindsonde data from both NOAA and Air Force aircraft. Global models such as the GFS cannot resolve TC core structures because of their coarse resolutions. Also, depending on the TC intensity and the eye size, dropwindsondes released in the eyewall may orbit more than halfway around the center; the “TEMPDROP” code used to transmit the data from the aircraft provides only one location with 0.1° latitude–longitude (about 10 km) resolution. This may result in the data assimilation attempting to utilize data more than 180° azimuthally from its correct location relative to the TC center (Fig. 5). Assimilation of accurate dropwindsonde data in the TC core can therefore lead to unrepresentative structures in the model initial conditions. Aberson (2003) and Majumdar et al. (2006) showed that the TC itself is nearly always a region of large perturbation growth, so these unrepresentative structures may grow rapidly, potentially leading to large forecast errors. Tests (“GFIC”) in which dropwindsonde data within 111.1 km of the specified TC center location were removed from each data assimilation cycle were conducted. GFIC nearly eliminated the average forecast degradations (Table 2). To show that the change does not negatively impact those cases in which the surveillance missions led to forecast track improvements, a test during Hurricane Charley, in which the dropwindsondes caused improvements of 14%–36% through 3 days (Table 3), were also conducted. These cases were further improved at almost all forecast times. Beginning early in the 2005 hurricane season, dropwindsonde data within 111.1 km (or 3 times the specified radius of maximum wind, whichever is larger) of the TC center location were removed from the data assimilation.4

5. Hurricanes Ophelia (2005) and Wilma (2005)

Because of possible U.S. coastal impacts, a series of surveillance missions were conducted around Hurricanes Ophelia and Wilma in 2005. The first Ophelia mission was flown on 8 September, when the storm was centered about 150 km east of Cape Canaveral, with missions following every 24 h for a week. The first Wilma mission was flown on 18 October, when the hurricane was centered about 200 km northeast of Cabo Gracias a Diós on the Nicaragua–Honduras border; a 36-h period with no surveillance followed, and missions resumed every 12 h thereafter until landfall on the morning of 24 October.

The data from the first Ophelia mission caused a substantial forecast degradation through the entire 5-day forecast (Table 4; Fig. 6). Figure 7 shows the dropwindsonde data impact in the GFS DLM wind at the initial time and 12 h later. Though the differences are due to dropwindsonde data between Florida and North Carolina, one maximum impact near 35°N, 60°W was associated with Tropical Storm Nate, and another at the northeastern edge of the figure was associated with Tropical Storm Maria; both features moved rapidly northeastward and did not affect the Ophelia model forecast track. The small increment near 40°N, 105°W in the initial condition is of particular note; this increment grew during the first 12 h of the forecast as it moved northward. This feature can be tracked as a maximum in midlevel vorticity in the GFSO to the western tip of Lake Superior 30 h into the forecast and subsequently southeastward (Fig. 8). A similar feature can be tracked in the GFSN until it dissipates over the Ohio Valley by 48 h into the forecast. The relatively strong GFSO vorticity maximum eroded the subtropical ridge causing Ophelia to move northward and make landfall in North Carolina; in the GFSN, the vorticity maximum dissipated before affecting the subtropical ridge, keeping Ophelia meandering offshore and providing a forecast superior to that from the GFSO. The data from the subsequent Ophelia mission and the second and third Wilma missions (Table 5) similarly degraded the GFS forecasts because of a feature originating in the same region (not shown). However, in the Wilma cases, the assimilation of the dropwindsonde data weakened a vorticity maximum in the midlatitude flow allowing the subtropical ridge to remain strong.

The origin of this initial increment is unknown. One possibility is that the GFSN and GFSO were run on two different computers, causing different numerical representations of model variables (Aberson and Etherton 2006). However, tests of the GFS using different numbers of processors on the same computer show that the differences are likely to be small into the medium range (about 3 days), suggesting that this is an important factor only in a very few cases and at long ranges. A second possibility is that the data assimilation and model are run in spectral space (Parrish and Derber 1992) so that local observations produce global increments, some in regions of rapid perturbation growth. A third possible cause is that the quality control (Woollen 1991) is conducted with multiple iterations; the increments may reflect different quality control decisions in that particular region in any particular iteration, especially if the data are close to the criterion. Hodyss and Majumdar (2007) also showed that data impacts quickly become contaminated by initially small instabilities in locations dynamically unrelated to those in which the observations are taken.

Majumdar et al. (2006) and Reynolds et al. (2007) compared five targeting techniques in the Atlantic basin during the 2004 season: the NCEP global ensemble DLM wind variance, the ETKF with either the NCEP global ensemble or the European Centre for Medium-Range Weather Forecasts (ECMWF) global ensemble, and total energy SVs from both the Navy and ECMWF. Three of these were available for this case (Fig. 9),5 and all show an area of possible rapid perturbation growth near the initial increment described above, though the maximum is smallest in the DLM wind variance and largest in the SV guidance. The second Ophelia degradation was caused by the same feature that remained in the model. The Wilma case (not shown) was caused by an increment in nearly the same region, and the targeting techniques similarly showed that this region was one of likely model perturbation growth. NCEP replaced the spectral statistical interpolation data assimilation system with a gridpoint system; this eliminated the global impact of data in the initial conditions. However, because of the spectral model, small local differences must have some, even very small, impact globally. Though these impacts are likely to be small in the 6-h data assimilation cycle, Zhang et al. (2003) showed that the smallest differences grow most rapidly in the models and differences of all sizes saturate at time scales on the order of a day. So, though the growth may be small for just one set of dropwindsonde data, they grow rapidly in time in the cycle. Thus, in any spectral model system, such problems will continue. Regardless of the cause of this feature, these cases show that small perturbations to model initial conditions can have a large impact downstream in a short time.

6. Conclusions and recommendations

Large GFS forecast degradations from synoptic surveillance during the 2004 and 2005 hurricane seasons are investigated. The reasons for the degradations are found to be erroneous data assimilated into the model, problems with the quality control in the assimilation process, or issues with the complex data assimilation itself. Because of these studies, modifications to the NCEP data assimilation system and to data acquisition systems aboard aircraft have been made. A planned upgrade to the dropwindsonde processing system will allow for viewing synoptic data in the hopes of flagging erroneous data before they are transmitted from the aircraft. Changes to the assimilation of dropwindsonde data in the TC core have been implemented. Further unrelated upgrades to the data assimilation system may also help to alleviate the problems discussed.

The TC surveillance generally leads to 15%–20% smaller GFS track forecast errors relative to those obtained from a parallel cycle without the additional data. The causes of large individual degradations during 2004 and 2005 have been found, and some have been eliminated. Burpee et al. (1996) and Aberson (2002, 2003) showed that, despite substantial average model forecast improvements due to the assimilation of dropwindsonde observations, not all individual forecasts are improved. Because of uncertainties in the data, the model, and the data assimilation, some surveillance missions will necessarily lead to forecast degradations. The goal of this study was to find particular problems leading to exceptionally large forecast degradations. Even with improvements already implemented or recommended, some degradations will continue to occur. Since the goal is to remove the correctable errors from the system, studies of very large degradations should continue so as to eliminate problems if they arise, to improve overall statistics, and to help to eliminate the largest forecast errors.

Acknowledgments

The author thanks NCEP/EMC for their help in running the GFS and in providing the computer resources to make the study possible and Steve Lord and James Franklin for helpful discussions. Nelsie Ramos provided much of the graphics in the Frances and Ophelia cases and engaged in many fruitful discussions. Sharan Majumdar and Brian Etherton provided the ETKF guidance, and Carolyn Reynolds provided the SV target guidance. Bob Kohler and Bill Barry provided computer support and guidance at HRD. The author also thanks the NOAA/Aircraft Operations Center (AOC) flight crews and dropwindsonde operators, Jack Parrish, the AOC G-IV project manager, and HRD personnel who participated in the flights. Jason Dunion, Mike Jankulak, Robert Rogers, and two anonymous reviewers provided helpful comments on an earlier version of this manuscript.

REFERENCES

  • Aberson, S. D., 2002: Two years of hurricane synoptic surveillance. Wea. Forecasting, 17 , 11011110.

  • Aberson, S. D., 2003: Targeted observations to improve operational tropical cyclone track forecast guidance. Mon. Wea. Rev., 131 , 16131628.

    • Search Google Scholar
    • Export Citation
  • Aberson, S. D., , and M. DeMaria, 1994: Verification of a nested barotropic hurricane track forecast model (VICBAR). Mon. Wea. Rev., 122 , 28042815.

    • Search Google Scholar
    • Export Citation
  • Aberson, S. D., , and J. L. Franklin, 1999: Impact on hurricane track and intensity forecasts of GPS dropwindsonde observations from the first-season flights of the NOAA Gulfstream-IV jet aircraft. Bull. Amer. Meteor. Soc., 80 , 421427.

    • Search Google Scholar
    • Export Citation
  • Aberson, S. D., , and B. J. Etherton, 2006: Targeting and data assimilation studies during Hurricane Humberto (2001). J. Atmos. Sci., 63 , 175186.

    • Search Google Scholar
    • Export Citation
  • Barrett, B. S., , L. M. Leslie, , and B. H. Fiedler, 2006: An example of the value of strong climatological signals in tropical cyclone track forecasting: Hurricane Ivan (2004). Mon. Wea. Rev., 134 , 15681577.

    • Search Google Scholar
    • Export Citation
  • Burpee, R. W., , J. L. Franklin, , S. J. Lord, , R. E. Tuleya, , and S. D. Aberson, 1996: The impact of Omega dropwindsondes on operational hurricane track forecast models. Bull. Amer. Meteor. Soc., 77 , 925933.

    • Search Google Scholar
    • Export Citation
  • Hodyss, D., , and S. J. Majumdar, 2007: The contamination of ‘data impact’ in global models by rapidly growing mesoscale instabilities. Quart. J. Roy. Meteor. Soc., 133 , 18651875.

    • Search Google Scholar
    • Export Citation
  • Liu, Q., , T. Marchok, , H-L. Pan, , M. Bender, , and S. Lord, 2000: Improvements in hurricane initialization and forecasting at NCEP with global and regional (GFDL) models. NCEP Tech. Proc. Bull. 472, 7 pp.

  • Majumdar, S. J., , S. D. Aberson, , C. H. Bishop, , R. Buizza, , M. S. Peng, , and C. A. Reynolds, 2006: A comparison of adaptive observing guidance for Atlantic tropical cyclones. Mon. Wea. Rev., 134 , 23542372.

    • 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
  • Reynolds, C. A., , M. S. Peng, , S. J. Majumdar, , S. D. Aberson, , C. H. Bishop, , and R. Buizza, 2007: Interpretation of adaptive observing guidance for Atlantic tropical cyclones. Mon. Wea. Rev., 135 , 40064029.

    • Search Google Scholar
    • Export Citation
  • Woollen, J. R., 1991: New NMC operational OI quality control. Preprints, Ninth Conf. on Numerical Weather Prediction, Denver, CO, Amer. Meteor. Soc., 24–27.

  • Zhang, R., , C. Snyder, , and R. Rotunno, 2003: Effects of moist convection on mesoscale predictability. J. Atmos. Sci., 60 , 11731185.

Fig. 1.
Fig. 1.

GFS track forecasts for Hurricane Frances initialized at 0000 UTC 2 Sep 2004. Positions are plotted every 6 h through 5 days and are marked by a number every 12 h. BEST is the best track.

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Fig. 2.
Fig. 2.

Dropwindsonde observations from the 1 Sep 2004 surveillance mission at (a) 925, (b) 500, (c) 400, and (d) 300 hPa, in standard format except that the moisture data are indicated by relative humidity (%).

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Fig. 3.
Fig. 3.

Skew T–logp diagram of the dropwindsonde observation taken near Grand Bahama Island at 2331 UTC 1 Sep 2004. Values near 275 hPa are from the aircraft flight values reported along with the dropwindsonde data.

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Fig. 4.
Fig. 4.

The 400-hPa geopotential heights for the (a)–(d) GFSO and (e)–(h) GFSN at 0, 12, 36, and 60 h into the forecast initialized at 0000 UTC 9 Sep 2005.

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Fig. 4.
Fig. 4.

(Continued)

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Fig. 5.
Fig. 5.

Trajectory of a dropwindsonde released in the eyewall of Hurricane Emily 16 Jul 2005. The “X” marks the location provided on the TEMPDROP message and represents the only location available for assimilation of the dropwindsonde data.

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Fig. 6.
Fig. 6.

GFS track forecasts for Hurricane Ophelia initialized at 0000 UTC 9 Sep 2005. Positions are plotted every 6 h and marked by a number every 12 h. BEST is the best track.

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Fig. 7.
Fig. 7.

(top) Initial condition and (bottom) 12-h forecast differences in the 850–200-hPa mean wind between the GFSN and GFSO for the 0000 UTC 9 Sep 2005 surveillance mission.

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Fig. 8.
Fig. 8.

The 0-, 12-, 36-, 60-, and 72-h forecasts of 400-hPa vorticity from the (a)–(e) GFSO and (f)–(i) GFSN initialized at 0000 UTC 9 Sep 2005.

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Fig. 8.
Fig. 8.

(Continued)

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Fig. 9.
Fig. 9.

Targeting guidance from (top) Navy total energy SVs, (middle) ETKF from the NCEP global ensemble, and (bottom) NCEP global ensemble DLM wind variance. The box and circle define the verification regions for the respective technique.

Citation: Monthly Weather Review 136, 8; 10.1175/2007MWR2192.1

Table 1.

GFS track forecast errors (km) for Hurricane Frances initialized at 0000 UTC 2 Sep 2004. GFSN is the GFS with dropwindsonde data removed from the data assimilation; GFSO is the operational GFS.

Table 1.
Table 2.

As in Table 1, but for a homogeneous sample of Hurricane Ivan forecasts from 1800 UTC 6 Sep to 1200 UTC 16 Sep 2004. GFIC is the GFS with dropwindsonde data within 111.1 km of the specified TC center location removed from the data assimilation, and N is the number of cases. Boldface values are statistically significantly better than the GFSO forecasts at the 90% level.

Table 2.
Table 3.

As in Table 2, but for a homogeneous sample of Hurricane Charley forecasts from 1800 UTC 11 Aug to 0000 UTC 13 Aug 2004.

Table 3.
Table 4.

As in Table 1, but for Hurricane Ophelia initialized at 0000 UTC 9 Sep 2005.

Table 4.
Table 5.

As in Table 1, but for Hurricane Wilma initialized at 0000 UTC 20 Oct 2005.

Table 5.

1

Dropwindsonde humidity data were not assimilated into the GFS during 2004.

2

Because of space constraints at NCEP, operational data are only available for 1 yr. A rerun of this case removing the suspect observation is therefore not possible.

3

Statistical significance calculations and the removal of serial correlation between successive tracks are presented in Aberson and DeMaria (1994).

4

Barrett et al. (2006) report a similar northward bias for Hurricane Emily (2005). The modification to operations reported here occurred the week after Hurricane Emily.

5

Details of each technique and their likely utility for tropical cyclone track targeting are provided in Majumdar et al. (2006) and Reynolds et al. (2007). The important point is that the available techniques suggest that small perturbations in the region of the increment are likely to grow rapidly in time.

Save