• Alves, J. H. G. M., , Greenslade D. J. M. , , and Banner M. L. , 2002: Impact of a saturation-dependent dissipation source function on operational hindcasts of wind-waves in the Australian region. Global Atmos. Oceanic Syst., 8 , 239267.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Atlas, R., and Coauthors, 2001: The effects of marine winds from scatterometer data on weather analysis and forecasting. Bull. Amer. Meteor. Soc., 82 , 19651990.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Beljaars, A. C. M., 1995: The parametrization of surface fluxes in large-scale models under free convection. Quart. J. Roy. Meteor. Soc., 121 , 255270.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Beljaars, A. C. M., , and Betts A. K. , 1992: Validation of the boundary layer representation in the ECMWF model. Proc. Seminar on Validation of Models over Europe, Vol. II, European Centre for Medium-Range Weather Forecasting, Shinfield Park, United Kingdom, 159–195.

  • Bender, L. C., 1996: Modification of the physics and numerics in a third-generation ocean wave model. J. Atmos. Oceanic Technol., 13 , 726750.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bosart, L. F., , and Lin S. C. , 1984: Diagnostic analysis of the Presidents’ Day storm of February 1979. Mon. Wea. Rev., 112 , 21482177.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Buehner, M., 2002: Assimilation of ERS-2 scatterometer winds using the Canadian 3D-var. Atmos.–Ocean, 40 , 361376.

  • CERSAT, 1999: Off-line wind scatterometer ERS products—User manual. C2-MUT-W-01-IF, IFREMER, Centre de Brest, 86 pp.

  • Charnock, H., 1955: Wind stress on a water surface. Quart. J. Roy. Meteor. Soc., 81 , 639640.

  • Chelton, D. B., , and Freilich M. H. , 2005: Scatterometer-based assessment of 10-m wind analysis from the operational ECMWF and NCEP numerical weather prediction models. Mon. Wea. Rev., 133 , 409429.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., , Schlax M. G. , , Freilich M. H. , , and Milliff R. F. , 2004: Satellite radar measurements reveal short-scale features in the wind stress field over the World Ocean. Science, 303 , 978983.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cohn, S. E., , da Silva A. , , Guo J. , , Sienkiewicz M. , , and Lamich D. , 1998: Assessing the effects of data selection with the DAO physical-space statistical analysis system. Mon. Wea. Rev., 126 , 29132926.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • ECMWF, 1999: Integrated forecasting system technical documentation. Part IV: Physical processes. ECMWF, Shinfield Park, United Kingdom, 142 pp.

  • Garratt, J. R., 1992: The Atmospheric Boundary Layer. Cambridge University Press, 316 pp.

  • Greenslade, D. J. M., , Schulz E. W. , , Kepert J. D. , , and Warren G. , 2005: The impact of the assimilation of scatterometer winds on surface wind and wave forecasts. J. Atmos. Ocean Sci., 10 , 261287.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hess, G. D., , and Garratt J. R. , 2002: Evaluating models of the neutral, barotropic planetary boundary layer using integral measures. Part I. Overview. Bound.-Layer Meteor., 104 , 333358.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hess, G. D., , Colman R. A. , , and McAvaney B. J. , 1995: On computing screen temperatures, humidities and anemometer-height winds in large-scale models. Aust. Meteor. Mag., 44 , 139145.

    • Search Google Scholar
    • Export Citation
  • Holland, G. J., , Lynch A. H. , , and Leslie L. M. , 1987: Australian east-coast cyclones. Part I: Synoptic overview and case study. Mon. Wea. Rev., 115 , 30243036.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hopkins, L. J., 1997: Australian east-coast cyclones and potential vorticity interactions Ph.D. thesis, Monash University, 170 pp.

  • Isaksen, L., , and Janssen P. A. E. M. , 2004: Impact of ERS scatterometer winds in ECMWF’s assimilation system. Quart. J. Roy. Meteor. Soc., 130 , 17931814.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • JPL, 2001: QuikSCAT science data product user’s manual. Version 2.1, Jet Propulsion Laboratory, California Institute of Technology, 86 pp.

  • Kaimal, J. C., , and Finnigan J. J. , 1994: Atmospheric Boundary Flows: Their Structure and Measurement. Oxford University Press, 289 pp.

  • Kepert, J. D., , Greenslade D. J. M. , , and Hess G. D. , 2005: Assessing and improving the marine surface winds in the Bureau of Meteorology numerical weather prediction systems. BMRC Research Rep. 105, Bureau of Meteorology, 61 pp.

  • Kitaigorodskii, S. A., , and Volkov Y. A. , 1965: On the roughness parameter of the sea surface and the calculation of momentum flux in the near-water layer of the atmosphere. Akad. Nauk SSR, Izv. Atmos. Oceanic Phys., 1 , 566574.

    • Search Google Scholar
    • Export Citation
  • Large, W. G., , Morzel J. , , and Crawford G. B. , 1995: Accounting for surface wave distortion of the marine wind profile in low-level ocean storms wind measurements. J. Phys. Oceanogr., 25 , 29592971.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Leidner, S. M., , Isaksen L. , , and Hoffman R. N. , 2003: Impact of NSCAT winds on tropical cyclones in the ECMWF 4DVAR assimilation system. Mon. Wea. Rev., 131 , 326.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Leslie, L. M., , Holland G. J. , , and Lynch A. M. , 1987: Australian east-coast cyclones. Part II: Numerical modeling study. Mon. Wea. Rev., 115 , 30373053.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lorenc, A. C., 1981: A global three-dimensional multivariate statistical interpolation scheme. Mon. Wea. Rev., 109 , 701721.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lorenc, A. C., 1986: Analysis methods for numerical weather prediction. Quart. J. Roy. Meteor. Soc., 112 , 11771194.

  • Louis, J-F., 1979: A parametric model of vertical eddy fluxes in the atmosphere. Bound.-Layer Meteor., 17 , 187202.

  • Louis, J. F., , Tiedtke M. , , and Geleyn J. F. , 1982: A short history of the operational PBL parametrization at ECMWF. Proc. Workshop on Boundary Layer Parametrization, ECWMF, Shinfield Park, United Kingdom, 59–79.

  • Lynch, P., , and Huang X-Y. , 1992: Initialization of the HIRLAM model using a digital filter. Mon. Wea. Rev., 120 , 10191034.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • McIntosh, P. C., , and Hubbert G. D. , 1992: Ocean winds for marine modeling. Aust. Meteor. Mag., 40 , 6169.

  • Mears, C. A., , Smith D. K. , , and Wentz F. J. , 2001: Comparison of Special Sensor Microwave Imager and buoy-measured wind speeds from 1987 to 1997. J. Geophys. Res., 106 , 1171911729.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Monaldo, F., 1988: Expected differences between buoy and radar altimeter estimates of wind speed and significant wave height and their implications on buoy–altimeter comparisons. J. Geophys. Res., 93 , 22852302.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Park, K-A., , Cornillon P. , , and Codiga D. L. , 2006: Modification of surface winds near ocean fronts: Effects of Gulf Stream rings on scatterometer (QuikSCAT, NSCAT) wind observations. J. Geophys. Res., 111 .C03021, doi:10.1029/2005JC003016.

    • Search Google Scholar
    • Export Citation
  • Puri, K., , Dietachmayer G. , , Mills G. A. , , Davidson N. E. , , Bowen R. A. , , and Logan L. W. , 1998: The new BMRC Limited Area Prediction System, LAPS. Aust. Meteor. Mag., 47 , 203233.

    • Search Google Scholar
    • Export Citation
  • Qi, L., , Leslie L. M. , , and Speer M. S. , 2006: Climatology of cyclones over the southwest Pacific: 1992–2001. Meteor. Atmos. Phys., 91 , 201209.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sanders, G., , and Gyakum J. R. , 1980: Synoptic–dynamic climatology of the “bomb.”. Mon. Wea. Rev., 108 , 15891606.

  • Seaman, R., , Bourke W. , , Steinle P. , , Hart T. , , Embery G. , , Naughton M. , , and Rikus L. , 1995: Evolution of the Bureau of Meteorology’s Global Assimilation and Prediction System, Part 1: Analyses and initialization. Aust. Meteor. Mag., 44 , 118.

    • Search Google Scholar
    • Export Citation
  • Spark, E., , Batt K. , , Gage B. , , and Zmijewski P. , 2002: Forecasts for the Sydney to Southport yacht race. Economic benefits of meteorological forecasts—A case study. Abstracts, Ninth Australian Meteorological and Oceanographic Society Annual Conf., Melbourne, Victoria, Australia, Melbourne University, 66.

  • Steinle, P. J., 2005: Generalized statistical interpolation. Fourth Symp. on Assimilation of Observations for Meteorology and Oceanography, Prague, Czech Republic, WMO, CD-ROM.

  • Stoffelen, A., 1998: Toward the true near-surface wind speed: Error modeling and calibration using triple collocation. J. Geophys. Res., 103 , 77557766.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., , and Anderson D. , 1997: Scatterometer data interpretation: Measurement space and inversion. J. Atmos. Oceanic Technol., 14 , 12981313.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tang, Y. M., , Smith N. , , and Greenslade D. J. M. , 2000: Comparison of model and observed surface winds. Aust. Meteor. Mag., 49 , 2335.

  • Troen, I., , and Mahrt L. , 1986: A simple model of the atmospheric boundary layer: Sensitivity to surface evaporation. Bound.-Layer Meteor., 37 , 129148.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wilkerson, J. C., , and Earle M. D. , 1990: A study of differences between environmental reports by ships in the voluntary observing program and measurements from buoys. J. Geophys. Res., 95 , 33733385.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Yuan, X., 2004: High-wind-speed evaluation in the Southern Ocean. J. Geophys. Res., 109 .D13101, doi:10.1029/2003JD004179.

  • View in gallery

    The ratio of 10-m wind speed to the geostrophic wind speed according to Rossby number similarity theory, as a function of (top) Charnock coefficient and (bottom) sea temperature (Ts) and air temperature (Ta) difference, for geostrophic wind speeds of 5 (solid), 10 (dashed), and 20 m s−1 (dotted). Other parameters are (top) Coriolis ( f ) = 10−4 s−1 and TsTa = 0, and (bottom) αC = 0.0185. (top) Charnock values of 0.011 and 0.018 are indicated.

  • View in gallery

    Location of buoys for which 10-m wind velocity observations are available.

  • View in gallery

    Time series of buoy 10-m wind observations (solid line), GASP-analyzed 10-m wind speed (dashed), and GASP 24-h forecast 10-m winds (dotted) during September 2001.

  • View in gallery

    As in Fig. 3 but during January 2002.

  • View in gallery

    Scatterplots of the GASP (analysis and 24-h forecast) vs buoy-mounted anemometer 10-m wind speeds for September 2001 and January 2002, shown in Figs. 3 and 4.

  • View in gallery

    (a) GASP U10, (b) LAPS-375 U10, and (c) the difference between the two wind fields (GASP − LAPS-375) for analysis forecasts valid at 0000 UTC 22 Sep 2001. Areas where GASP > LAPS-375 are shaded.

  • View in gallery

    Comparison between LAPS-375 and GASP operational 10-m wind speeds, for analyses and 24-h forecasts, at 12-hourly intervals for (a), (b) September 2001 and (c), (d) January 2002. Statistics are displayed in Table 3. Contours are at 1000, 2000, 4000, 6000, and 8000 observations.

  • View in gallery

    Comparison between LAPS-375 and GASP analyses valid at 1200 UTC 17 Sep 2001.

  • View in gallery

    (top) Scatterometer wind speed (m s−1) and analyzed LAPS-375 MSLP (hPa, contoured every 4 hPa), (middle) LAPS-375 18-h forecasts of wind speed and MSLP, all valid at 1100 UTC 18 Jan 2002, and (bottom) wind speed difference.

  • View in gallery

    Contoured 2D histograms for GASP over the Australian domain 24-h forecast (y axis) against scatterometer winds during 2003 and 2004 for (a) wind speed, (b) wind direction, and (c) zonal and (d) meridional components. The line y = x (light) and the line of best fit described in the text (heavy) are also shown. Contours are in geometric progression, every 101/2. The mean and standard deviation of the model − observation difference, together with the slope and correlation parameter, are described in the text and Table 4.

  • View in gallery

    As in Fig. 10 but for LAPS-375.

  • View in gallery

    The 24-h forecast − observed wind speed vs observed wind speed (in 2 m s−1 bins) for the GASP global domain. Statistics are mean (thick bar), median (thin bar), 1st and 3d quartiles (bottom and top of box), and 1st and 99th percentiles (bottom and top whisker).

  • View in gallery

    Monthly verification statistics averaged over 2003 and 2004 for GASP (global domain) − QuikSCAT (a) wind speed bias, (b) slope of linear best fit through zero for speed, and (c) zonal and (d) meridional wind components.

  • View in gallery

    As in Fig. 13 but for GASP (Australian domain) − QuikSCAT.

  • View in gallery

    As in Fig. 13 but for LAPS-375 (Australian domain) − QuikSCAT.

  • View in gallery

    As in Fig. 13 but for Meso-LAPS (Australian domain) − QuikSCAT.

  • View in gallery

    Mean monthly QuikSCAT-observed zonal winds for 2003 and 2004. Calculated for GASP global (solid), LAPS-375 (dashed), and Meso-LAPS (dotted) domains.

  • View in gallery

    The slope b0 of the line of best fit through the origin of model against scatterometer winds, as a function of forecast period (days). Models analyzed are GASP global domain (asterisks), GASP Australian region (circles), LAPS-375 Australian region (triangles), and Meso-LAPS Australian region (diamonds). Statistics are for (a), (d) wind speed, and (b), (e) zonal and (c), (f) meridional wind components for (left) September 2001 and (right) January 2002.

  • View in gallery

    Comparison of LAPS-375 18-h forecast winds and MSLP with scatterometer observations and analyzed MSLP for the Sydney–Southport storm on (top) 27, (middle) 28, and (bottom) 29 July. Columns from left to right are scatterometer observations and analyzed MSLP at 0500 UTC, LAPS-375 18-h forecast wind and MSLP forecast at 0500 UTC, scatterometer observations and analyzed MSLP at 1700 UTC, and LAPS-375 18-h forecast wind and MSLP forecast at 1700 UTC. Locations of MSLP observations are indicated for Sydney (S); Lord Howe Island (LH); buoy near 37°S, 157°E (B); and Green Cape (GC). Note that model winds are only plotted at scatterometer observation locations.

  • View in gallery

    (left) Pressure at Sydney (top), Lord Howe Island (second from top), Green Cape (third from top), and the buoy (bottom). (top right) Pressure differences between Sydney and Lord Howe Island; (middle right) Sydney and buoy near 37°S, 157°E; and (bottom right) Sydney and Green Cape, as a function of the day of July 2001: heavy line, observations; crosses, analysis; and circles, 18-h forecast.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 29 29 11
PDF Downloads 15 15 4

An Assessment of Marine Surface Winds from the Australian Bureau of Meteorology Numerical Weather Prediction Systems

View More View Less
  • 1 Bureau of Meteorology Research Centre, Melbourne, Victoria, Australia
© Get Permissions
Full access

Abstract

A method for routinely verifying numerical weather prediction surface marine winds with satellite scatterometer winds is introduced. The marine surface winds from the Australian Bureau of Meteorology’s operational global and regional numerical weather prediction systems are evaluated. The model marine surface layer is described. Marine surface winds from the global and limited-area models are compared with observations, both in situ (anemometer) and remote (scatterometer). A 2-yr verification shows that wind speeds from the regional model are typically underestimated by approximately 5%, with a greater bias in the meridional direction than the zonal direction. The global model also underestimates the surface winds by around 5%–10%. A case study of a significant marine storm shows that where larger errors occur, they are due to an underestimation of the storm intensity, rather than to biases in the boundary layer parameterizations.

Corresponding author address: Eric W. Schulz, Bureau of Meteorology Research Centre, GPO Box 1289, Melbourne, VIC 3001, Australia. Email: e.schulz@bom.gov.au

Abstract

A method for routinely verifying numerical weather prediction surface marine winds with satellite scatterometer winds is introduced. The marine surface winds from the Australian Bureau of Meteorology’s operational global and regional numerical weather prediction systems are evaluated. The model marine surface layer is described. Marine surface winds from the global and limited-area models are compared with observations, both in situ (anemometer) and remote (scatterometer). A 2-yr verification shows that wind speeds from the regional model are typically underestimated by approximately 5%, with a greater bias in the meridional direction than the zonal direction. The global model also underestimates the surface winds by around 5%–10%. A case study of a significant marine storm shows that where larger errors occur, they are due to an underestimation of the storm intensity, rather than to biases in the boundary layer parameterizations.

Corresponding author address: Eric W. Schulz, Bureau of Meteorology Research Centre, GPO Box 1289, Melbourne, VIC 3001, Australia. Email: e.schulz@bom.gov.au

1. Introduction

Marine wind estimates from numerical weather prediction (NWP) systems are an important input to the Australian Bureau of Meteorology (hereafter the Bureau) forecast and warning services to shipping. They are also used indirectly as the forcing for ocean models. As such, the diagnosis (and, if necessary, correction) of any systematic biases and errors is an important ongoing activity. Here, we describe some recent comparisons of the analyzed and forecast marine winds from the models with observations.

The Bureau has developed two operational NWP systems: a global spectral model and a finite-difference limited-area model. Currently, they use the same data assimilation scheme, and will soon use a common physics package and boundary layer scheme. Further details are provided in later sections of this paper.

The Bureau’s global and regional NWP marine winds have been verified in the past, and such comparisons are reviewed below. Other authors (e.g., Buehner 2002; Leidner et al. 2003; Isaksen and Janssen 2004; Yuan 2004) have reported on NWP wind verifications against satellite-borne scatterometers with model minus observation bias ranging from 1 to −1 m s−1. Atlas et al. (2001) provides a comprehensive review of scatterometer-related research. It is however appropriate to revisit this question in the context of the Bureau’s systems for a number of reasons. First is the current availability of scatterometer data. Previous comparisons have been severely limited by the availability of representative marine wind observations. The SeaWinds scatterometer aboard the Quick Scatterometer (QuikSCAT) satellite supplies over 1 million high quality marine wind vectors daily over the global oceans and, thus, enables a much more geographically comprehensive and statistically significant comparison than was hitherto possible (Chelton and Freilich 2005).

A second reason is that the marine winds have not been comprehensively verified in the current operational configuration of the Bureau’s NWP systems. This factor is particularly pertinent, because forecasters and other users noted a negative bias in the surface marine winds from the current regional model, the Limited- Area Prediction System with 0.375° resolution (LAPS-375; Puri et al. 1998), almost from inception, and over time LAPS-375 has developed a reputation for underforecasting marine wind speed. The marine wind bias shows up in direct use of the winds by forecasters, and also in use of the winds for forcing ocean models, particularly the wave model. There is some suggestion that it may be more serious at high winds. While the evidence from forecasters is largely anecdotal, the wave model predictions are routinely verified against buoy measurements, as documented by Kepert et al. (2005) and Greenslade et al. (2005), who conclude that observed bias in wave forecasts (and in particular significant wave height), can be considerably reduced by the use of bias-corrected wind forcing.

Sources of systematic model errors in marine near-surface winds can be conveniently divided into two types: those due to biases in the boundary layer parameterization, and those due to model biases in the intensity of weather systems. The latter can include a host of causes, such as model resolution, physical parameterizations including horizontal diffusion and cloud processes, and the initial condition. Marine boundary conditions could also be a source of error in NWP. Poor representation of sea surface temperature (SST) fields and sea state could impact on both the synoptic and boundary layer structures.

In summary, there is some reason to believe that the marine winds in LAPS-375 contain a systematic negative bias. However, this is based upon studies that are out of date, incomplete, nonsystematic, and/or anecdotal. The marine winds from LAPS-375 have not been systematically verified, and it is unclear how much of the perceived bias is real, and how much is due to inadequacies in the informal verification, or a reaction to the removal of a small positive bias in the previous version.

Here, we introduce a method to routinely verify marine surface winds, and comprehensively evaluate and document the quality of the marine wind fields in the Bureau’s operational NWP systems as a necessary first step to fixing any systematic shortcomings.

Section 2 comprises descriptions of the salient features of the atmospheric models, and a brief review of previous verifications of marine winds. Section 3 outlines the current boundary layer parameterizations within the atmospheric models, and includes a brief numerical experiment on the sensitivity of a simple boundary layer model to the Charnock coefficient. Section 4 introduces the observations used in the verification. Section 5 describes the method used to perform routine wind verifications and presents a number of new NWP surface wind verification results, while section 6 presents a case study of a storm in the Tasman Sea. Conclusions are drawn in section 7.

2. The Bureau’s NWP system

a. The atmospheric models

The Bureau currently has two operational NWP systems. These are LAPS and the Global Assimilation and Prediction System (GASP; Seaman et al. 1995). The former is run over a variety of horizontal resolutions and domains, including LAPS-375 [0.375° resolution (38 km at 25°S); 16.75°N–65°S, 65°–184.25°E], Meso-LAPS [0.125° resolution (13 km at 25°S); 4.875°N–55°S, 95°–169.875°E], TC-LAPS [0.15° resolution (17 km at the equator); only run over tropical cyclones], TXLAPS [0.375° resolution (42 km at equator); tropical domain], and LAPS-05 (0.05° resolution; various domains). Verification of the LAPS models presented here will focus on LAPS-375 and Meso-LAPS. LAPS’s vertical resolution also varies, but all the versions considered here have 29 levels, with the lowest sigma level varying near 10-m height. The global system GASP is a spectral model currently run operationally at T239 resolution, approximately equivalent to a grid spacing at the equator of 0.75°(or 83 km). The lowest of the 29 vertical sigma levels is at approximately 70-m height. The physical parameterizations in the global and local systems are largely in common, apart from some parameter settings. A significant exception to this unified physics is in the boundary layer parameterization, where the LAPS family uses the European Centre for Medium-Range Weather Forecasts (ECMWF; ECMWF 1999) scheme, while GASP uses the Louis scheme (Louis 1979; Louis et al. 1982).

The data assimilation scheme is common to both systems. At the beginning of 2003 an observation-space multivariate statistical interpolation scheme similar to that described in Lorenc (1981) was used. This was replaced by a generalized statistical interpolation scheme (Steinle 2005) using an iterative solution similar to that described by Cohn et al. (1998), in May 2004 for the global system and September 2004 for the limited-area model.

Observation sources include conventional (surface synoptic, ship, synthetic, and radiosonde), automatic (buoy and land based), satellite [radiances and retrieved profiles from the Television Infrared Operational Satellite (TIROS) Operational Vertical Sounder (TOVS) instrument, and cloud drift winds], and aircraft. Satellite scatterometer winds were not being assimilated operationally during the 2003 and 2004 study periods examined later in this paper. The following are the main differences between the global and regional systems at the time of this study.

  • The global system uses a later data cutoff time relative to the time of analysis and so receives more data in the region (especially satellite data).
  • The global system is “warm run,” while the regional system is “cold started” every 12 h. That is, the global system uses its own forecast from a previous analysis as the background field for the analysis, while the regional system relies upon a forecast from the global system as the first guess.
  • The global system uses a 1D variational technique to analyze TOVS radiances, while the regional system (at the time of this study) uses surface air temperature retrievals from the National Environmental Satellite, Data, and Information Service.
  • There are some differences in the background error covariance models and quality control, due largely to the differing resolutions.
  • The tropical members of the LAPS family (TLAPS and TC-LAPS) use diabatic initialization schemes based around satellite-derived cloud heating rates, the other members of the LAPS family use the digital filter initialization of Lynch and Huang (1992), and GASP uses nonlinear normal mode initialization.
  • The global system in its present operational configuration saves its state at 6-h intervals of forecast time, while LAPS uses 3-h intervals.

b. Previous verifications of marine winds

The previous version of the operational limited-area system, LAPS-75, was replaced by the current configuration in 1999. The main differences of relevance here were an increase in the horizontal and vertical resolutions (from 0.75° and 19 levels to 0.375° and 29 levels), and some updates to the physical parameterizations, including the replacement of the Louis (Louis 1979; Louis et al. 1982) boundary layer scheme by that of the ECMWF (1999). Subsequent changes have been confined to minor tuning of the physical constants.

The marine winds in LAPS-75 are believed to have been quite good. This is based partly on informal feedback from users, but more particularly on the verification studies by Tang et al. (2000). This involved the comparison of the winds derived from the operational model, with European Remote Sensing Satellite-2 (ERS-2) altimeter wind speed products and coastal automatic weather station observed winds. This study showed that there was generally satisfactory agreement, although LAPS-75 usually showed a small positive bias over the longer averaging periods. The discrepancies between the model and observations were more marked for shorter averaging periods, which could have been due to the model underestimating the strength of intense systems, and were most marked in the Tasman Sea (based on automatic weather station data), and to a lesser extent in Bass Strait and the Southern Ocean.

Bender (1996) briefly considered the quality of the surface wind fields from the Bureau’s regional atmospheric modeling system for the month of July 1992. At that time, the lowest-level winds from the model were adjusted to 10 m through the surface layer model of McIntosh and Hubbert (1992) rather than the present scheme of Hess et al. (1995). Bender (1996) compared observed winds from the Cape Grim Baseline station in northwest Tasmania with modeled winds at the nearest grid point, and found an rms error of 5 m s−1 and a bias (model − observation) of approximately −2 m s−1. This case, however, is a far from ideal comparison because the Cape Grim observations are taken at the top of a large cliff, and so there is likely to be a significant flow distortion effect that was not taken into account.

The quality of the Bureau’s operational surface winds was also considered by Alves et al. (2002). In that work, a comparison of LAPS 10-m winds with in situ observations from four sites around the Australian coast was performed. Two time periods were considered, covering both the current (LAPS-375) and previous (LAPS-75) operational configurations: February 2000 and 20 March–20 May 1998. No consistent pattern in the quality of the LAPS winds was detected for either time period, with the rms error of the LAPS winds at each site ranging from 2 to 3 m s−1 and the bias ranging from −1 to 1.5 m s−1. A shortcoming of this study was that no attempt was made to adjust the observations, which ranged in height from 4.5 to 44 m, to the standard height of 10 m.

Alves et al. (2002) also selected two storm events within the 1998 time period, for which modeled winds over a portion of the LAPS-75 domain were compared with in situ observations and ERS-2 scatterometer winds. The portion of the domain considered was the “fetch” of the west Tasmanian buoy, that is, the oceans to the south and west of Australia. There are no significant biases associated with ERS-2 observations of 10-m wind speed (U10) within the range 3.5 m s−1 < U10 < 15 m s−1, so comparisons and statistics were only calculated for observations falling within this range (CERSAT 1999). It was found that the rms error was approximately 3 m s−1 and that the LAPS-75 winds were biased high (approximately 0.75 m s−1). Because the mean U10 during these storm events was >9 m s−1, this indicates that the quality of the modeled winds was particularly good. However, it should be borne in mind that these time periods were chosen specifically because the modeled winds satisfied particular accuracy criteria and so these results are not necessarily representative of the overall quality of the LAPS-75 winds at that time.

Kepert et al. (2005) performed a verification of the Bureau’s NWP systems that mostly focused on September 2001 and January 2002. They found LAPS-375 underestimated the surface winds compared with QuikSCAT by around 10% with a greater bias in the meridional component. GASP was found to have a smaller bias. These earlier findings may differ from later results (including the work presented here) due to improvements in verification methods and evolution of the NWP models.

A 2-yr study (2001 and 2002) of the ECMWF 10-m wind analysis with QuikSCAT by Chelton and Freilich (2005) found the model was biased low by about 0.4 m s−1 (or 6%) after corrections for atmospheric stability. Yuan (2004) also compared ECMWF and QuikSCAT winds in the Southern Ocean and found the model biased low. The bias was more pronounced as wind speed increased [0.4 m s−1 (or 2%–3%) bias for wind speeds 15–20 m s−1, 1 m s−1 (or 5%) bias for wind speeds >20 m s−1]. These results are of relevance to this study because the Bureau’s LAPS models use the ECMWF boundary layer scheme.

3. Current wind modeling procedures

This section presents an overview of the parameterizations presently used in LAPS and GASP for the marine boundary layer. The global and limited-area models use an identical marine surface-layer parameterization, while they diverge significantly on the treatment of the boundary layer.

a. Parameterization of the atmospheric boundary layer

The Bureau’s GASP and LAPS family of models share common physics parameterizations except in the boundary layer where GASP uses the Louis scheme (Louis 1979; Louis et al. 1982), while LAPS uses the ECMWF Research Department (1995) scheme. The schemes are summarized here, with a more detailed description available in Kepert et al. (2005).

The Louis boundary layer scheme parameterizes the turbulent viscosity and diffusivity above the lowest model layer as a function of the bulk Richardson number for stable and unstable conditions. A computational value of the height of the boundary layer, used to determine the appropriate stability dependence for the turbulent mixing, is found by calculating two height scales: a dynamic height scale and a convective height scale. The top of the boundary layer is given by the maximum of these two heights. This scheme is computationally efficient as it relies on lookup tables, but tends to give mixed layer heights that are too low in convective conditions, due to an underestimation of the entrainment at the top of the boundary layer (Beljaars and Betts 1992).

The boundary layer parameterization used in LAPS changed in 1999 from the Louis scheme to that used by the ECMWF Research Department (1995). The differences in the schemes are primarily limited to unstable atmospheric conditions where the turbulent viscosity and diffusivity above the lowest layer are based on the work of Troen and Mahrt (1986) and are computed as a function of the turbulent Prandtl number.

b. Parameterization of the marine surface layer

A scheme for parameterizing the vertical profiles of wind velocity, temperature, and moisture between the lowest dynamic sigma levels in GASP and LAPS and the surface, together with their fluxes through this layer, is based on Monin–Obukhov similarity theory (MOST). The theory is developed in detail in many books (e.g., Garratt 1992, 49–60; Kaimal and Finnigan 1994, 10–21), and a full description of the Bureau’s implementation can be found in Kepert et al. (2005); hence, only a brief description will be provided here.

The surface fluxes of momentum, heat, and moisture are used to define characteristic scales u* (the friction velocity), θ*, and q*, respectively, through
i1520-0434-22-3-613-e1
i1520-0434-22-3-613-e2
i1520-0434-22-3-613-e3
where τ is the surface stress, Hs the surface sensible flux, Hl the surface latent flux, ρ is the air density, Cp the specific heat of air at constant pressure, and Le the latent heat of vaporization of water.
The characteristics of the surface are parameterized through the roughness lengths. A key issue in applying MOST over the sea is that the roughness lengths can be expected to vary with the sea state. This in turn depends on the surface stress, among other things. The roughness length for momentum z0, used here, employs the parameterization of Beljaars (1995), consisting of a low wind speed viscous limit and the usual Charnock (1955) formula:
i1520-0434-22-3-613-e4
where aM = 0.11, ν is the kinematic molecular viscosity of air, and the Charnock coefficient αc is constant at 0.011. This has the effect of imposing a nonzero lower bound on z0, and accounts for the aerodynamically smooth flow at wind speeds of less than about 2 m s−1. Estimates of αc range from 0.0067 (Charnock 1955) to 0.035 (Kitaigorodskii and Volkov 1965), with 0.011 being a widely used value. Garratt (1992, Table 4.1) gives an overview of commonly used Charnock coefficients. The Charnock coefficient is known to depend on wave age (younger waves are steeper) and water depth (shallow water waves are steeper) among others, as well as related parameters; for example, short fetch implies younger wind sea and hence a higher αc.

A particular issue in the verification of near-surface winds is that both the observation and the model need to be adjusted to a common height. This is best accomplished using a boundary layer model. Here, we shall use the operational marine 10-m winds diagnosed by the physically based method of Hess et al. (1995) from both global and regional NWP systems. The operational marine 10-m winds, together with the derived surface stress, are saved at high temporal resolution for forcing the wave and other ocean models. GASP outputs a 6-h time-averaged wind every 6 h, while LAPS outputs instantaneous values hourly.

c. Boundary layer sensitivity to the Charnock coefficient

It is clear that the surface-layer schemes are subject to a number of empirically determined functional forms and parameter values. Determining these has been and remains an active research topic. It will be helpful in understanding the remainder of this paper to have some feel for the sensitivity of the boundary layer structure to the parameterization of the surface roughness lengths. For this purpose a simplified boundary layer model is sufficient, and here we adopt Rossby number similarity theory (e.g., Garratt 1992, 43–48). This matches a MOST-based surface layer to a spiral layer in the rest of the boundary layer, and has as its inputs the surface roughness and sea–air temperature difference, the geostrophic wind, Coriolis parameter, and boundary layer depth. It has been extensively studied and tested against observations, and a discussion of the neutral stability case may be found in Hess and Garratt (2002).

The variation of the ratio U10/Ug with the Charnock coefficient is shown in the top panel of Fig. 1, where U10 is the wind speed at 10 m and Ug is the geostrophic wind speed. Charnock values of 0.018 and 0.011 are indicated. As expected, the ratio decreases with increasing roughness. However, the changes are not large, and the range plotted (which exceeds the range of measurements in the literature) corresponds to about a 5% change in the 10-m wind speed. The ratio is lower for high wind speeds largely because the Charnock relation gives greater surface roughness there. For comparison, the sensitivity to stability, as measured by the sea–air temperature difference, is shown in the lower panel of Fig. 1, and can be seen to be somewhat larger, particularly at low wind speeds.

In December 2001, the Charnock coefficient was changed in LAPS-375, from 0.018 to 0.011, and the resulting impacts on surface marine winds will be examined in the following sections. This simple model suggests that the impact of the changes in the Charnock coefficient, while significant, are less important than the correct specification of the stability effects in parameterizing the boundary layer winds, particularly at low to moderate wind speeds. It also suggests that large errors in the surface wind speed are unlikely to be repairable by adjusting the Charnock coefficient, although there may be small impacts.

4. Data

In this work, NWP verification is performed by comparison with observations obtained from in situ and remotely sensed methods. Here, we are using observations to “ground truth” the NWP wind, which implies that the errors and uncertainties associated with the observations are significantly smaller than those associated with the modeled winds. Therefore, it is instructive to first discuss the limitations in accuracy and applicability associated with the observations used in our verification, before describing the observations in some detail.

When using conventional observations of marine winds from, for example, coastal buoys, the verification method may be confounded by coastal effects in two ways. The observations may include the effects of local processes (e.g., sea breezes) that are not resolved by the model, and care must be taken in interpolating the model winds to the observation point to avoid using grid points over the land. The first of these considerations eliminates a large part of the conventional observation database; there are not many wind observations with true open-ocean exposure. Thus, we shall rely heavily on remotely sensed data, in particular satellite scatterometer winds.

A general discussion of scatterometers and the interpretation of their data may be found in Stoffelen and Anderson (1997), while Stoffelen (1998) discusses the relative errors of ship, buoy, and scatterometer winds. For the purposes of comparison with models, scatterometer winds are shown to be the most accurate, not least because they are a spatial average over a horizontal scale of about 25 km, and so are less subject to errors of representativeness (Lorenc 1986) than point measurements. Comparisons between the U.S. National Data Buoy Center (NDBC) open-ocean observations and QuikSCAT over a 2-yr period reveal accuracy in the scatterometer winds of around 1.7 m s−1 over a large range of wind speeds (Chelton and Freilich 2005).

Two potential sources of uncertainty are introduced by our verification method. First, QuikSCAT winds are calculated for a neutrally stable atmosphere, while NWP products are an estimate of the actual wind. This discrepancy may cause the NWP winds to be biased low by around 0.2 m s−1 compared with the equivalent neutral-stability wind (Chelton and Freilich 2005; Mears et al. 2001). Poor representation of the SST fields that control the buoyancy forcing of the atmospheric boundary layer will exacerbate uncertainties in the NWP atmospheric stability (Chelton and Freilich 2005). Observations of QuikSCAT winds over warm- and cold-core eddies show changes in wind speeds of up to 10% due to modification of the atmospheric stability (Park et al. 2006). Second, the QuikSCAT winds are calculated relative to the ocean surface, while the NWP wind is relative to a fixed earth. For regions of strong ocean currents, there could be biases as large as 1 m s−1 (Chelton et al. 2004). This could be particularly important for areas affected by the East Australian Current and the Antarctic Circumpolar Current in the LAPS-375 domain. Another potential source of error in model–observation comparisons is a “pseudobias” that can arise due to the nonlinear dependence of wind speed on the wind vector components, although this effect only becomes significant in low wind speeds (Stoffelen 1998).

NWP wind verification against in situ observations also introduces uncertainties. The stated rms accuracy of anemometers mounted on buoys is 0.5 m s−1 or 10% of the wind speed, whichever is greater (Monaldo 1988). In addition, there are other impacts on the accuracy of anemometers in marine conditions that should be considered. For example, the anemometer may be so close to the surface that is it partly shielded by waves in strong wind conditions (Large et al. 1995). The quality of the wind speed measurements from buoys is discussed in Wilkerson and Earle (1990) and in Monaldo (1988).

a. In situ observations

There are very few in situ observations of marine winds within the Australian region, and the few that do exist were unfortunately of poor quality during the time periods of interest here, containing substantial portions of missing data. Therefore, the in situ observations used here are obtained from anemometers mounted on three discus buoys operated by the NDBC, and located in the Gulf of Mexico (see Fig. 2). The NDBC maintains a large database of marine winds from anemometers mounted on buoys; however, these are typically measured at 5 m above the surface. The observations used here are limited to those that are measured directly at 10 m above the surface. This avoids any error involved in converting the observed winds to the standard height of 10 m. The wind speed obtained from the anemometers is averaged over an 8-min period and reported hourly. These data are not presently used in the assimilation system and so provide an independent test of the modeled surface winds.

b. QuikSCAT scatterometer data

Scatterometer wind data are ideal for comparisons with the NWP wind field (Chelton and Freilich 2005), as (i) they are available over the whole of the marine part of the model domain with high spatial and temporal density; (ii) apart from some known quality control issues such as rain contamination and directional ambiguities, they are highly accurate; and (iii) they are not currently used in the operational LAPS and GASP systems,1 and so provide an independent test.

The SeaWinds instrument aboard the QuikSCAT satellite is described in detail in JPL (2001), from which we summarize a few important points. In the mode used here, it produces winds on an 1800-km-wide swath at a horizontal spacing of 25 km, with an orbital period of 101 min, covering about 90% of the ice-free oceans per day. QuikSCAT employs a conical-scanning pencil-beam antenna, in contrast to the fan-beam antenna of earlier spaceborne scatterometers. This has the advantage of a wide swath, but the disadvantage that the edges and nadir of the swath are observed from azimuths that differ by close to 0° or 180°, rather than the optimal 90°. This lack of so-called azimuthal diversity means that the winds in this part of the swath are of lower accuracy. Another feature of QuikSCAT is that it operates in the Ku band, rather than the C band, and is thus more sensitive to the rain attenuation of the radar signal. In heavy rain, similar radar backscatter is received from all azimuths, which the retrieval interprets as a wind direction that is normal to the satellite direction of travel. Rain flags are included in the wind vector field that can be used to exclude rain-affected wind observations. Finally, the process of calculating the wind that best fits the observed radar backscatter generally produces from two to six solutions in the minimization of the statistical cost function. These solutions are known as ambiguities, and an important part of the retrieval is to select the correct one. This process, which includes tests of the goodness of fit, neighbor checks, and comparison with NWP data, does nevertheless occasionally select an incorrect vector. Earlier scatterometers had relatively simple antenna geometries and the ambiguities were often close to 180° in direction apart. However, the more complex geometry of the QuikSCAT antenna system means it is difficult to make generalizations about the nature of the ambiguities.

5. Verification of surface winds

In this section, we present the results of some recent verification studies of the marine near-surface winds from LAPS and GASP. Comparisons are between GASP and in situ buoys in the Gulf of Mexico, a model intercomparison between GASP and LAPS-375, and GASP and LAPS against satellite observations. We present results for both long-term statistics and for case studies. Within this section, the marine winds are examined over a 2-yr period, 2003 and 2004, and also two months, September 2001 and January 2002, which represent LAPS’s performance before and after the Charnock coefficient was changed from 0.018 to 0.011. Table 1 summarizes the various comparisons that are presented in this and the following section.

a. Comparison of GASP against in situ observations

To ensure compatibility with the time scales included in the GASP winds, the buoy data are smoothed with a 7-hourly boxcar window. GASP winds used are the operational 10-m winds interpolated to a 1° spatial resolution. Both the analysis and the 24-h forecasts are considered. Figures 3 and 4 show time series of observed 10-m wind speed at each buoy along with the analysis and 24-h forecasts from GASP for the two monthly time periods. These plots show that the variability in the wind speed on scales of 5–10 days is captured well by GASP. Buoy 42003 disagrees with GASP between 12 and 15 September 2001, with an approximate 1-day lag in the forecast. This event is associated with the presence of Hurricane Gabrielle (11–19 September 2001), which formed in the east of the Gulf of Mexico before moving across Florida and out into the Atlantic.

Scatterplots and verification statistics for each time period are displayed in Fig. 5 and Table 2. During the first time period (September 2001; Figs. 5a and 5b), it can be seen that the GASP analysis and 24-h forecasts have negligible systematic bias. (The mean observed wind speed at these buoys during this time period is 5.3 m s−1). During January 2002 (Figs. 5c and 5d), the bias in the global wind speeds is more pronounced, with a negative bias of almost 1 m s−1 in the analysis. (Mean observed wind speed during this time period is 7.2 m s−1.) This bias is reduced in the 24-h forecasts. As expected, the variable errors (R and SI) are larger for the 24-h forecast than the analysis.

b. Comparison of LAPS and GASP

In this section, an intercomparison of 10-m wind speeds between GASP and LAPS-375 is performed over the LAPS-375 domain. Similar to the GASP winds described above, the LAPS-375 10-m winds used here are the operational 10-m winds, interpolated to 1°, and archived at 12-hourly intervals.

Figures 6a and 6b show examples of an analyzed 10-m wind field from GASP and LAPS-375, respectively. The main features of these two wind fields are similar; however, it can be seen that there is more variability on small scales in the LAPS-375 wind field, and the highest wind speeds over the domain at this time occur in LAPS-375. These effects are likely to be due to the fact that LAPS-375 winds are instantaneous winds whereas the GASP winds represent averages over a 6-h time period. The difference between these two wind fields (GASP − LAPS-375) is shown in Fig. 6c. There does not seem to be any systematic pattern to the difference between the LAPS-375 and GASP surface winds here. The overall bias (GASP − LAPS-375) at this time is approximately 0.4 m s−1.

Figure 7 shows contoured scatterplots of LAPS versus GASP-375 10-m wind speeds for the months of September 2001 and January 2002. These time periods were chosen to allow for the evaluation of the impact of changing the Charnock coefficient in LAPS-375. Comparison statistics are also displayed (Table 3), with the bias defined as GASP − LAPS-375. During September 2001 (Figs. 7a and 7b), there does not seem to be any significant systematic bias between LAPS-375 and GASP for low wind speeds (<5 m s−1). However, for wind speeds above 5 m s−1, there is a tendency for LAPS-375 to underestimate the wind speed compared with GASP. This is reflected in the bias statistics, with the LAPS-375 24-h forecast wind speeds being more than 0.5 m s−1 lower than the GASP 24-h forecast winds.

During the second time period, January 2002 (Figs. 7c and 7d), there is negligible systematic bias between the LAPS-375 and GASP analyses. However, it can be seen that there is still a tendency for LAPS-375 to underestimate the wind speed compared to GASP for high wind speeds, as the peak of the 1000-level contour lies below the y = x line. This tendency for LAPS-375 to be lower than GASP for high wind speeds is of concern when the instantaneous nature of the LAPS-375 winds is taken into account. Figure 8 shows a comparison of the LAPS-375 and GASP wind speeds for one time level: 1200 UTC 17 September. It can be seen that there are several LAPS-375 winds in the range 20–35 m s−1, while the maximum wind speed in GASP is approximately 21.5 m s−1. This difference arises because the LAPS-375 winds are instantaneous “snapshots” of the surface winds and therefore contain more variability than the smoothed (in time) GASP fields. However, despite these high LAPS-375 values, on average, the contours show that the GASP winds are higher than LAPS-375 for high wind speeds. If the LAPS-375 winds were smoothed to the same temporal resolution as the GASP winds, then it is likely that the systematic bias would become more pronounced.

c. Comparison of LAPS and GASP winds with scatterometer data

Here, we present comparisons of the LAPS-375, Meso-LAPS, and GASP forecast marine winds with scatterometer data.

1) example wind field

Before deriving overall verification statistics for the various models, single time-level snapshots were examined. Here, we provide an example of the spatial distribution of the NWP and observed surface winds and their differences. This is provided here to set the context for the domain-wide, time-averaged error statistics introduced in later sections. Additional insight is gained from this case study by overlaying mean sea level pressure (MSLP) information. Figure 9 shows the scatterometer (with rain-contaminated observations removed) and LAPS-375 18-h forecast wind speeds, and analyzed and 18-h forecast MSLPs, valid for 1100 UTC 18 January. Differences are small, but include the following.

  • (i) The low pressure system near the southern tip of New Zealand and another near 48°S, 122°E are weaker in the forecast than in the analysis valid at the same time.
  • (ii) The band of southerlies to the east of the front associated with the low pressure system at 48°S, 122°E is 2–3 m s−1 weaker in the model.
  • (iii) The effects of rain contamination north of 15°S are apparent, where the speckled character of the scatterometer speed observations is suggestive of convective rain.

It should be noted that some of the difference in speed near the low centers might also be due to rain contamination of the scatterometer data. There will also be some error associated with the assumption that the scatterometer data within a 6-h collection period are instantaneous. This can be observed as the swath mismatch southwest of New Zealand (Fig. 9, top panel). It is clear that the largest differences are associated with spatially discrete but large-scale synoptic elements, although it is difficult to ascribe the errors to a particular source. We believe the differences are most likely due to errors in the model depiction of those particular synoptic features, but we cannot discount the possibility that the model boundary layer, which exhibits a small systematic wind bias on average, may have degraded performance under low pressure systems.

2) A method for routine verification applied to a 2-yr statistical study

All available QuikSCAT data for the 2 yr from 1 January 2003 to 31 December 2004 have been used to derive verification statistics. They were collected into 6-hourly periods centered about 0000, 0600, 1200, and 1800 UTC. Verification is applied over a domain of approximately 5° inside each respective model domain to avoid boundary effects: the full globe for GASP; 60°S–10°N, 70°E–180° for LAPS-375; and 0°–50°S, 100°–160°E for Meso-LAPS. Additionally, the GASP verification is performed over the reduced LAPS-375 verification domain. Approximately 45 000 observations were available at each 6-hourly period, 5.6 million for a month, or 134 million for a 2-yr period. Modeled winds from the 12- and 24-h forecasts are interpolated to 10-m height as described in Hess et al. (1995), and horizontally to the observation point by cubic splines, and compared to observations where the observed wind speed is greater than 3 m s−1. In the verification, we focus on the 12- and 24-h forecasts, because differences in earlier periods are likely to be due to initialization errors, while differences later in the period are more likely to be due to the synoptic pattern being forecast incorrectly. The 12–24-h forecast period should be a good diagnostic of the quality of the boundary layer physical parameterizations. The QuikSCAT data are of sufficiently high quality, particularly when rain flags are used to eliminate data affected by precipitation. No additional quality control was applied except where noted below.

Two-dimensional histograms of the comparison between NWP forecasts and observations have been generated. Here, we restrict ourselves to histogram examples for GASP on the Australian domain (Fig. 10) and LAPS-375 (Fig. 11) 24-h forecasts for 2003 and 2004. Note that histograms of GASP on the global domain and Meso-LAPS are qualitatively similar to Figs. 10 and 11. In each figure the forecast wind (speed, direction, westerly and southerly components) is on the y axis, and the corresponding observations on the x axis. The histogram counts in each cell are contoured, with logarithmic contour spacing with a ratio of 101/2 between contours. Thus, the diagrams typically cover three orders of magnitude of count density, from approximately 3 to 3000. Means and standard deviations of the differences, together with the slope and correlation coefficient of the line of best fit through the origin, are displayed in Table 4 for GASP and LAPS-375, as well as for the other models and domains.

There is on average a 0.7 m s−1 negative bias of the model wind speed relative to the scatterometer-derived wind for the GASP 24-h forecasts (and 0.4 m s−1 for LAPS-375). Some of this is due to the lobe where the model winds are ∼5 m s−1 but the scatterometer reports ∼15 m s−1. Careful examination of some representative cases demonstrated that many of these points are poor quality data, being either near the center and edges of the swath or in some cases rain affected. The directions show little bias. The vertical bands (Figs. 10b and 11b) near the observed directions of 90° and 270° are due to the cross-swath bias in rain-affected scatterometer data (e.g., see Chelton and Freilich 2005). Apparently, the rain flag does not detect the presence of all rain-affected data. It is evident from the histograms that QuikSCAT wind speed observations with values less than 3 m s−1 are not used in the verification. If we additionally exclude modeled wind speeds with magnitudes less than 3 m s−1 from the verification, then the bias in speed for all models is improved by about 0.1 m s−1, or 15%–30%.

An alternative to examining the mean bias is to fit a functional relationship between the model and observed winds. A first-order least squares best fit, forced through the origin is appropriate.2 The results of the regressions (i.e., the slopes, b0) for global GASP 24-h forecasts for the speed s, zonal u, and meridional υ components were
i1520-0434-22-3-613-eq1

A linear function suites our purpose here, although we note that the “fit” is not optimal at the strongest wind speeds (Fig. 10a), where, for example, mean observations of 30 m s−1 are matched by 25 m s−1 for the model distribution and 27 m s−1 for the linear best fit.

As a measure of the goodness of fit, the analog to the usual correlation coefficient, for a line passing through the origin, was calculated (r02), and found to be 0.96, 0.90, and 0.85 for the speed, and the u and υ components, respectively, confirming the strength of the linear relationship. For the LAPS-375 24-h forecasts, regression values are
i1520-0434-22-3-613-eq2
with r02 values of 0.96, 0.91, and 0.85, respectively. Thus, it appears that the model winds are roughly 5%–10% lighter than those of the scatterometer, with LAPS-375 showing better agreement than GASP, and with the bias being substantially more marked in the meridional component than the zonal. Note that winds containing this bias can have a significant impact when used to force wave models (Greenslade et al. 2005). Over the 2-yr period, LAPS-375 performs better than GASP (global and Australian regions), for the zonal component (b0 ∼ 0.9), but markedly worse in the meridional component where b0 ∼ 0.8. LAPS-375 is expected to perform better as it has higher vertical resolution with the lowest level at 10 m (as opposed to approximately 70 m for GASP), and also uses more sophisticated boundary layer physics. The 12-h forecast generally shows better agreement with the scatterometer winds than does the 24-h forecast.

Once again, we reiterate that QuikSCAT data are collected over a 6-h window, while LAPS is an instantaneous snapshot, and GASP a 6-h mean wind field. For the Australian region QuikSCAT has ascending passes between 2000 and 0000 UTC, and descending passes between 0800 and 1200 UTC. Each pass occurs over approximately 25 min, and there is a total of six to eight passes occurring every 24 h. This mismatch in temporal scales between the observations and the global and local area models may cause some degradation of the verification, although it may have a larger effect on the instantaneous LAPS wind fields.

The biases are also presented as a function of the observed wind speed, for GASP (global domain) for the 24-h forecast (Fig. 12). Figure 12 displays the mean, median, and 1st, 25th, 75th and 99th percentiles of the model minus observation values for each 2 m s−1 observation wind speed bin, with the first bin spanning 2–4 m s−1. Statistics calculated in the 7–14 m s−1 observed wind speed range tend to have 1–5 million observations per bin, one to two orders of magnitude more data than that contained in the bins above about 25 m s−1. There is a linear trend in the bias for observed wind speeds of 2–23 m s−1, with an overestimation of around 2 m s−1 at the lowest speeds, a small bias in the range 6–12 m s−1, and progressively larger underestimation as wind speed increases. For observed speeds greater than about 23 m s−1, the bias increases rapidly. The other models, domains, and forecast periods exhibit similar behavior. This trend is consistent with that found by Yuan (2004) for ECMWF surface winds in the Southern Ocean during 2000, although the GASP bias is somewhat larger.

Seasonal variation in the verification statistics is examined (Figs. 13 –16) by calculating monthly averages from the 2-yr period for all models and domains. Statistics displayed are wind speed bias μ, and slope of best fit b0, for the wind speed, and the zonal and meridional components. A seasonal dependence of the agreement between the model and scatterometer winds can be observed based on the 2-yr dataset, particularly in Meso-LAPS. All models display better agreement with the scatterometer during the austral winter where b0 may increase by 0.2–0.5. The seasonal variability in the bias is possibly related to the increased prevalence of observed scatterometer easterly winds compared with westerlies in the verification domain during the austral summer (Fig. 17). This effect is present in all three study domains (global, Australian region LAPS-375, and Australian region Meso-LAPS) and shows increased seasonal amplitude and noise as the domain decreases in size.

3) Comparison between september 2001 and january 2002

The verification described above was repeated for the months of September 2001 and January 2002 to evaluate the impact of changing the Charnock coefficient in LAPS-375 from 0.018 to 0.011. In this section we focus primarily on the LAPS-375 18- and 24-h forecasts, as some interesting features became evident for these forecast periods.

Verification statistics are presented in Table 5. There is a slight (∼0.8 m s−1) negative bias of the model wind speed relative to the scatterometer-derived wind during September 2001 and January 2002. The directions show little bias, apart from a consistent tendency for the model’s east to southeast winds to be about 10° more easterly than the observations (not shown). The LAPS-375 domain is dominated by the trade winds, and presumably the bias can be attributed to this regime. It appears that the model winds during these time periods are roughly 10% lighter than those of the scatterometer, with the bias being substantially more marked in the meridional component than in the zonal. Results for the 24-h forecast show similar performance. There appear to be no significant changes between the September 2001 and January 2002 verification statistics beyond what can be attributed to seasonal variations; thus, the small reduction in the Charnock coefficient made in December 2001 (from 0.0185 to 0.011) does not seem to have had a major effect on the forecast wind speeds. This is consistent with the study using Rossby number similarity theory in section 3c.

The LAPS-375 slope parameter b0 is plotted as a function of forecast period (out to 48 h) in Fig. 18, for September 2001 and January 2002, and also for GASP (on the LAPS-375 and global domains, out to 120 h) and Meso-LAPS (out to 36 h). GASP and Meso-LAPS show a somewhat better performance in the Australian region in September than January, which is most likely a seasonal bias, and overall are closer to unity than those from LAPS-375. This difference is barely significant except for the meridional component. Differences between the verification statistics for LAPS-375 and Meso-LAPS (which utilize common model physics) can be attributed to the different boundary forcings where Meso-LAPS is nested within LAPS-375, while LAPS-375 is nested within GASP. A gradual decrease in the slope parameter with forecast period is consistent with a decline in model accuracy with time.

Verification of forecast periods that are even multiples of 6 h (e.g., 12- and 24-h forecasts) corresponds to data over roughly the eastern half of the Australian domain, and the odd multiples (e.g., 6- and 18-h forecasts) to the western half.3 The slope parameters display a marked 12-h oscillation in speed and direction for all models, although they are much weaker in the global domain, suggesting a regional difference in forecast skill. An extensive series of calculations was undertaken to try and find a geographical reason for the variation. These showed that the oscillation in the zonal component tends to be more marked at high latitudes, while the meridional component oscillates more strongly in the Tropics, but the underlying reason for this behavior is still not understood. Yuan (2004) noted large regional biases between ECMWF and QuikSCAT wind speeds, particularly in the southwest Indian Ocean, south of Australia, and the southwest Atlantic, consistent with findings here where a larger (smaller) bias in NWP corresponds to verification performed in the Indian (Pacific) Ocean.

6. A case study: The Sydney–Southport storm of July 2001

In late July of 2001 an intense midlatitude cyclone developed off the New South Wales (NSW) coast and produced strong southerly gales in the coastal waters. Such developments in the eastern part of the Tasman Sea are reasonably common, and known locally as east coast lows. They have dynamical similarities with the “bombs” of Sanders and Gyakum (1980) and the “Presidents’ Day storm” (Bosart and Lin 1984), and have been studied by Holland et al. (1987), Leslie et al. (1987), and Hopkins (1997), for example. This particular example coincided with a yacht race from Sydney to Southport (in southern Queensland) and thus produced a significant amount of public interest. It was studied by Spark et al. (2002), who found that while the LAPS-375 system produced good forecast guidance, the model winds at the σ = 0.9875 level (about 100 m above the surface) compared better to the winds experienced by the fleet, than the 10-m winds. Here, we shall compare the model forecasts with QuikSCAT data as a further evaluation of the NWP winds.

Figure 19 shows the observed QuikSCAT wind speeds and LAPS-375 MSLP analyses over the Tasman Sea every 12 h, starting from 0500 UTC 27 July, as well as the 18-h LAPS-375 forecasts of MSLP and wind speed valid at the same times. The low formed initially in a broad trough with its axis near 150°E on 26 July. After drifting over the Tasman Sea, it deepened rapidly in situ, with its central pressure dropping 16 hPa in 24 h according to the LAPS-375 analyses. It then drifted slowly eastward without weakening significantly.

The model 10-m wind speeds are compared with all available rain-flagged quality controlled scatterometer data. It can be seen that there is very good agreement in the depiction of the bands of strong winds throughout the period. However, there is a tendency for the modeled winds to be about 5 m s−1 lighter than those from the scatterometer at 1700 UTC 17 July, with a lesser bias at the adjacent times. This bias is significantly larger than that found in the earlier comparisons in this paper. Observations where the retrieved probability of rain exceeded 0.5 were not plotted, so the bias here is unlikely to be due to rain contamination of the scatterometer, which for this instrument in this wind speed range tends to produce a positive bias in the retrieved wind speed. Rather, it is clear from a comparison of the forecast and analyzed MSLP fields that the forecast has substantially misrepresented both the structure and the intensity of the low during its development, and that this error is consistent with the wind speed error.

Comparison of the observed wind directions (not shown here) confirms that there were some deficiencies in the structure of the modeled low. For instance, at 0500 UTC 28 July, the erroneously light winds north of 30°S were depicted as westerly in LAPS-375, while the scatterometer had them as south-southwesterly. Thus, the wind speed forecast error was largely due to a problem with the modeled surface pressure pattern.

Further into the period, by 0500 UTC 28 July, the forecast and analysis of MSLP agreed within 2 hPa on the central pressure, had very similar gradients (as the ridge to the southeast was also a little stronger in the forecast), and had surface winds that showed a much reduced bias relative to the scatterometer. The comparison between scatterometer and model wind, and model and analyzed MSLP, then continued to improve further as the low drifts east.

Testing the accuracy of the system’s depiction of the MSLP field against observations is difficult, because of an absence of marine observations near the NSW coast. However, comparisons between the forecast and analyzed pressure differences and between various locations [Sydney (33.86°S, 151.21°E) and Lord Howe Island (31.54°S, 159.08°E), a buoy near 37.0°S, 157.0°E, and Green Cape on the NSW south coast (37.26°S, 150.05°E) (positions marked in Fig. 19)] yields some modest further insight (Fig. 20). The pressure differences between Sydney and the two stations in the Tasman Sea are generally very well handled in both the analysis and forecast, although these two stations are too far east to provide a reliable measurement of the near-coastal pressure gradient. The Sydney − Green Cape difference is much less well depicted. While the forecast valid at 1700 UTC 27 July has already been noted as being particularly poor, there are also some problems apparent early in the period. Examination of the pressure series at the individual stations (also in Fig. 20) reveals that all stations tended to measure a lower pressure compared to the analysis during the period.

In summary, early in the period, the substantial differences between forecast winds and the scatterometer observations contain a large signal due to errors in the depiction of the MSLP field. In particular, the period of greatest negative bias of the model marine winds, relative to the scatterometer, coincides with a particularly poor MSLP forecast, with the near-coastal southerly gradient substantially underpredicted. By 28 July, the MSLP field was significantly better depicted, and only a small bias in the model winds remains, consistent with the bias of roughly 5% already discussed. Qi et al. (2006) noted that the quality of the High Resolution NWP model (HIRES) MSLP field representation for a subtropical cyclone in the Tasman Sea was improved by the inclusion of scatterometer winds in the initial state. Spark et al. (2002) noted that the 100-m model winds corresponded most closely to the observed surface winds with the implication that this could be useful operationally as a proxy for the 10-m wind. We recommend some caution in using this method because of the wind error in this case having multiple causes, and suggest that a correction of the order of 5%, as already discussed, is a preferable approach.

Weather system intensity depends on all aspects of the modeling system, including the data assimilation, dynamics, and physical parameterizations. Given the variation in forecast quality through this particular event, it is probably reasonable to infer that deficiencies in the initial condition are the most likely explanation for the less satisfactory performance in the early forecasts. Verifying this is, however, beyond the scope of this paper.

7. Summary

A method for verifying NWP marine surface winds with QuikSCAT satellite winds is presented. The forecast and analysis winds from the Bureau of Meteorology’s current operational LAPS-375, Meso-LAPS, and GASP models are first compared with each other (GASP vs LAPS-375) and then with in situ winds (GASP only) before applying the new verification technique to all models. In all cases, the operational 10-m wind field, which is obtained by a physically based interpolation from the lowest model grid level (Hess et al. 1995), was used for the comparison.

The usefulness of the verification technique, when coupled with an extensive high quality dataset like that provided by QuikSCAT has been demonstrated. The system has been implemented at the Australian Bureau of Meteorology as part of the operational verification suite and provides estimates of the performance of NWP marine surface wind analysis and forecasts. The quality of the NWP surface marine winds is of particular relevance to ocean and wave models, which are critically dependent on the accuracy of the atmospheric forcing.

All the models examined tend to underestimate marine near-surface winds. A 2-yr 2003–04 verification shows that GASP has a seasonally varying bias of −0.4 to −0.8 m s−1, and an underestimation of 5%–10%. LAPS-375 displays a bias of −0.2 to −0.5 m s−1, and an underestimation of around 5%, where deficiencies in the meridional component of LAPS-375 seem to be countered by good agreement in the zonal component. The higher-resolution Meso-LAPS displays slightly larger variability in wind speed bias and b0 than does LAPS-375, averaging around −0.4 m s−1, with much greater interseasonal variation. This may be due to the LAPS-375 domain extending farther into the Tropics where there is less synoptic-scale variability. In these cases the correlation parameter (r20) does not vary significantly from one model to the next, suggesting that much of the difference in performance is due to a systematic bias, rather than “random” errors in the forecast. The bias becomes larger for all models as the observed wind speed increases. Wind verification of LAPS-375 performed between September 2001 and January 2002 shows slightly larger bias than that present in the more extensive 2-yr study, although within the expected noise due to seasonal variation and model changes. Differences between the two 1-month periods are likely to be due to seasonal variability rather than any change in the Charnock coefficient.

This study supports the regional model’s (LAPS-375) reputation for underestimating marine surface winds. However, it should be pointed out that these errors, while significant, are not overwhelming. For instance in the presence of gale force winds, the errors are less than 2.5 m s−1 (∼5 kt). As such, their major impact could be argued to be in the forcing of oceanographic models, rather than in their direct use by forecasters.

A case study of an intense Australian east coast cyclone suggests that the source of the bias in that case is more related to the model underpredicting the intensity of intense systems, than to deficiencies in the depiction of the boundary layer structure, although the above systematic bias also applies. Persistent wind bias due to deficiencies in the NWP boundary layer can be corrected based on the wind verification method, but occasional large errors due to the model depiction of synoptic features require a different solution. The practice has developed among forecasters of using the winds at approximately 100 m as a proxy for the perceived deficiencies in the 10-m wind. A better approach would be to increase the wind speeds by approximately 5%–10%, consistent with the comparison with the very large volume of scatterometer data. This approach must be coupled with the ongoing monitoring of the verification results as the bias is expected to decrease as improvements in the model and data assimilation yield improvements in performance.

This work could be extended through the use of a greater number of in situ observations for the wind verification. In addition there is ongoing work to extend the surface wind verification to evaluations of wind stress curl and divergence. Further improvements to the NWP marine surface wind forecast will be achieved through the assimilation of scatterometer data (see Greenslade et al. 2005), additional model levels in the boundary layer, adopting the ECMWF boundary layer scheme in GASP, the implementation of NWP–wave model coupling to allow parameterization of the surface roughness as a function of sea state, and more frequently updated and higher-resolution SST fields. The ongoing availability of global satellite coverage for surface marine winds is essential for continued monitoring and the quantification of future improvements in the NWP systems.

Acknowledgments

The authors appreciate the input of Dale Hess, Noel Davidson, and Gary Brassington, and the helpful suggestions of reviewers. We thank one reviewer in particular for a very thorough review that led to significant improvements in the manuscript. The work presented here is based in part on a BMRC research report (Kepert et al. 2005).

REFERENCES

  • Alves, J. H. G. M., , Greenslade D. J. M. , , and Banner M. L. , 2002: Impact of a saturation-dependent dissipation source function on operational hindcasts of wind-waves in the Australian region. Global Atmos. Oceanic Syst., 8 , 239267.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Atlas, R., and Coauthors, 2001: The effects of marine winds from scatterometer data on weather analysis and forecasting. Bull. Amer. Meteor. Soc., 82 , 19651990.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Beljaars, A. C. M., 1995: The parametrization of surface fluxes in large-scale models under free convection. Quart. J. Roy. Meteor. Soc., 121 , 255270.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Beljaars, A. C. M., , and Betts A. K. , 1992: Validation of the boundary layer representation in the ECMWF model. Proc. Seminar on Validation of Models over Europe, Vol. II, European Centre for Medium-Range Weather Forecasting, Shinfield Park, United Kingdom, 159–195.

  • Bender, L. C., 1996: Modification of the physics and numerics in a third-generation ocean wave model. J. Atmos. Oceanic Technol., 13 , 726750.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bosart, L. F., , and Lin S. C. , 1984: Diagnostic analysis of the Presidents’ Day storm of February 1979. Mon. Wea. Rev., 112 , 21482177.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Buehner, M., 2002: Assimilation of ERS-2 scatterometer winds using the Canadian 3D-var. Atmos.–Ocean, 40 , 361376.

  • CERSAT, 1999: Off-line wind scatterometer ERS products—User manual. C2-MUT-W-01-IF, IFREMER, Centre de Brest, 86 pp.

  • Charnock, H., 1955: Wind stress on a water surface. Quart. J. Roy. Meteor. Soc., 81 , 639640.

  • Chelton, D. B., , and Freilich M. H. , 2005: Scatterometer-based assessment of 10-m wind analysis from the operational ECMWF and NCEP numerical weather prediction models. Mon. Wea. Rev., 133 , 409429.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., , Schlax M. G. , , Freilich M. H. , , and Milliff R. F. , 2004: Satellite radar measurements reveal short-scale features in the wind stress field over the World Ocean. Science, 303 , 978983.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cohn, S. E., , da Silva A. , , Guo J. , , Sienkiewicz M. , , and Lamich D. , 1998: Assessing the effects of data selection with the DAO physical-space statistical analysis system. Mon. Wea. Rev., 126 , 29132926.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • ECMWF, 1999: Integrated forecasting system technical documentation. Part IV: Physical processes. ECMWF, Shinfield Park, United Kingdom, 142 pp.

  • Garratt, J. R., 1992: The Atmospheric Boundary Layer. Cambridge University Press, 316 pp.

  • Greenslade, D. J. M., , Schulz E. W. , , Kepert J. D. , , and Warren G. , 2005: The impact of the assimilation of scatterometer winds on surface wind and wave forecasts. J. Atmos. Ocean Sci., 10 , 261287.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hess, G. D., , and Garratt J. R. , 2002: Evaluating models of the neutral, barotropic planetary boundary layer using integral measures. Part I. Overview. Bound.-Layer Meteor., 104 , 333358.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hess, G. D., , Colman R. A. , , and McAvaney B. J. , 1995: On computing screen temperatures, humidities and anemometer-height winds in large-scale models. Aust. Meteor. Mag., 44 , 139145.

    • Search Google Scholar
    • Export Citation
  • Holland, G. J., , Lynch A. H. , , and Leslie L. M. , 1987: Australian east-coast cyclones. Part I: Synoptic overview and case study. Mon. Wea. Rev., 115 , 30243036.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hopkins, L. J., 1997: Australian east-coast cyclones and potential vorticity interactions Ph.D. thesis, Monash University, 170 pp.

  • Isaksen, L., , and Janssen P. A. E. M. , 2004: Impact of ERS scatterometer winds in ECMWF’s assimilation system. Quart. J. Roy. Meteor. Soc., 130 , 17931814.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • JPL, 2001: QuikSCAT science data product user’s manual. Version 2.1, Jet Propulsion Laboratory, California Institute of Technology, 86 pp.

  • Kaimal, J. C., , and Finnigan J. J. , 1994: Atmospheric Boundary Flows: Their Structure and Measurement. Oxford University Press, 289 pp.

  • Kepert, J. D., , Greenslade D. J. M. , , and Hess G. D. , 2005: Assessing and improving the marine surface winds in the Bureau of Meteorology numerical weather prediction systems. BMRC Research Rep. 105, Bureau of Meteorology, 61 pp.

  • Kitaigorodskii, S. A., , and Volkov Y. A. , 1965: On the roughness parameter of the sea surface and the calculation of momentum flux in the near-water layer of the atmosphere. Akad. Nauk SSR, Izv. Atmos. Oceanic Phys., 1 , 566574.

    • Search Google Scholar
    • Export Citation
  • Large, W. G., , Morzel J. , , and Crawford G. B. , 1995: Accounting for surface wave distortion of the marine wind profile in low-level ocean storms wind measurements. J. Phys. Oceanogr., 25 , 29592971.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Leidner, S. M., , Isaksen L. , , and Hoffman R. N. , 2003: Impact of NSCAT winds on tropical cyclones in the ECMWF 4DVAR assimilation system. Mon. Wea. Rev., 131 , 326.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Leslie, L. M., , Holland G. J. , , and Lynch A. M. , 1987: Australian east-coast cyclones. Part II: Numerical modeling study. Mon. Wea. Rev., 115 , 30373053.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lorenc, A. C., 1981: A global three-dimensional multivariate statistical interpolation scheme. Mon. Wea. Rev., 109 , 701721.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lorenc, A. C., 1986: Analysis methods for numerical weather prediction. Quart. J. Roy. Meteor. Soc., 112 , 11771194.

  • Louis, J-F., 1979: A parametric model of vertical eddy fluxes in the atmosphere. Bound.-Layer Meteor., 17 , 187202.

  • Louis, J. F., , Tiedtke M. , , and Geleyn J. F. , 1982: A short history of the operational PBL parametrization at ECMWF. Proc. Workshop on Boundary Layer Parametrization, ECWMF, Shinfield Park, United Kingdom, 59–79.

  • Lynch, P., , and Huang X-Y. , 1992: Initialization of the HIRLAM model using a digital filter. Mon. Wea. Rev., 120 , 10191034.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • McIntosh, P. C., , and Hubbert G. D. , 1992: Ocean winds for marine modeling. Aust. Meteor. Mag., 40 , 6169.

  • Mears, C. A., , Smith D. K. , , and Wentz F. J. , 2001: Comparison of Special Sensor Microwave Imager and buoy-measured wind speeds from 1987 to 1997. J. Geophys. Res., 106 , 1171911729.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Monaldo, F., 1988: Expected differences between buoy and radar altimeter estimates of wind speed and significant wave height and their implications on buoy–altimeter comparisons. J. Geophys. Res., 93 , 22852302.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Park, K-A., , Cornillon P. , , and Codiga D. L. , 2006: Modification of surface winds near ocean fronts: Effects of Gulf Stream rings on scatterometer (QuikSCAT, NSCAT) wind observations. J. Geophys. Res., 111 .C03021, doi:10.1029/2005JC003016.

    • Search Google Scholar
    • Export Citation
  • Puri, K., , Dietachmayer G. , , Mills G. A. , , Davidson N. E. , , Bowen R. A. , , and Logan L. W. , 1998: The new BMRC Limited Area Prediction System, LAPS. Aust. Meteor. Mag., 47 , 203233.

    • Search Google Scholar
    • Export Citation
  • Qi, L., , Leslie L. M. , , and Speer M. S. , 2006: Climatology of cyclones over the southwest Pacific: 1992–2001. Meteor. Atmos. Phys., 91 , 201209.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sanders, G., , and Gyakum J. R. , 1980: Synoptic–dynamic climatology of the “bomb.”. Mon. Wea. Rev., 108 , 15891606.

  • Seaman, R., , Bourke W. , , Steinle P. , , Hart T. , , Embery G. , , Naughton M. , , and Rikus L. , 1995: Evolution of the Bureau of Meteorology’s Global Assimilation and Prediction System, Part 1: Analyses and initialization. Aust. Meteor. Mag., 44 , 118.

    • Search Google Scholar
    • Export Citation
  • Spark, E., , Batt K. , , Gage B. , , and Zmijewski P. , 2002: Forecasts for the Sydney to Southport yacht race. Economic benefits of meteorological forecasts—A case study. Abstracts, Ninth Australian Meteorological and Oceanographic Society Annual Conf., Melbourne, Victoria, Australia, Melbourne University, 66.

  • Steinle, P. J., 2005: Generalized statistical interpolation. Fourth Symp. on Assimilation of Observations for Meteorology and Oceanography, Prague, Czech Republic, WMO, CD-ROM.

  • Stoffelen, A., 1998: Toward the true near-surface wind speed: Error modeling and calibration using triple collocation. J. Geophys. Res., 103 , 77557766.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., , and Anderson D. , 1997: Scatterometer data interpretation: Measurement space and inversion. J. Atmos. Oceanic Technol., 14 , 12981313.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tang, Y. M., , Smith N. , , and Greenslade D. J. M. , 2000: Comparison of model and observed surface winds. Aust. Meteor. Mag., 49 , 2335.

  • Troen, I., , and Mahrt L. , 1986: A simple model of the atmospheric boundary layer: Sensitivity to surface evaporation. Bound.-Layer Meteor., 37 , 129148.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wilkerson, J. C., , and Earle M. D. , 1990: A study of differences between environmental reports by ships in the voluntary observing program and measurements from buoys. J. Geophys. Res., 95 , 33733385.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Yuan, X., 2004: High-wind-speed evaluation in the Southern Ocean. J. Geophys. Res., 109 .D13101, doi:10.1029/2003JD004179.

Fig. 1.
Fig. 1.

The ratio of 10-m wind speed to the geostrophic wind speed according to Rossby number similarity theory, as a function of (top) Charnock coefficient and (bottom) sea temperature (Ts) and air temperature (Ta) difference, for geostrophic wind speeds of 5 (solid), 10 (dashed), and 20 m s−1 (dotted). Other parameters are (top) Coriolis ( f ) = 10−4 s−1 and TsTa = 0, and (bottom) αC = 0.0185. (top) Charnock values of 0.011 and 0.018 are indicated.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 2.
Fig. 2.

Location of buoys for which 10-m wind velocity observations are available.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 3.
Fig. 3.

Time series of buoy 10-m wind observations (solid line), GASP-analyzed 10-m wind speed (dashed), and GASP 24-h forecast 10-m winds (dotted) during September 2001.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 4.
Fig. 4.

As in Fig. 3 but during January 2002.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 5.
Fig. 5.

Scatterplots of the GASP (analysis and 24-h forecast) vs buoy-mounted anemometer 10-m wind speeds for September 2001 and January 2002, shown in Figs. 3 and 4.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 6.
Fig. 6.

(a) GASP U10, (b) LAPS-375 U10, and (c) the difference between the two wind fields (GASP − LAPS-375) for analysis forecasts valid at 0000 UTC 22 Sep 2001. Areas where GASP > LAPS-375 are shaded.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 7.
Fig. 7.

Comparison between LAPS-375 and GASP operational 10-m wind speeds, for analyses and 24-h forecasts, at 12-hourly intervals for (a), (b) September 2001 and (c), (d) January 2002. Statistics are displayed in Table 3. Contours are at 1000, 2000, 4000, 6000, and 8000 observations.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 8.
Fig. 8.

Comparison between LAPS-375 and GASP analyses valid at 1200 UTC 17 Sep 2001.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 9.
Fig. 9.

(top) Scatterometer wind speed (m s−1) and analyzed LAPS-375 MSLP (hPa, contoured every 4 hPa), (middle) LAPS-375 18-h forecasts of wind speed and MSLP, all valid at 1100 UTC 18 Jan 2002, and (bottom) wind speed difference.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 10.
Fig. 10.

Contoured 2D histograms for GASP over the Australian domain 24-h forecast (y axis) against scatterometer winds during 2003 and 2004 for (a) wind speed, (b) wind direction, and (c) zonal and (d) meridional components. The line y = x (light) and the line of best fit described in the text (heavy) are also shown. Contours are in geometric progression, every 101/2. The mean and standard deviation of the model − observation difference, together with the slope and correlation parameter, are described in the text and Table 4.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 11.
Fig. 11.

As in Fig. 10 but for LAPS-375.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 12.
Fig. 12.

The 24-h forecast − observed wind speed vs observed wind speed (in 2 m s−1 bins) for the GASP global domain. Statistics are mean (thick bar), median (thin bar), 1st and 3d quartiles (bottom and top of box), and 1st and 99th percentiles (bottom and top whisker).

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 13.
Fig. 13.

Monthly verification statistics averaged over 2003 and 2004 for GASP (global domain) − QuikSCAT (a) wind speed bias, (b) slope of linear best fit through zero for speed, and (c) zonal and (d) meridional wind components.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 14.
Fig. 14.

As in Fig. 13 but for GASP (Australian domain) − QuikSCAT.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 15.
Fig. 15.

As in Fig. 13 but for LAPS-375 (Australian domain) − QuikSCAT.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 16.
Fig. 16.

As in Fig. 13 but for Meso-LAPS (Australian domain) − QuikSCAT.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 17.
Fig. 17.

Mean monthly QuikSCAT-observed zonal winds for 2003 and 2004. Calculated for GASP global (solid), LAPS-375 (dashed), and Meso-LAPS (dotted) domains.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 18.
Fig. 18.

The slope b0 of the line of best fit through the origin of model against scatterometer winds, as a function of forecast period (days). Models analyzed are GASP global domain (asterisks), GASP Australian region (circles), LAPS-375 Australian region (triangles), and Meso-LAPS Australian region (diamonds). Statistics are for (a), (d) wind speed, and (b), (e) zonal and (c), (f) meridional wind components for (left) September 2001 and (right) January 2002.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 19.
Fig. 19.

Comparison of LAPS-375 18-h forecast winds and MSLP with scatterometer observations and analyzed MSLP for the Sydney–Southport storm on (top) 27, (middle) 28, and (bottom) 29 July. Columns from left to right are scatterometer observations and analyzed MSLP at 0500 UTC, LAPS-375 18-h forecast wind and MSLP forecast at 0500 UTC, scatterometer observations and analyzed MSLP at 1700 UTC, and LAPS-375 18-h forecast wind and MSLP forecast at 1700 UTC. Locations of MSLP observations are indicated for Sydney (S); Lord Howe Island (LH); buoy near 37°S, 157°E (B); and Green Cape (GC). Note that model winds are only plotted at scatterometer observation locations.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Fig. 20.
Fig. 20.

(left) Pressure at Sydney (top), Lord Howe Island (second from top), Green Cape (third from top), and the buoy (bottom). (top right) Pressure differences between Sydney and Lord Howe Island; (middle right) Sydney and buoy near 37°S, 157°E; and (bottom right) Sydney and Green Cape, as a function of the day of July 2001: heavy line, observations; crosses, analysis; and circles, 18-h forecast.

Citation: Weather and Forecasting 22, 3; 10.1175/WAF996.1

Table 1.

Information on comparisons and verifications. “Snapshot” indicates that the verification is performed on a single forecast or analysis.

Table 1.
Table 2.

Monthly statistics for GASP (analysis and 24-h forecast) − observed buoy 10-m wind observations for September 2001 and January 2002, for data displayed in Figs. 3 –5: systematic bias (bias), linear correlation coefficient (R), root-mean-square error (rms), and scatter index (SI; standard deviation of error as a proportion of the mean observed value).

Table 2.
Table 3.

Statistics for GASP − LAPS-375 difference for the months of September 2001 and January 2002, as displayed in Fig. 7.

Table 3.
Table 4.

NWP wind verification statistics for the 2-yr period 2003 and 2004, relating to Figs. 10 and 11, for 12- and 24-h forecasts for various models and domains. See text for full description.

Table 4.
Table 5.

LAPS-375 NWP wind verification statistics for September 2001 and January 2002, for 18- and 24-h forecasts. See text for full description.

Table 5.
1

Operational assimilation of QuikSCAT winds commenced March 2005 for LAPS-375. GASP commenced nonoperational assimilation in 2001. Only operational products are verified in this paper.

2

The method regresses the model onto scatterometer winds, which assumes no error in the observations. Double regression, which assumes equal model and observation error, consistently yields a 0.04 increase in b0. A model error twice the magnitude of the observation error yields increases in b0 of approximately 0.02.

3

QuikSCAT is in a sun-synchronous orbit, crossing the equator at about 0600 and 1800 local standard time.

Save