• Aikman, F., , Mellor G. L. , , Rao D. B. , , and Waters M. P. , 1994: A feasibility study of a coastal nowcast/forecast system. Abstracts, Spring Meeting of the American Geophysical Union, Baltimore, MD, AGU, 197.

    • Search Google Scholar
    • Export Citation
  • Aikman, F., , Sheinin D. , , Chen P. , , Breaker L. , , Bosley K. , , and Rao D. B. , 1996: Towards an operational nowcast/forecast system for the U.S. East Coast. Modern Approaches to Data Assimilation in Ocean Modeling. Vol. 61, P. Malanotte-Rizzoli, Ed., Elsevier, 347–376.

    • Search Google Scholar
    • Export Citation
  • Aikman, F., , Wei E. J. , , and Schultz J. R. , 1998: Water level evaluation for the Coastal Ocean Forecast System. Preprints, Second Conf. on Coastal Atmospheric and Oceanic Prediction and Processes, Phoenix, AZ, Amer. Meteor. Soc., 1–3.

    • Search Google Scholar
    • Export Citation
  • Behringer, D. W., 1994: Sea surface height variations in the Atlantic Ocean: A comparison of TOPEX altimeter data with results from an ocean data assimilation system. J. Geophys. Res., 99 , 2468524690.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Behringer, D. W., , Ming J. , , and Leetma A. , 1998: An improved coupled model for ENSO prediction and implications for ocean initialization. Part I: The ocean data assimilation system. Mon. Wea. Rev., 126 , 10131021.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bell, M. J., , Forbes R. M. , , and Hines A. , 2000: Assessment of the FOAM global data assimilation system for real-time operational ocean forecasting. J. Mar. Syst., 25 , 122.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Black, T., 1994: The new NMC Mesoscale Eta Model: Description and forecast examples. Wea. Forecasting, 9 , 265278.

  • Blumberg, A. F., , and Grehl B. J. , 1987: A river climatology for the United States Atlantic coast. HydroQual, Inc., Mahwah, NJ, 6 pp. [Available from HydroQual, Inc., 1 Lethbridge Plaza, Mahwah, NJ 07430.].

    • Search Google Scholar
    • Export Citation
  • Breaker, L. C., , and Rao D. B. , 1998: Experience gained during the implementation of NOAA Coastal Ocean Forecast System. Proc. The Marine Technology Society Annual Conf., Baltimore, MD, The Marine Technology Society, 235–241.

    • Search Google Scholar
    • Export Citation
  • Chalikov, D., , Breaker L. , , and Łobocki L. , 1996: Parameterization of mixing in the upper ocean. National Centers for Environmental Prediction Tech. Note, 40 pp. [Available from National Weather Service, NCEP, Environmental Modeling Center, World Weather Building, 5200 Auth Road, Camp Springs, MD 20746.].

    • Search Google Scholar
    • Export Citation
  • Chen, P., , and Mellor G. L. , 1999: Determination of tidal boundary forcing using tide station data. Coastal Ocean Prediction, C. N. K. Mooers, Ed., Amer. Geophys. Union, 329–351.

    • Search Google Scholar
    • Export Citation
  • Dengg, J. A., , Beckmann A. , , and Gerdes R. , 1996: The Gulf Stream separation problem. The Warm Water Sphere of the North Atlantic Ocean, W. Krauss, Ed., Gebrader Borntreger, 253–290.

    • Search Google Scholar
    • Export Citation
  • Derber, J., , and Rosati A. , 1989: A global oceanic data assimilation system. J. Phys. Oceanogr., 19 , 13331347.

  • Ezer, T., , Ko D. S. , , and Mellor G. L. , 1992: Modeling and forecasting the Gulf Stream. Oceanic and Atmospheric Nowcasting and Forecasting, Vol. 26(2), D. L. Durham and J. W. Lewis, Eds., Marine Technology Society, 5–14.

    • Search Google Scholar
    • Export Citation
  • Forbes, R. M., 1995: Experiments with the assimilation of surface temperature and thermal profile observations into a dynamical model of the Atlantic Ocean. Forecasting Research Tech. Report 167, Forecasting Research Division, United Kingdom Meteorological Office, 31 pp.

    • Search Google Scholar
    • Export Citation
  • Fox, D. N., , Carnes M. R. , , and Mitchell J. L. , 1992: Characterizing major frontal systems: A nowcast/forecasts system for the northwest Atlantic. Oceanography, 5 , 4954.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Galperin, B., , Kantha L. H. , , Hassid S. , , and Rosati A. , 1988: A quasi-equilibrium turbulent energy model for geophysical flows. J. Atmos. Sci., 45 , 5562.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gilhousen, D. B., 1987: A field evaluation of NDBC moored buoy winds. J. Atmos. Oceanic Technol., 4 , 94104.

  • Gilhousen, D. B., . 1998: Improved real-time quality control of NDBC measurements. Preprints, 10th Symp. on Meteorological Observations and Instrumentation, Phoenix, AZ, Amer. Meteor. Soc., 363–366.

    • Search Google Scholar
    • Export Citation
  • Gill, P. E., , Murray W. , , and Wright M. H. , 1981: Practical Optimization. Academic Press, 401 pp.

  • Haltiner, G. J., , and Williams R. T. , 1980: Numerical Prediction and Dynamic Meteorology. John Wiley and Sons, 477 pp.

  • Ji, M., , Leetmaa A. , , and Derber J. , 1995: An ocean analysis system for seasonal to interannual climate studies. Mon. Wea. Rev., 123 , 460481.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kelley, J. G. W., 1998: Generation of three-dimensional lake model forecasts for Lake Erie. Wea. Forecasting, 13 , 659687.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kelley, J. G. W., , Behringer D. W. , , and Thiebaux H. J. , 1999: Description of the SST Data Assimilation System used in the NOAA Coastal Ocean Forecast System (COFS) for the U.S. East Coast. National Centers for Environmental Prediction Tech. Note, 49 pp. [Available from National Weather Service, NCEP, Environmental Modeling Center, World Weather Building, 5200 Auth Road, Camp Springs, MD 20746.].

    • Search Google Scholar
    • Export Citation
  • Koutitonsky, V. G., , and Bugden G. L. , 1991: The physical oceanography of the Gulf of St. Lawrence: A review with emphasis on the synoptic variability of the motion. The Gulf of St. Lawrence: Small Ocean or Large Estuary. Vol. 113, J.-C. Therriault, Ed., Canadian Dept. of Fisheries and Oceans, 57–90.

    • Search Google Scholar
    • Export Citation
  • Kuan, C-F., 1995: Performance evaluation of the Princeton Circulation Model for Lake Erie. Ph.D. dissertation, The Ohio State University, 376 pp. [Available from Dept. of Civil and Environmental Engineering and Geodetic Science, The Ohio State University, 470 Hitchcock Hall, 2070 Neil Avenue, Columbus, OH, 43210-1275.].

    • Search Google Scholar
    • Export Citation
  • Leaman, K. D., , Molinari R. L. , , and Vertes P. S. , 1987: Structure and variability of the Florida Current at 27°N: April 1982–July 1984. J. Phys. Oceanogr., 17 , 565583.

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

  • Mellor, G. L., 1996: Users guide for a three-dimensional, primitive equation, numerical ocean model. Program in Atmospheric and Oceanic Sciences, Princeton University, 39 pp. [Available from Program in Atmospheric and Oceanic Sciences, Princeton University, P.O. Box CN710, Sayre Hall, Princeton, NJ 08544-0710.].

    • Search Google Scholar
    • Export Citation
  • Mellor, G. L., , and Yamada T. , 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys. Space Phys, 20 , 851875.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Navon, I. M., , and Legler D. M. , 1987: Conjugate-gradient methods for large-scale minimization in meteorology. Mon. Wea. Rev., 115 , 14791502.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • NOS, 1999: NOS procedures for developing and implementing operational nowcast and forecast systems for PORTS. NOAA Tech. Rep. NOS CO-OPS 0020, Center for Operational Oceanographic Products and Services, 33 pp. [Available from NOS/CO-OPS, National PORTS Program Manager, SSMC4, 1305 East-West Highway, Silver Spring, MD 20910-3281.].

    • Search Google Scholar
    • Export Citation
  • Pichel, W. G., , Clemente-Colon P. , , Sapper J. F. , , Duda C. , , and Krasnopolsky V. M. , 1995: Validation and quality control of CoastWatch satellite-derived water-surface-temperature imagery. Proc. Third Thematic Conf. on Remote Sensing for Marine and Coastal Environments, Seattle, WA, Environmental Research Institute of Michigan, I-83–I-91.

    • Search Google Scholar
    • Export Citation
  • Richardson, P. L., 1985: Average velocity and transport of the Gulf Stream near 55°W. J. Mar. Res., 43 , 83111.

  • Robinson, A. R., 1992: Shipboard prediction with a regional forecast model. Oceanography, 5 , 4248.

  • Rogers, E., and Coauthors. 1998: Changes to the NCEP Operational “Early” Eta Analysis/Forecast System. NWS Tech. Procedures Bulletin No. 447, National Centers for Environmental Prediction, 31 pp. [Available online from http://www.nws.noaa.gov/om/447body.htm.].

    • Search Google Scholar
    • Export Citation
  • Rogers, E., , Parrish D. , , and DiMego G. , 1999: Changes to the NCEP Operational Eta Analysis. NWS Tech. Procedures Bulletin (Draft), National Weather Service, 5 pp. [Available online at http://tgsv5.nws.noaa.gov/om/tpb/3d-eta.html.].

    • Search Google Scholar
    • Export Citation
  • Schmalz, R. A. Jr,, 1998: Design of a nowcast/forecast system for Galveston Bay. Preprints, Second Conf. on Coastal Atmospheric and Oceanic Prediction and Processes, Phoenix, AZ, Amer. Meteor. Soc., 15–22.

    • Search Google Scholar
    • Export Citation
  • Schultz, J. R., , and Aikman F. I. , 1998: Evaluation of subtidal water level in NOAA's Coastal Ocean Forecast System for the U.S. East Coast. Proc., Fifth Int. Conf. on Estuarine and Coastal Modeling, Alexandria, VA, Amer. Soc. Civil Eng., 766–780.

    • Search Google Scholar
    • Export Citation
  • Schwab, D. J., , and Bedford K. W. , 1994: Initial implementation of the Great Lakes Forecasting System: A real-time system for predicting lake circulation and thermal structure. Water Pollut. Res. J. Can., 29 , 203220.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Smagorinsky, J., 1963: General circulation experiments with the primitive equations. I. The basic experiments. Mon. Wea. Rev., 91 , 99164.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Teague, W. J., , Carron M. J. , , and Hogan P. J. , 1990: A comparison between the generalized digital environmental model and Levitus climatologies. J. Geophys. Res., 95 , 71677183.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Thiebaux, J., , Katz B. , , and Wang W. , 2001: New sea-surface temperature analysis implemented at NCEP. Preprints, 18th Conf. on Weather Analysis and Forecasting, Ft. Lauderdale, FL, Amer. Meteor. Soc., J159–J163.

    • Search Google Scholar
    • Export Citation
  • Thompson, J. D., , and Schmitz Jr. W. J. , 1989: A limited-area model of the Gulf Stream: Design, initial experiments, and model–data intercomparison. J. Phys. Oceanogr., 19 , 792814.

    • Search Google Scholar
    • Export Citation
  • Thompson, J. D., , Townsend T. L. , , Wallcraft A. , , and Schmitz Jr. W. J. , 1992: Ocean prediction and the Atlantic Basin: Scientific issues and technical challenges. Oceanography, 5 , 3641.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wei, E. J., 1992: Simulation of three-dimensional hydrodynamics in Long Island Sound: Seasonal timescale. Proc. Second Int. Conf. on Estuarine and Coastal Modeling, Tampa, FL, Amer. Soc. Civil Eng., 430–440.

    • Search Google Scholar
    • Export Citation
  • Wei, E. J., , and Sun C. L. , 1998: Development of a New York/New Jersey Harbor water level and current nowcast/forecast system. Preprints, Second Conf. on Coastal Atmospheric and Oceanic Prediction and Processess, Phoenix, AZ, Amer. Meteor. Soc., 11–13.

    • Search Google Scholar
    • Export Citation
  • Xue, H., , Chai F. , , and Pettigrew N. R. , 2000: A model study of the seasonal circulation in the Gulf of Maine. J. Phys. Oceanogr., 30 , 11111133.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Yen, C-C., , Bedford K. W. , , and Schwab D. J. , 1992: Nowcast protocol for the Great Lakes Forecasting System. Proc. Second Int. Conf. on Estuarine and Coastal Modeling, Tampa, FL, Amer. Soc. Civil Eng., 140–148.

    • Search Google Scholar
    • Export Citation
  • Yen, C-C., , Kelley J. G. W. , , and Bedford K. W. , 1994: Daily procedure for GLFS nowcasts. Proc. Natl. Conf. on Hydraulic Engineering, Buffalo, NY, Amer. Soc. Civil Eng., 202–206.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Cycle configurations of (a) COFS 3.1, the nonassimilating version, and (b) COFS 3.2, the assimilating version

  • View in gallery

    Locations of COFS grid points, contours of model bathymetry, names, and approximate locations of rivers for which outflow is specified, along with the location and amount of transport on the lateral open boundaries

  • View in gallery

    Locations of fixed platforms in the COFS domain whose SST observations were used or withheld from data assimilation. The curved dotted line represents the southern and eastern boundary of the COFS curvilinear grid domain. The 200-m isobath and a 5-yr mean Gulf Stream landward surface edge are also depicted

  • View in gallery

    Example of the sources and locations of SST and MCSST observations available in the COFS domain during a 48-h period. The identifiers for in situ platforms (except drifting buoys) are given. Also shown are the 200-m isobath and the mean Gulf Stream landward surface edge. The NOAA-14 and NOAA-15 MCSST observations are plotted as red and green dots, respectively

  • View in gallery

    Schematic of the mixed-layer assimilation scheme used in COFS 3.2 (a) when the estimated SST (E) is colder than the model SST (M) and (b) when the estimated SST is warmer than the model SST

  • View in gallery

    Locations of observed thermal profiles used in the evaluation of COFS-3.1 and -3.2 subsurface temperature predictions during Feb 1998. The letter next to each profile corresponds to the locations referred to in the text and in Figs. 10–14

  • View in gallery

    Time series plots of COFS-3.1 SST initial conditions, COFS-3.2 SST nowcasts, and in situ observations at (a) Portland, ME, buoy 44007; (b) Chesapeake Bay Light Tower C-MAN station, CHLV2; (c) Southwest Grand Banks buoy 44138; and (d) South Hatteras buoy 41002 from 1 Oct 1997 to 31 Mar 1998. The observations and the COFS predictions have both been averaged over a day

  • View in gallery

    (a) COFS-3.1 SST initial condition, (b) COFS-3.2 SST nowcast, and (c) NESDIS 14-km MCSST analysis valid at 0000 UTC 28 Feb 1998

  • View in gallery

    (Continued)

  • View in gallery

    Difference maps of (a) COFS-3.1 SST initial condition minus the NESDIS 14-km MCSST analysis, and (b) COFS-3.2 SST nowcast minus the same analysis valid at 0000 UTC 28 Feb 1998

  • View in gallery

    Observed thermal profiles (Data) and corresponding COFS-3.1 and -3.2 predictions along a shipping lane to the east of Bermuda valid at (a) 1652 UTC and (b) 2257 UTC 21 Feb and (c) 0451 UTC 22 Feb 1998

  • View in gallery

    Same as in Fig. 10 except that the profiles are along the shipping lane from New York City toward Puerto Rico (NW to SE) valid at (a) 2228 UTC 14 Feb and (b) 0503 UTC, (c) 1028 UTC, (d) 1717 UTC, and (e) 2231 UTC 15 Feb 1998.

  • View in gallery

    Same as in Fig. 10 except that the profiles are along the shipping lane from New York City toward Bermuda (NW to SE) valid 14 Feb 1998 at (a) 0206 UTC, (b) 0302 UTC, (c) 0355 UTC, (d) 0449 UTC, (e) 0605 UTC, (f) 0954 UTC, (g) 1110 UTC, (h) 1303 UTC, (i) 1458 UTC, (j) 1618 UTC, (k) 1701 UTC, (l) 1801 UTC, and (m) 1923 UTC

  • View in gallery

    (Continued)

  • View in gallery

    Same as in Fig. 10 except that the profiles are across the continental slope just to the north of the New York City–to-Bermuda shipping lane valid at (a) 2349 UTC 18 Feb and (b) 0157 UTC and (c) 1248 UTC 19 Feb 1998

  • View in gallery

    Same as in Fig. 10 except that the profiles are along the shipping lanes from Boston to Nova Scotia (E to W) valid at (a) 2327 UTC 21 Feb and (b) 0200 UTC, (c) 0604 UTC, (d) 0758 UTC, and (e) 0952 UTC 22 Feb 1998

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 61 61 2
PDF Downloads 14 14 1

Assimilation of SST Data into a Real-Time Coastal Ocean Forecast System for the U.S. East Coast

View More View Less
  • 1 NOAA/National Ocean Service Coast Survey Development Laboratory, Silver Spring, Maryland
  • | 2 NOAA/National Weather Service/National Centers for Environmental Prediction Environmental Modeling Center, Washington, D.C
© Get Permissions
Full access

Abstract

The real-time, three-dimensional, limited-area Coastal Ocean Forecast System (COFS) has been developed for the northwestern Atlantic Ocean and implemented at the National Centers for Environmental Prediction. COFS generates a daily nowcast and 1-day forecast of water level, temperature, salinity, and currents. Surface forcing is provided by 3-h forecasts from the National Weather Service's Eta Model, a mesoscale atmospheric prediction model. Lateral oceanic boundary conditions are based on climatic data. COFS assimilates in situ sea surface temperature (SST) observations and multichannel satellite SST retrievals for the past 48 h. SST predictions from the assimilating and nonassimilating versions of COFS were compared with independent observations and a 14-km-resolution multichannel SST analysis. The assimilation of SST data reduced the magnitude and the geographic extent of COFS's characteristic positive temperature bias north of the Gulf Stream. The root-mean-square SST differences between the COFS predictions and in situ observations were reduced by up to 47%–50%. Qualitative comparisons were also made between predictions from the assimilating and nonassimilating versions and thermal profiles measured by expendable bathythermographs. These comparisons indicated that the assimilation scheme had positive impact in reducing temperature differences in the top 300 m at most locations. However, the subsurface comparisons also show that, in dynamically complex regions such as the Gulf Stream, the continental slope, or the Gulf of Maine, the data assimilation system has difficulty reproducing the observed ocean thermal structure and would likely benefit from the direct assimilation of observed profiles.

Corresponding author address: Dr. John G. W. Kelley, Coast Survey Development Laboratory, National Ocean Service, N/CS13, SSMC3, Rm 7825, 1315 East-West Highway, Silver Spring, MD 20910-3282. Email: john.kelley@noaa.gov

Abstract

The real-time, three-dimensional, limited-area Coastal Ocean Forecast System (COFS) has been developed for the northwestern Atlantic Ocean and implemented at the National Centers for Environmental Prediction. COFS generates a daily nowcast and 1-day forecast of water level, temperature, salinity, and currents. Surface forcing is provided by 3-h forecasts from the National Weather Service's Eta Model, a mesoscale atmospheric prediction model. Lateral oceanic boundary conditions are based on climatic data. COFS assimilates in situ sea surface temperature (SST) observations and multichannel satellite SST retrievals for the past 48 h. SST predictions from the assimilating and nonassimilating versions of COFS were compared with independent observations and a 14-km-resolution multichannel SST analysis. The assimilation of SST data reduced the magnitude and the geographic extent of COFS's characteristic positive temperature bias north of the Gulf Stream. The root-mean-square SST differences between the COFS predictions and in situ observations were reduced by up to 47%–50%. Qualitative comparisons were also made between predictions from the assimilating and nonassimilating versions and thermal profiles measured by expendable bathythermographs. These comparisons indicated that the assimilation scheme had positive impact in reducing temperature differences in the top 300 m at most locations. However, the subsurface comparisons also show that, in dynamically complex regions such as the Gulf Stream, the continental slope, or the Gulf of Maine, the data assimilation system has difficulty reproducing the observed ocean thermal structure and would likely benefit from the direct assimilation of observed profiles.

Corresponding author address: Dr. John G. W. Kelley, Coast Survey Development Laboratory, National Ocean Service, N/CS13, SSMC3, Rm 7825, 1315 East-West Highway, Silver Spring, MD 20910-3282. Email: john.kelley@noaa.gov

1. Introduction

Since the mid-1980s, several three-dimensional coastal ocean circulation forecast systems have been developed and tested for limited time periods and areas of the northwest Atlantic Ocean (Fox et al. 1992; Robinson 1992). However, these forecast systems do not cover the entire northwest Atlantic Ocean, and some rely in part on non–operationally available data, and others have not been implemented operationally for general use. The first successful attempt at producing real-time 3D nowcasts was done for Lake Erie using the Great Lakes Forecasting System developed by The Ohio State University and the Great Lakes Environmental Research Laboratory of the National Oceanic and Atmospheric Administration (NOAA) in the spring of 1992 (Yen et al. 1992). The National Weather Service's National Centers for Environmental Prediction (NCEP) and the National Ocean Service (NOS) subsequently developed, in collaboration with Princeton University, the real-time, limited-area Coastal Ocean Forecast System (COFS) for the U.S. East Coast. COFS was implemented experimentally at NCEP in August of 1993 (Aikman et al. 1994; Aikman et al. 1996).

From August of 1993 to March of 1998, COFS generated 24-h forecasts once per day for the northwest Atlantic Ocean with surface forcing provided by NCEP's operational eta-coordinate mesoscale atmospheric prediction model without oceanic data assimilation (Fig. 1a). Analyses of errors from these forecasts pointed out a need to modify the Eta Model surface flux parameterizations and also provided valuable knowledge on the sensitivity of COFS to lateral boundary conditions (Breaker and Rao 1998) and the accuracy of its water-level predictions (Schultz and Aikman 1998). The simulations also showed the tendency of COFS to produce a Gulf Stream that “overshoots” near the separation site at Cape Hatteras; that is, it hugs the continental slope and separates near 40°N instead of at 35°N as observed. This tendency, which has occurred in other ocean models, is not fully understood (Thompson and Schmitz 1989; Thompson et al. 1992; Dengg et al. 1996). As expected, COFS was able to predict the statistical properties of features such as the Gulf Stream, but it did not properly depict the location of individual meanders and eddies. The prediction of the evolution and movement of eddies, meanders, frontal positions, and correction of the Gulf Stream separation problem require oceanic data assimilation to provide initial conditions for the COFS forecast cycle.

In designing a data assimilation system for COFS, the emphasis was placed on first assimilating sea surface temperature (SST) data instead of altimeter data and/or thermal profiles from expendable bathythermographs (XBTs). This decision was based on both data and operational issues, including 1) the low density and nonuniform distribution of XBTs relative to in situ and remotely sensed SST data available operationally at NCEP, 2) the success of the Met Office in extrapolating SST information into the mixed layer of a daily ocean model, 3) no operational access to sea surface height anomalies derived from the Ocean Topography Experiment/Poseidon (TOPEX/Poseidon) or European Remote Sensing Satellite-2 (ERS-2) satellite altimeter data, and 4) uncertainty in the appropriate method for translating sea surface height changes into subsurface temperature and salinity changes in a daily, real-time coastal ocean forecast model.

A new, two-cycle configuration of COFS was implemented in April of 1998 and consisted of an SST data assimilation cycle and a 24-h forecast cycle (Fig. 1b).The assimilation cycle incorporates both in situ and remotely sensed SST data to generate a daily nowcast. A nowcast is defined as a model-based prediction that incorporates recently observed oceanographic, meteorological, and/or river flow rate data; that covers the period from the recent past to the present; and that makes a prediction for locations at which data are not available (NOS 1999). The COFS nowcast serves the initial conditions for a daily 24-h forecast and the next day's assimilation cycle.

The purpose of this paper is to present the results of comparisons between COFS temperature predictions with and without SST data assimilation. COFS SST predictions were evaluated for a 3-month period from 1 January to 31 March 1998, and subsurface predictions were assessed for February of 1998. A description of the version of the coastal ocean circulation model used in COFS is given in section 2, which includes discussions of surface and lateral boundary conditions. Discussion of the data and quality-control procedure is given in section 3, and a description of the assimilation scheme is given in section 4. Section 5 discusses the type of COFS available from NCEP. The method of evaluation and the results from the evaluation are described in sections 6 and 7, respectively. A summary and conclusions are presented in section 8.

2. Coastal ocean model

COFS uses the Princeton Ocean Model (POM), a three-dimensional, primitive equation, hydrostatic, coastal ocean prediction model (Mellor 1996). The model explicitly predicts the three-dimensional velocity distribution, temperature, salinity, free surface water elevation, and turbulent quantities. The parameterization of turbulent mixing is accomplished using a simplified version of the level-2.5 turbulence closure scheme (Mellor and Yamada 1982; Galperin et al. 1988). Horizontal diffusion is based on the formulation of Smagorinsky (1963), which takes into account the vertical variability in the horizontal viscosity and diffusivity.

POM has been used in a variety of research studies of the Great Lakes (Kuan 1995; Kelley 1998), estuaries (Wei 1992), and the northwest Atlantic Ocean (Ezer et al. 1992). The model has also been used to produce real-time, numerical nowcasts and forecasts for the Great Lakes (Schwab and Bedford 1994; Yen et al. 1994), Port of New York–New Jersey (Wei and Sun 1998), and Galveston Bay, Texas, (Schmalz 1998). As mentioned earlier, the version of POM used in COFS was developed by a collaborative project including NOS, NCEP, and Princeton University. A brief description of the model configuration is given in this section. A more detailed discussion of COFS can be found in Kelley et al. (1999).

a. Grid configuration

The model uses a bottom-following sigma vertical coordinate system and curvilinear orthogonal horizontal grid. The COFS grid domain covers the region off the U.S. East Coast from approximately 30° to 47°N latitude and from the coast to 50°W longitude (Fig. 2). The domain has 181 grid points in the x direction and 101 points in the y direction. The spatial resolution of the model grid varies from 20 km offshore to 10 km near shore. In the vertical, 19 sigma layers have been used, with at least one-half of the layers contained in the upper 100 m. Sigma is defined as
i1520-0434-17-4-670-e1
where z is the vertical height, η(x, y, t) is the surface elevation, and H(x, y) is the bottom topography relative to model datum.

The 19 sigma levels are: 0.0, −0.004, −0.008, −0.0016, −0.0032, −0.0064, −0.0128, −0.0256, −0.05, −0.1, −0.2, −0.3, −0.4, −0.5, −0.6, −0.7, −0.8, −0.9, and −1.0. The coastal boundary corresponds to the 10-m isobath on the continental shelf. The model bathymetry is based on the U.S. Navy Digital Bathymetric Data Base on a 5-min grid (DBDB-5), modified over the continental shelf with a more accurate NOS bathymetry database on a 15-s grid (NOS-15).

b. Lateral boundary conditions

Because the COFS domain has open boundaries at its southern and eastern extremities, it is necessary to prescribe the forcing on this domain by the open ocean through the specification of appropriate boundary conditions. Such conditions on open boundaries normally should be specified by using the model fields obtained from a time-dependent basin-scale model that is also running in real time. However, because no real-time basin-scale forecast fields are available for the North Atlantic Ocean, the lateral boundary conditions are based on monthly climatological estimates of temperature and annual climatological estimates of salinity and transport. The temperature and salinity values are based on the U.S. Navy's Generalized Digital Environmental Model (Teague et al. 1990). The transport of water in and out of the domain is specified on the open boundaries as depicted in Fig. 2. On the southern boundary, inflows of 58.25 Sverdrups (Sv) of water and outflows of 36.25 Sv of water are prescribed and distributed horizontally based in part on the measurements from the Subtropical Atlantic Climate Studies program (Leaman et al. 1987). One Sverdrup is equal to 106 m3 s−1. On the eastern boundary, a total of 90 Sv of water are allowed to exit the domain between 37° and 40°N while a total inflow of 38 Sv of water enter north of the Gulf Stream, along the continental slope, and 30 Sv of water enter south of the Gulf Stream. The inflows and outflows on the eastern boundary, which are based on diagnostic calculations and observations (Richardson 1985), represent the northern recirculation gyre and southern subtropical gyre, respectively. The total inflow and outflow across the COFS boundaries are constrained to be equal to zero.

Monthly values of freshwater input are prescribed on the model coastal boundary for 15 major rivers or estuaries and the Gulf of St. Lawrence based on the monthly climatic means of Blumberg and Grehl (1987) and Koutitonsky and Bugden (1991), respectively. The location of specified transports and the names of rivers for which water input is prescribed are also depicted in Fig. 2.

c. Tidal forcing

Tidal forcing was added to POM on 15 November 1996. It includes tidal boundary forcing on the eastern and southern lateral open boundaries via tidal barotropic currents and astronomical tidal forcing within the model domain (Chen and Mellor 1999). Both the boundary and astronomical tidal forcing include three semidiurnal constituents: principal lunar, principal solar, and lunar elliptic, and three diurnal constituents: luni-solar, principal lunar, and principal solar. The inclusion of tides in COFS improved subtidal water levels by approximately 10% (Aikman et al. 1998).

d. Surface boundary conditions

The model is driven at its upper boundary by heat, moisture, and momentum fluxes from the 0000 UTC cycle of NCEP's eta mesoscale atmospheric forecast model (Black 1994). The fluxes are not taken directly from the Eta Model's forecast fluxes but are instead calculated from the model's surface (10 m AGL) wind forecasts and the temperature and humidity forecasts for the surface (2 m AGL) and the estimates for the sea surface. The fluxes are calculated using the standard bulk parameterization with stability effects included based on Monin–Obukhov similarity theory. The Eta Model's downward shortwave and longwave radiation fluxes are adjusted to account for known overwater biases in the model; shortwave radiation values are multiplied by 0.8 and downward longwave radiation values are multiplied by 1.054. Further information on the need for adjusting Eta Model radiation fluxes is given by Breaker and Rao (1998).

During the time period discussed in this paper, the Eta Model used a horizontal resolution of 29 km with 50 layers in the vertical. For its specification of SSTs, the Eta Model used the NOAA National Environmental Satellite, Data and Information Service (NESDIS) 50-km-resolution Multichannel Sea Surface Temperature (MCSST) Analysis, which was updated approximately 2 times per week. Initial conditions for the Eta forecast cycle were provided by the Eta Data Assimilation System (Rogers et al. 1998, 1999).

3. Assimilation data and quality-control procedure

COFS assimilates SST data, including in situ and remotely sensed observations, for the most recent 48 h. The assimilation only uses data for the past 48 h in order for COFS to respond to sudden changes in SSTs resulting from the passage of tropical and extratropical cyclones along the East Coast. The time window is similar to other forecast and analysis systems designed to provide daily updates on SST conditions and short-term forecasts. Both the Met Office's Forecasting Ocean–Atmosphere Model (FOAM) for the Arctic and Atlantic Oceans (Forbes 1995; Bell et al. 2000) and NCEP's real-time global SST analysis use observations for the past 24 h (Thiebaux et al. 2001).

In situ observations are obtained from U.S. and Canadian fixed buoys, drifting buoys, Coastal Marine Automated Network (C-MAN) stations, and ships participating in the Voluntary Observing Ship (VOS) Program. The accuracies of the in situ SST observations are given in Table 1. Within the COFS domain there are 27 fixed buoys and C-MAN stations and 5–10 drifting buoys that report SST on any particular day. The locations of the fixed buoys and C-MAN stations that report SST in near real time is depicted in Fig. 3.

The remotely sensed observations consist of MCSST retrievals. These retrievals are derived from multichannel data from the Advanced Very High Resolution Radiometer (AVHRR) on NOAA's operational polar-orbiting satellites, NOAA-14 and NOAA-15. Separate equations are used to derive MCSST data from daytime and nighttime data. Each retrieval represents approximately an 8 km × 8 km area. The retrieval is based on an average of four AVHRR Global Area Coverage (GAC) spots arranged as a 2 × 2 unit array. Each GAC spot in the unit array is approximately a 4 km × 4 km square of 1-km-horizontal-resolution AVHRR data. Both daytime and nighttime MCSST retrievals are used by COFS. The accuracies of the MCSST retrievals are given in Table 1.

The MCSST retrievals are the most important data for COFS because of the broad coverage in the grid domain, especially in the area of the Gulf Stream. The number of retrievals in the COFS domain on a given day ranges from 400 to 7000, depending on cloud cover. An example of a 2-day period of excellent data coverage in the COFS grid domain is shown in Fig. 4.

Quality control of the in situ buoy and C-MAN observations is performed by the National Data Buoy Center at the National Weather Service Telecommunication Gateway prior to the in situ fixed buoy data being received at NCEP. This quality control consists of range check and a standard deviation–based, time continuity check (Gilhousen 1998). Limits for the surface water temperature check are −2° to 40°C. In the time continuity check, the maximum allowable temperature difference is dependent on the time difference in hours since the last acceptable observation and the standard deviation of the measurement. The time difference is never greater than 3 h even when the actual time difference may be greater. The standard deviation used for the time continuity check of water temperature is 8.6°C at all stations other than those in the vicinity of the Gulf Stream, where it is 12.1°C.

The MCSST retrievals are quality controlled by NESDIS using climatological and range checks. Any retrieval that differs by more than 10°C from monthly mean climate is eliminated. Any retrieval outside of the range from −2° to 35°C is also eliminated (J. Sapper 1999, personal communication). In addition, COFS preprocessing software performs a gross check to remove all data with a temperature less than 1°C or greater than 33°C. Additional checks, including ones for geographic position error, are planned for COFS.

4. Data assimilation system

The data assimilation system is based on three assimilation steps. In the first step, observed SST data are compared with the model's top-layer temperature to obtain a surface correction using the method of Derber and Rosati (1989) and Behringer (1994). In the second step, a correction field for the top model layer is projected downward into the mixed layer following the method of Chalikov et al. (1996). A similar approach has been used in FOAM for the Arctic and Atlantic Oceans (Forbes 1995; Bell et al. 2000). Last, a nudging procedure is used to apply a three-dimensional correction field slowly into the model's mixed layer. The assimilation of SST data into COFS is done continuously at every internal time step (0.2 h) of the model to prevent imbalance between the velocity and density fields.

In the Derber–Behringer method, the model-simulated SSTs, also referred to as first-guess or background estimates, are obtained at SST observation sites by bilinear interpolation. The background estimates at the observation sites are then subtracted from the observations to generate observation increments or observed-minus-background differences. Next, analysis increments, called analysis-minus-background differences, or collectively called the correction field, are obtained by an objective analysis of the observation increments. The analysis increment, or correction field, is then added to the background estimate to obtain the final analysis. The correction field is determined by statistical interpolation, with the objective analysis equation solved using an equivalent variational formulation (Lorenc 1986). This method seeks to combine the model field and observations to estimate the correction field in a manner consistent with the estimated accuracy of each. This requires estimates of the spatial error covariances for the model and the observations. The formulation uses all observations to determine the correction increment at each grid point.

In the variational formulation, the goal is to minimize an objective function, also referred to as a functional. The objective function minimized in the Derber–Behringer method consists of two terms. The first term is a measure of the fit of the corrected temperature field to the uncorrected model temperature field and the second is a measure of the fit of the corrected temperature field to the observations. The solution is a temperature correction field that balances information from the observations with the model (Behringer et al. 1998). The form of the functional is
i1520-0434-17-4-670-e2
where T is an N-component vector of corrections to the model temperature field, 𝗲 is an approximation to the N × N model error covariance matrix, To is an M-component vector of differences between the SST observations and the top-model-layer temperatures at the location of the observations, D is an interpolation operator from grid points to observation locations, and 𝗳 is an approximation to the M × M observational error covariance matrix. The number of grid points is N and the number of observations is M. The transpose of the vector is indicated by ( )T and the inverse of a matrix by ( )−1.
The first term of the functional,
TT−1T
is an estimate of the fit of the model, weighted by the inverse of the model error covariance matrix, to the corrected temperature field. The second term,
DTToT−1DTTo
is an estimate of the fit of the observations, weighted by the inverse of the observational error covariance matrix, to the corrected temperature field. Thus, for the first term, a larger error covariance diminishes the contribution of the first guess to the final analysis; for the second term, a large error covariance diminishes the contribution of the observations.

Error covariance matrices are poorly known and thus are specified in a simple ad hoc fashion (Behringer 1994). The observational error covariance matrix 𝗳 is computed from estimates of measurement errors. No attempt is made to incorporate errors of representativeness; measurement errors are treated as if they were uncorrelated, giving a diagonal error matrix. Because 𝗳 is a diagonal matrix, its inverse, 𝗳−1, can be determined directly. The diagonal elements of 𝗳−1 are the reciprocals of estimates of the observational error variances. Estimates of observation error variances can depend on the data type. In the current version of the COFS assimilation system, MCSST retrievals and in situ SST observations are assigned the errors given in Table 1. Time is incorporated into 𝗳 by the multiplication of the diagonal elements of 𝗳−1 by a time factor. A time factor of 0.5 is assigned to yesterday's (−48 to −24 h) observations and 1.0 to today's (−24 to 0 h) observations. Thus, today's observations are given more weight.

In the Derber–Rosati method, the matrix 𝗲 is considered to be independent of depth; hence, the spatial correlations are the same for each model level. To reduce computational costs, the horizontal covariances for a layer are modeled by repeated applications of a Laplacian smoother. In COFS, the Derber–Behringer method is used to assimilate the SST data into the top layer only. The model error covariance between any two points on the model's top layer is approximated by an axially symmetric Gaussian function:
i1520-0434-17-4-670-e5
where a is the first-guess error variance, r is the horizontal distance between the two model grid points of the calculation, and b is the estimate of the correlation spatial scale of the model error. The model error variance a influences the relative weight between the observations and the first guess. The values of a and b were arbitrarily set to (0.50°C)2 and 30 km, respectively. The objective function is minimized using a preconditioned conjugate gradient algorithm (Gill et al. 1981; Navon and Legler 1987).

The relatively short correlation spatial scale was selected based on the following reason. An evaluation was conducted of the actual correlations of COFS predictions—buoy SSTs based on observations from the fixed buoys in the northwest Atlantic Ocean. These correlations are strongly inhomogeneous/anisotropic in the high-gradient region of the Gulf Stream. These results indicated two problems. First, there are far too few data to compute correlations everywhere and, second, for reasons of computational efficiency, the assimilation scheme can accommodate only a limited amount of anisotropy. Because of these difficulties, a conservative approach was taken, and a relatively short correlation spatial scale of 30 km was chosen. This short correlation length gives negligible influence to analysis increments at separations beyond which the parameterized correlation function is surely wrong. The downside of this approach is that a more accurate correlation could extend the influence of the analysis increments and thus utilize more efficiently the SST data. The short correlation scale also serves to enhance further the influence of the abundant MCSST retrievals as compared with the geographically sparse in situ data from buoys, C-MAN stations, and ships.

In the second step, the corrected temperature field is used by the mixed-layer extrapolation scheme of Chalikov et al. (1996) to estimate a new subsurface temperature structure down through the mixed layer (Fig. 5). The scheme is physically based on the assumption that the surface temperature is correlated well with the temperature of the mixed layer and thus that it is possible to project surface information downward into the mixed layer.

To be specific, the scheme checks whether the corrected surface temperature field is warmer than the model's first guess. If the corrected field is warmer than the first guess, a new temperature profile is calculated by distributing the surface temperature throughout the mixed layer. If the corrected field is colder than the first guess, a new profile is calculated by assigning the corrected surface temperature field to a depth at which the first guess and new profiles become equal. Last, the difference is calculated between the new profile and the model's temperature profile. A limitation of this scheme is that the calculation of the mixed-layer depth is based on the first guess or model's thermal field. The thermal field is known to be in error in some areas of the model domain.

In the third step, the three-dimensional temperature differences are used to modify the model temperatures over time by Newtonian relaxation or nudging. The technique of nudging drives the model variables toward the observations by including extra forcing terms in the model equations (Haltiner and Williams 1980). In COFS, the horizontal temperature equation for scalar temperature was modified to include a term in which the temperature difference is multiplied by a constant that is an order of magnitude smaller than the inverse of the model's internal mode time step. The action of this term serves to reduce the temperature difference between the model and observations as the model is stepped forward in time. Additional details on the SST data assimilation scheme can be found in Kelley et al. (1999).

A simple sensitivity test on the impact of data assimilation was conducted on the nowcast and forecast cycles. The results (not shown) indicated that the model's thermal structure is maintained during the data assimilation cycle on most days but is gradually lost during the forecast cycle beyond a certain number of days. In the data assimilation cycle, the temperature structure is maintained by using the previous day's nowcast as the initial conditions for the present assimilation cycle.

5. Output

COFS generates each day a nowcast and 24-h forecast of temperature, salinity, and currents in three dimensions and of water level. The output is available on the POM native grid at 19 vertical sigma levels and also on a standard latitude–longitude, 10-km-resolution grid at 34 vertical Z levels. The output is packed in World Meteorological Organization Gridded Binary (GRIB) exchange format. In addition, hourly forecasts of SST at observing sites in the COFS region are made available in text format. The output is archived and available at NOAA's National Oceanographic Data Center in Silver Spring, Maryland. At the time of writing, displays of daily COFS forecasts along with information on how to obtain gridded output could be found online at NCEP's Ocean Modeling Branch Web site: http://polar.wwb.noaa.gov.

6. Method of evaluation

a. Surface temperatures

SST nowcasts from the data-assimilating version of COFS (COFS 3.2) were compared with initial conditions from its nonassimilating control run (COFS 3.1) and to observations for a 3-month period from 1 January to 31 March 1998. COFS SST predictions were evaluated at four sites whose observations were not assimilated into the model (Fig. 3). These sites were chosen to evaluate the predictions in different parts of the domain: near the coast and close to the Gulf Stream. The four sites are 1) the Portland (Maine) buoy (44007), 2) the Chesapeake Bay Light C-MAN station (CHLV2), 3) the southwest Grand Banks buoy (44138), and 4) the South Hatteras buoy (41002). Observations from Buoy 44138 were unfortunately not available after early February, and observations from Buoy 41002 were missing after mid-March.

In addition, the COFS-3.1 initial conditions and COFS-3.2 nowcasts were compared with NESDIS 14-km MCSST analyses. These analyses used the same MCSST retrievals assimilated by COFS 3.2 but were generated by NESDIS using a statistically based analysis scheme with no dynamical model. Therefore, these analyses cannot provide an independent dataset for verification. However, the analyses do provide a qualitative assessment of the spatial impact of assimilating SST data in areas in which there are no in situ observations available for verification.

Comparisons between COFS SST forecasts and observations were done using graphical techniques and statistical measures. These statistical measures included mean algebraic and absolute differences (MAD, MBD) and root-mean-square difference (rmsd).

b. Subsurface temperatures

COFS-3.1 and -3.2 subsurface temperatures during February of 1998 were compared with thermal profiles measured by expendable bathythermographs from the VOS program. These XBTs were generally confined to the top 800 m of the ocean and are concentrated along shipping lanes. The XBTs were obtained from the U.S. National Oceanographic Data Center's (NODC) Oceanographic Profile Data Base (OPDB). Although NODC performs quality control prior to the insertion of profiles into the OPDB, additional quality control was done to remove spikes and to identify unusable profiles. Some XBTs were not considered for the evaluation because they either were too noisy or were obviously unrealistic (e.g., several profiles in the Gulf of Maine showed a constant value of 0 throughout the water column). Overall, 43 profiles were available, of which 29 profiles were retained for verification of the COFS subsurface temperature predictions on the continental shelf, over the continental slope across the Gulf Stream, and in the Sargasso Sea. The profiles were located along five tracks: 1) to the east of Bermuda (3 profiles), 2) from New York City toward Puerto Rico (5 profiles), 3) from New York City toward Bermuda (13 profiles), 4) across the continental slope north of the New York City–to-Bermuda shipping lane (3 profiles), and 5) from Boston to Nova Scotia in the Gulf of Maine (5 profiles). The locations of the profiles are depicted in Fig. 6. The COFS gridpoint predictions closest to the location of each profile were extracted from the model output files.

Comparisons between COFS subsurface temperatures were done visually. A statistical evaluation of the subsurface forecasts was not warranted because of the small number of observed profiles along the shipping lanes. However, these qualitative comparisons do provide insight into the subsurface response of the model to the COFS data assimilation scheme in parts of the domain and can serve to illustrate some limitations of the scheme.

7. Results and discussion

a. Evaluation of sea surface temperatures

1) Comparison with observations

The COFS-3.1 SST initial conditions and observations from 1 October 1997 to 31 March 1998 are depicted in Fig. 7. Comparison of the COFS-3.1 values with the observations illustrates the systematic bias of the nonassimilating COFS 3.1 to overpredict SSTs. Rmsd ranged from 1.3°C at 41002 to 3.6°C at 44138 (Table 2). The positive bias of COFS 3.1 increases during the winter months, especially at CHLV2 and 44138. Algebraic differences of 5°C can be seen at CHLV2 (Fig. 7b) and 44138 (Fig. 7c) in late February.

The COFS-3.2 SST nowcasts are also shown in Fig. 7 starting from 1 January to 31 March 1998. The assimilation of SST data had a significant impact on the model. Rmsd was reduced at all four sites (Table 2). The greatest improvement occurred at 44007, CHLV2, and 44138, for which the rmsd was reduced by between 47%–50%. The least improvement (15%) was at 41002, south of the Gulf Stream where SSTs from COFS 3.1 already agreed well with in situ observations. At the two sites for which observations were available for the entire 3-month period, 44007 and CHLV2, the difference between the nowcasts and observations decreased steadily as the length of assimilation period increased.

2) Comparison with MCSST analysis

The COFS-3.2 nowcasts were also compared with the latest NESDIS 14-km-resolution MCSST analysis on a daily basis. As mentioned earlier, these analyses are created from MCSST retrievals and thus are not an independent verification. However, the analyses do provide insight into the spatial response of COFS to SST data assimilation in areas of the ocean for which there are no daily in situ observations for verification. The COFS predictions (Figs. 8a,b) are valid for 0000 UTC 28 February 1998, which is after 59 days of SST data assimilation. The NESDIS analysis is also valid for 28 February 1998. Representative difference plots between COFS 3.1 and the analysis and between COFS 3.2 and the same analysis are depicted in Fig. 9.

The assimilation of SST data had an impact in several geographic areas in the domain. The COFS-3.1 simulation has its characteristic tendency to overshoot the Gulf Stream too far to the north, near the separation site at Cape Hatteras with its accompanying surge of warm waters (Fig. 8a). This feature is less dominant in the COFS-3.2 nowcast (Fig. 8b). In addition, the north side of the Gulf Stream is more clearly defined but still not as sharp as that depicted in the MCSST analysis. Another geographic area in which the assimilation had an impact was in the continental shelf waters from Cape Hatteras to Newfoundland. COFS-3.2 SSTs on the shelf were reduced by 2°–6°C when compared with COFS 3.1.

The impact of the assimilation is highlighted in the difference maps. In the COFS-3.1-minus-the-analysis difference map (Fig. 9a), a large area of 10°C positive departures extends from Cape Hatteras to Georges Bank. This map also depicts a positive bias of 1°–6°C along the continental shelf from Cape Hatteras to the Grand Banks. The COFS-3.2-minus-the-analysis difference plot (Fig. 9b) shows a significantly smaller area of 10°C departures. The difference map also depicts significantly smaller departures of only 1°C along the continental shelf from Cape Hatteras to the Grand Banks. However, both COFS 3.1 and 3.2 failed to capture the cold eddy at 40°N and 53°W.

b. Evaluation of subsurface temperatures

The COFS-3.1 initial conditions and the COFS-3.2 nowcasts were compared with five sets of thermal profiles as measured by XBTs in the model domain during February of 1998, as mentioned in the previous section. The first set of profiles (Fig. 10) was located to the east of Bermuda in the Sargasso Sea and in the southeast quadrant of the COFS domain (Fig. 6) on 21 and 22 February. The observed profiles depict the wintertime mixed layer down to 125–150 m. The COF-3.1 and -3.2 subsurface temperatures are almost identical in two of the three profiles (Figs. 10a,c) from 0 to 800 m, and thus both deviate from the observed profiles by the same amount. Only for the profiles depicted in Fig. 10b is there a slight improvement made in the COFS-3.2 profile from 0 to 100 m. This reflects the fact the COFS data assimilation scheme relies on the presence of a temperature difference at the surface in order for an adjustment to be made below the surface.

The second set of profiles (Fig. 11) follows a track from New York City toward Puerto Rico on 14 and 15 February. The first of the five profiles was collected near the continental slope (Fig. 6) and, as seen in Fig. 11a, shows cold shelf water in the upper 50 m overlying slope water. At this location, the data assimilation has resulted in COFS 3.2 being closer to the observed profile than COFS 3.1 is by approximately 1°–2°C. However, COFS-3.2 temperatures still differ from observations by as much as 8°C in the top 100 m. This persistent large error in the SST in this region was also seen in the comparison of COFS with the MCSST analysis (Fig. 9b). At the other locations along the track (Figs. 11b–e), the COFS-3.2 profiles closely approximated the observed profile. The data assimilation only made a noticeable impact between 400 and 800 m at the second site (Fig. 11b) and between 0 and 200 m and 300 and 600 m at the third site (Fig. 11c). The COFS-3.1 and -3.2 profiles are almost identical at the fourth and fifth sites (Figs. 11d,e). The lack of impact by the assimilation scheme is due to the small temperature difference between the model and observed SSTs that determines, in part, the magnitude of the adjustment below the surface.

The third set of profiles (Fig. 12) was from New York City toward Bermuda on 14 February and consisted of 13 profiles. The first four profiles were taken on the shelf and are nearly isothermal. At three of these locations, the data assimilation reduced temperature differences between COFS and observations, by up to 3°C at one site (Fig. 12d). Figure 12e depicts a profile on the outer shelf in which the observed profile showed the isothermal shelf water down to approximately 60 m with warmer slope water below. On the other hand, COFS 3.1 showed isothermal conditions down to 35 m but with colder water below. As expected, COFS 3.2 kept the general thermal structure of 3.1; however, it had cooler temperatures down to 60 m, which resulted in reduced temperature differences throughout the column. At the next three locations, the temperature differences between COFS and the observations were reduced by 1°–2°C in the top 200 m and at one location for the entire profile (Fig. 12f). However, COFS 3.2 did not improve the shape of the thermal structure. This is due to the fact that the thermal structure of the model itself determines how the mixed-layer assimilation scheme projects the surface increments into the subsurface. At the next three locations (Figs. 12i,j,k), the assimilation scheme appears to cool the SST by 2°–3°C too much, resulting in a larger temperature difference at the surface than in COFS 3.1. When these surface increments were projected into the subsurface, the impact was beneficial on the entire thermal structure at one location (Fig. 12i) but detrimental at the other two sites (Figs. 12j,k). At the last two sites (Figs. 12l,m), COFS 3.1 closely approximated the observed profile and the assimilation had only minor impact, similar to that seen in Figs. 11b–e.

The fourth set of profiles crossed the continental slope on 18 and 19 February (Fig. 13). The three profiles were located just to the north of the New York City–to-Bermuda shipping route (Fig. 6). Two of the profiles (Figs. 13a,c) were in the warm water, probably to the south of the surface expression of the shelf-slope front, and the third (Fig. 13b) was in the colder surface water north of the shelf-slope front (Fig. 13b). COFS-3.1 profiles at these sites differed from the observed profiles by 6°–11°C at the surface and by 2°C at 800 m. Similar large differences in SSTs in this region were depicted in the COFS-3.1–MCSST analysis comparison (Fig. 8a). The assimilation scheme made a significant impact on all levels at all three locations, generally reducing the error by 2°–3°C. The largest improvement was made in the northern location (Fig. 13b), at which temperature differences in the 0–150-m layer were reduced by 3°–5°C.

The fifth set of profiles (Fig. 14) was across the Gulf of Maine from Nova Scotia toward Massachusetts Bay (Fig. 6) on 21 and 22 February. The Gulf of Maine is a midlatitude marginal sea that is affected by several processes of varying spatial and temporal scales, including river discharge, wind, surface heat flux, tides, and inflows of waters from the Scotian Shelf and the continental slope (Xue et al. 2000). Comparisons of COFS-3.1 and -3.2 profiles to the observed profiles show the difficulty that the model has in representing the complex thermal structure in the gulf during late February of 1998. The greatest departure occurred in the central gulf (Fig. 14c) at which COFS fails even to have a thermocline. Comparisons of COFS-3.1 and -3.2 profiles with the observed profiles show mixed results in improving the depiction of the thermal structure by the data assimilation scheme. Temperature differences are reduced by 1°–2°C in the top 40 m in the majority of the profiles. However, at four of the locations (Figs. 14a,b,c,e), the assimilation decreased the accuracy of the model in depicting the deeper thermal structure. At only one location (Fig. 14d) where the observed structure is the simplest does the assimilation improve the entire profile.

8. Summary and conclusions

A system for assimilating near–real time SST data into NOAA's COFS for the northwest Atlantic Ocean has been developed and tested in an operational environment. The system assimilates both in situ and remotely sensed SST data for the most recent 48 h using a three-step procedure. The procedure includes statistical interpolation, a mixed-layer assimilation scheme, and nudging.

Surface predictions from nonassimilating and assimilating versions of COFS during a 3-month period in 1998 were compared with independent in situ observations and with a MCSST analysis. In addition, COFS subsurface predictions were compared with thermal profiles measured by XBTs along five tracks during February of 1998.

The assimilation of SST data has reduced the magnitude and geographic extent of the positive surface temperature bias caused by the systematic error of POM to overshoot the Gulf Stream near Cape Hatteras, especially in winter. The assimilation has also improved SSTs near the immediate coast where the mean algebraic difference has been reduced to less than 1.3°C. Overall, the rmsd between COFS predictions and in situ observations was reduced by 47%–50%. Although the assimilation of SST has improved COFS nowcasts, the Gulf Stream still overshoots and follows the continental slope before separating near 40°N. This results in a large positive temperature bias and provides poor initial conditions for the COFS forecast cycle. Different explanations for the Gulf Stream overshoot have been put forth over the years, including model resolution, buoyancy and momentum forcing, model formulation, and topographic and coastal processes (Thompson et al. 1992). This systematic error in POM can not be eliminated by SST data assimilation alone. Other avenues, such as the treatment of open ocean boundaries, river inflows, and numerical problems associated with the use of sigma coordinates in the presence of steep bathymetry, need to be examined to improve the model performance.

The qualitative comparisons between the COFS predictions and observed profiles indicated that the data assimilation scheme had a positive impact in generally reducing temperature differences by about 2°C in the top 300 m. At the majority of XBT locations, the significant improvement of the SST forecasts was not at the expense of the subsurface thermal structure. However, the comparisons did illustrate the dependency of the scheme on the model's simulated thermal structure. The comparisons also demonstrated the need to assimilate observed profiles to depict correctly the coastal ocean thermal structure, especially in the vicinity of the Gulf Stream, continental slope, and the Gulf of Maine.

The quality of any data assimilation system depends on the quality of the error covariances used in the scheme. The error covariances are unfortunately very poorly known and are very difficult to specify. The available data from a few buoys suggests that covariances in this complex region are strongly inhomogeneous and anisotropic. Because there are far too few data to compute covariances everywhere, a conservative approach was taken in which the covariance was parameterized with a relatively short correlation scale. The advantage of this approach is that it gives negligible influence to analysis increments at separations beyond which our lack of knowledge prevents the specification of an accurate covariance function. The obvious disadvantage of this approach is that an accurate covariance function could extend the influence of the analysis increments and thus make more efficient use of the observations. Finding ways to improve the error covariances is a high priority.

The SST data assimilation system described in this paper was implemented on 1 April 1998, and the resultant nowcasts were used as the initial conditions for the daily COFS forecast cycle and the next day's assimilation cycle. A new version of COFS was implemented in September of 1999 that also assimilates altimeter data. COFS was implemented as an operational forecast model at NCEP on 12 March 2002.

Acknowledgments

COFS is the result of work conducted by two dozen individuals from NOS, NCEP, and Princeton University over the past seven years. Thanks are due to Vera Gerald of NCEP for her assistance in obtaining in situ SST observations, to Bert Katz and John Sapper for their help in accessing MCSST retrievals, and to Lech Łobocki for his instructions on the day-to-day operation of COFS. We also thank D. B. Rao, Frank Aikman, Larry Breaker, Kurt Hess, Richard Schmalz, William Hart, and Allen Greenberg for their useful reviews of this paper. Much of the data assimilation work described here was performed while the first author was a UCAR Visiting Postdoctoral Scientist in NCEP's Environmental Modeling Center, Ocean Modeling Branch.

REFERENCES

  • Aikman, F., , Mellor G. L. , , Rao D. B. , , and Waters M. P. , 1994: A feasibility study of a coastal nowcast/forecast system. Abstracts, Spring Meeting of the American Geophysical Union, Baltimore, MD, AGU, 197.

    • Search Google Scholar
    • Export Citation
  • Aikman, F., , Sheinin D. , , Chen P. , , Breaker L. , , Bosley K. , , and Rao D. B. , 1996: Towards an operational nowcast/forecast system for the U.S. East Coast. Modern Approaches to Data Assimilation in Ocean Modeling. Vol. 61, P. Malanotte-Rizzoli, Ed., Elsevier, 347–376.

    • Search Google Scholar
    • Export Citation
  • Aikman, F., , Wei E. J. , , and Schultz J. R. , 1998: Water level evaluation for the Coastal Ocean Forecast System. Preprints, Second Conf. on Coastal Atmospheric and Oceanic Prediction and Processes, Phoenix, AZ, Amer. Meteor. Soc., 1–3.

    • Search Google Scholar
    • Export Citation
  • Behringer, D. W., 1994: Sea surface height variations in the Atlantic Ocean: A comparison of TOPEX altimeter data with results from an ocean data assimilation system. J. Geophys. Res., 99 , 2468524690.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Behringer, D. W., , Ming J. , , and Leetma A. , 1998: An improved coupled model for ENSO prediction and implications for ocean initialization. Part I: The ocean data assimilation system. Mon. Wea. Rev., 126 , 10131021.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bell, M. J., , Forbes R. M. , , and Hines A. , 2000: Assessment of the FOAM global data assimilation system for real-time operational ocean forecasting. J. Mar. Syst., 25 , 122.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Black, T., 1994: The new NMC Mesoscale Eta Model: Description and forecast examples. Wea. Forecasting, 9 , 265278.

  • Blumberg, A. F., , and Grehl B. J. , 1987: A river climatology for the United States Atlantic coast. HydroQual, Inc., Mahwah, NJ, 6 pp. [Available from HydroQual, Inc., 1 Lethbridge Plaza, Mahwah, NJ 07430.].

    • Search Google Scholar
    • Export Citation
  • Breaker, L. C., , and Rao D. B. , 1998: Experience gained during the implementation of NOAA Coastal Ocean Forecast System. Proc. The Marine Technology Society Annual Conf., Baltimore, MD, The Marine Technology Society, 235–241.

    • Search Google Scholar
    • Export Citation
  • Chalikov, D., , Breaker L. , , and Łobocki L. , 1996: Parameterization of mixing in the upper ocean. National Centers for Environmental Prediction Tech. Note, 40 pp. [Available from National Weather Service, NCEP, Environmental Modeling Center, World Weather Building, 5200 Auth Road, Camp Springs, MD 20746.].

    • Search Google Scholar
    • Export Citation
  • Chen, P., , and Mellor G. L. , 1999: Determination of tidal boundary forcing using tide station data. Coastal Ocean Prediction, C. N. K. Mooers, Ed., Amer. Geophys. Union, 329–351.

    • Search Google Scholar
    • Export Citation
  • Dengg, J. A., , Beckmann A. , , and Gerdes R. , 1996: The Gulf Stream separation problem. The Warm Water Sphere of the North Atlantic Ocean, W. Krauss, Ed., Gebrader Borntreger, 253–290.

    • Search Google Scholar
    • Export Citation
  • Derber, J., , and Rosati A. , 1989: A global oceanic data assimilation system. J. Phys. Oceanogr., 19 , 13331347.

  • Ezer, T., , Ko D. S. , , and Mellor G. L. , 1992: Modeling and forecasting the Gulf Stream. Oceanic and Atmospheric Nowcasting and Forecasting, Vol. 26(2), D. L. Durham and J. W. Lewis, Eds., Marine Technology Society, 5–14.

    • Search Google Scholar
    • Export Citation
  • Forbes, R. M., 1995: Experiments with the assimilation of surface temperature and thermal profile observations into a dynamical model of the Atlantic Ocean. Forecasting Research Tech. Report 167, Forecasting Research Division, United Kingdom Meteorological Office, 31 pp.

    • Search Google Scholar
    • Export Citation
  • Fox, D. N., , Carnes M. R. , , and Mitchell J. L. , 1992: Characterizing major frontal systems: A nowcast/forecasts system for the northwest Atlantic. Oceanography, 5 , 4954.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Galperin, B., , Kantha L. H. , , Hassid S. , , and Rosati A. , 1988: A quasi-equilibrium turbulent energy model for geophysical flows. J. Atmos. Sci., 45 , 5562.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gilhousen, D. B., 1987: A field evaluation of NDBC moored buoy winds. J. Atmos. Oceanic Technol., 4 , 94104.

  • Gilhousen, D. B., . 1998: Improved real-time quality control of NDBC measurements. Preprints, 10th Symp. on Meteorological Observations and Instrumentation, Phoenix, AZ, Amer. Meteor. Soc., 363–366.

    • Search Google Scholar
    • Export Citation
  • Gill, P. E., , Murray W. , , and Wright M. H. , 1981: Practical Optimization. Academic Press, 401 pp.

  • Haltiner, G. J., , and Williams R. T. , 1980: Numerical Prediction and Dynamic Meteorology. John Wiley and Sons, 477 pp.

  • Ji, M., , Leetmaa A. , , and Derber J. , 1995: An ocean analysis system for seasonal to interannual climate studies. Mon. Wea. Rev., 123 , 460481.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kelley, J. G. W., 1998: Generation of three-dimensional lake model forecasts for Lake Erie. Wea. Forecasting, 13 , 659687.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kelley, J. G. W., , Behringer D. W. , , and Thiebaux H. J. , 1999: Description of the SST Data Assimilation System used in the NOAA Coastal Ocean Forecast System (COFS) for the U.S. East Coast. National Centers for Environmental Prediction Tech. Note, 49 pp. [Available from National Weather Service, NCEP, Environmental Modeling Center, World Weather Building, 5200 Auth Road, Camp Springs, MD 20746.].

    • Search Google Scholar
    • Export Citation
  • Koutitonsky, V. G., , and Bugden G. L. , 1991: The physical oceanography of the Gulf of St. Lawrence: A review with emphasis on the synoptic variability of the motion. The Gulf of St. Lawrence: Small Ocean or Large Estuary. Vol. 113, J.-C. Therriault, Ed., Canadian Dept. of Fisheries and Oceans, 57–90.

    • Search Google Scholar
    • Export Citation
  • Kuan, C-F., 1995: Performance evaluation of the Princeton Circulation Model for Lake Erie. Ph.D. dissertation, The Ohio State University, 376 pp. [Available from Dept. of Civil and Environmental Engineering and Geodetic Science, The Ohio State University, 470 Hitchcock Hall, 2070 Neil Avenue, Columbus, OH, 43210-1275.].

    • Search Google Scholar
    • Export Citation
  • Leaman, K. D., , Molinari R. L. , , and Vertes P. S. , 1987: Structure and variability of the Florida Current at 27°N: April 1982–July 1984. J. Phys. Oceanogr., 17 , 565583.

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

  • Mellor, G. L., 1996: Users guide for a three-dimensional, primitive equation, numerical ocean model. Program in Atmospheric and Oceanic Sciences, Princeton University, 39 pp. [Available from Program in Atmospheric and Oceanic Sciences, Princeton University, P.O. Box CN710, Sayre Hall, Princeton, NJ 08544-0710.].

    • Search Google Scholar
    • Export Citation
  • Mellor, G. L., , and Yamada T. , 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys. Space Phys, 20 , 851875.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Navon, I. M., , and Legler D. M. , 1987: Conjugate-gradient methods for large-scale minimization in meteorology. Mon. Wea. Rev., 115 , 14791502.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • NOS, 1999: NOS procedures for developing and implementing operational nowcast and forecast systems for PORTS. NOAA Tech. Rep. NOS CO-OPS 0020, Center for Operational Oceanographic Products and Services, 33 pp. [Available from NOS/CO-OPS, National PORTS Program Manager, SSMC4, 1305 East-West Highway, Silver Spring, MD 20910-3281.].

    • Search Google Scholar
    • Export Citation
  • Pichel, W. G., , Clemente-Colon P. , , Sapper J. F. , , Duda C. , , and Krasnopolsky V. M. , 1995: Validation and quality control of CoastWatch satellite-derived water-surface-temperature imagery. Proc. Third Thematic Conf. on Remote Sensing for Marine and Coastal Environments, Seattle, WA, Environmental Research Institute of Michigan, I-83–I-91.

    • Search Google Scholar
    • Export Citation
  • Richardson, P. L., 1985: Average velocity and transport of the Gulf Stream near 55°W. J. Mar. Res., 43 , 83111.

  • Robinson, A. R., 1992: Shipboard prediction with a regional forecast model. Oceanography, 5 , 4248.

  • Rogers, E., and Coauthors. 1998: Changes to the NCEP Operational “Early” Eta Analysis/Forecast System. NWS Tech. Procedures Bulletin No. 447, National Centers for Environmental Prediction, 31 pp. [Available online from http://www.nws.noaa.gov/om/447body.htm.].

    • Search Google Scholar
    • Export Citation
  • Rogers, E., , Parrish D. , , and DiMego G. , 1999: Changes to the NCEP Operational Eta Analysis. NWS Tech. Procedures Bulletin (Draft), National Weather Service, 5 pp. [Available online at http://tgsv5.nws.noaa.gov/om/tpb/3d-eta.html.].

    • Search Google Scholar
    • Export Citation
  • Schmalz, R. A. Jr,, 1998: Design of a nowcast/forecast system for Galveston Bay. Preprints, Second Conf. on Coastal Atmospheric and Oceanic Prediction and Processes, Phoenix, AZ, Amer. Meteor. Soc., 15–22.

    • Search Google Scholar
    • Export Citation
  • Schultz, J. R., , and Aikman F. I. , 1998: Evaluation of subtidal water level in NOAA's Coastal Ocean Forecast System for the U.S. East Coast. Proc., Fifth Int. Conf. on Estuarine and Coastal Modeling, Alexandria, VA, Amer. Soc. Civil Eng., 766–780.

    • Search Google Scholar
    • Export Citation
  • Schwab, D. J., , and Bedford K. W. , 1994: Initial implementation of the Great Lakes Forecasting System: A real-time system for predicting lake circulation and thermal structure. Water Pollut. Res. J. Can., 29 , 203220.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Smagorinsky, J., 1963: General circulation experiments with the primitive equations. I. The basic experiments. Mon. Wea. Rev., 91 , 99164.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Teague, W. J., , Carron M. J. , , and Hogan P. J. , 1990: A comparison between the generalized digital environmental model and Levitus climatologies. J. Geophys. Res., 95 , 71677183.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Thiebaux, J., , Katz B. , , and Wang W. , 2001: New sea-surface temperature analysis implemented at NCEP. Preprints, 18th Conf. on Weather Analysis and Forecasting, Ft. Lauderdale, FL, Amer. Meteor. Soc., J159–J163.

    • Search Google Scholar
    • Export Citation
  • Thompson, J. D., , and Schmitz Jr. W. J. , 1989: A limited-area model of the Gulf Stream: Design, initial experiments, and model–data intercomparison. J. Phys. Oceanogr., 19 , 792814.

    • Search Google Scholar
    • Export Citation
  • Thompson, J. D., , Townsend T. L. , , Wallcraft A. , , and Schmitz Jr. W. J. , 1992: Ocean prediction and the Atlantic Basin: Scientific issues and technical challenges. Oceanography, 5 , 3641.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wei, E. J., 1992: Simulation of three-dimensional hydrodynamics in Long Island Sound: Seasonal timescale. Proc. Second Int. Conf. on Estuarine and Coastal Modeling, Tampa, FL, Amer. Soc. Civil Eng., 430–440.

    • Search Google Scholar
    • Export Citation
  • Wei, E. J., , and Sun C. L. , 1998: Development of a New York/New Jersey Harbor water level and current nowcast/forecast system. Preprints, Second Conf. on Coastal Atmospheric and Oceanic Prediction and Processess, Phoenix, AZ, Amer. Meteor. Soc., 11–13.

    • Search Google Scholar
    • Export Citation
  • Xue, H., , Chai F. , , and Pettigrew N. R. , 2000: A model study of the seasonal circulation in the Gulf of Maine. J. Phys. Oceanogr., 30 , 11111133.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Yen, C-C., , Bedford K. W. , , and Schwab D. J. , 1992: Nowcast protocol for the Great Lakes Forecasting System. Proc. Second Int. Conf. on Estuarine and Coastal Modeling, Tampa, FL, Amer. Soc. Civil Eng., 140–148.

    • Search Google Scholar
    • Export Citation
  • Yen, C-C., , Kelley J. G. W. , , and Bedford K. W. , 1994: Daily procedure for GLFS nowcasts. Proc. Natl. Conf. on Hydraulic Engineering, Buffalo, NY, Amer. Soc. Civil Eng., 202–206.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Cycle configurations of (a) COFS 3.1, the nonassimilating version, and (b) COFS 3.2, the assimilating version

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 2.
Fig. 2.

Locations of COFS grid points, contours of model bathymetry, names, and approximate locations of rivers for which outflow is specified, along with the location and amount of transport on the lateral open boundaries

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 3.
Fig. 3.

Locations of fixed platforms in the COFS domain whose SST observations were used or withheld from data assimilation. The curved dotted line represents the southern and eastern boundary of the COFS curvilinear grid domain. The 200-m isobath and a 5-yr mean Gulf Stream landward surface edge are also depicted

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 4.
Fig. 4.

Example of the sources and locations of SST and MCSST observations available in the COFS domain during a 48-h period. The identifiers for in situ platforms (except drifting buoys) are given. Also shown are the 200-m isobath and the mean Gulf Stream landward surface edge. The NOAA-14 and NOAA-15 MCSST observations are plotted as red and green dots, respectively

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 5.
Fig. 5.

Schematic of the mixed-layer assimilation scheme used in COFS 3.2 (a) when the estimated SST (E) is colder than the model SST (M) and (b) when the estimated SST is warmer than the model SST

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 6.
Fig. 6.

Locations of observed thermal profiles used in the evaluation of COFS-3.1 and -3.2 subsurface temperature predictions during Feb 1998. The letter next to each profile corresponds to the locations referred to in the text and in Figs. 10–14

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 7.
Fig. 7.

Time series plots of COFS-3.1 SST initial conditions, COFS-3.2 SST nowcasts, and in situ observations at (a) Portland, ME, buoy 44007; (b) Chesapeake Bay Light Tower C-MAN station, CHLV2; (c) Southwest Grand Banks buoy 44138; and (d) South Hatteras buoy 41002 from 1 Oct 1997 to 31 Mar 1998. The observations and the COFS predictions have both been averaged over a day

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 8.
Fig. 8.

(a) COFS-3.1 SST initial condition, (b) COFS-3.2 SST nowcast, and (c) NESDIS 14-km MCSST analysis valid at 0000 UTC 28 Feb 1998

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 9.
Fig. 9.

Difference maps of (a) COFS-3.1 SST initial condition minus the NESDIS 14-km MCSST analysis, and (b) COFS-3.2 SST nowcast minus the same analysis valid at 0000 UTC 28 Feb 1998

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 10.
Fig. 10.

Observed thermal profiles (Data) and corresponding COFS-3.1 and -3.2 predictions along a shipping lane to the east of Bermuda valid at (a) 1652 UTC and (b) 2257 UTC 21 Feb and (c) 0451 UTC 22 Feb 1998

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 11.
Fig. 11.

Same as in Fig. 10 except that the profiles are along the shipping lane from New York City toward Puerto Rico (NW to SE) valid at (a) 2228 UTC 14 Feb and (b) 0503 UTC, (c) 1028 UTC, (d) 1717 UTC, and (e) 2231 UTC 15 Feb 1998.

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 12.
Fig. 12.

Same as in Fig. 10 except that the profiles are along the shipping lane from New York City toward Bermuda (NW to SE) valid 14 Feb 1998 at (a) 0206 UTC, (b) 0302 UTC, (c) 0355 UTC, (d) 0449 UTC, (e) 0605 UTC, (f) 0954 UTC, (g) 1110 UTC, (h) 1303 UTC, (i) 1458 UTC, (j) 1618 UTC, (k) 1701 UTC, (l) 1801 UTC, and (m) 1923 UTC

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 13.
Fig. 13.

Same as in Fig. 10 except that the profiles are across the continental slope just to the north of the New York City–to-Bermuda shipping lane valid at (a) 2349 UTC 18 Feb and (b) 0157 UTC and (c) 1248 UTC 19 Feb 1998

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Fig. 14.
Fig. 14.

Same as in Fig. 10 except that the profiles are along the shipping lanes from Boston to Nova Scotia (E to W) valid at (a) 2327 UTC 21 Feb and (b) 0200 UTC, (c) 0604 UTC, (d) 0758 UTC, and (e) 0952 UTC 22 Feb 1998

Citation: Weather and Forecasting 17, 4; 10.1175/1520-0434(2002)017<0670:AOSDIA>2.0.CO;2

Table 1. 

Measurement error assigned to different types of SST data

Table 1. 
Table 2. 

Performance measures (forecast minus observed, °C) of daily COFS-3.1 initial conditions and COFS-3.2 nowcasts for the period 1 Jan–31 Mar 1998

Table 2. 

*

National Centers for Environmental Prediction/Ocean Modeling Branch Contribution Number 195.

Save