• Ackerman, S. A., K. I. Strabala, W. P. Menzel, R. A. Frey, C. C. Moeller, and L. E. Gumley, 1998: Discriminating clear sky from clouds with MODIS. J. Geophys. Res., 103 , 3214132157.

    • Search Google Scholar
    • Export Citation
  • Cayula, J-F., and P. Cornillon, 1996: Cloud detection from a sequence of SST images. Remote Sens. Environ., 55 , 8088.

  • Chelton, D. B., 2005: The impact of SST specification on ECMWF surface wind stress fields in the eastern tropical Pacific. J. Climate, 18 , 530550.

    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., and F. J. Wentz, 2005: Global microwave satellite observations of sea surface temperature for numerical weather prediction and climate research. Bull. Amer. Meteor. Soc., 86 , 10971115.

    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., M. G. Schlax, M. H. Freilich, and R. F. Milliff, 2004: Satellite measurements reveal persistent small-scale features in ocean winds. Science, 303 , 978983.

    • Search Google Scholar
    • Export Citation
  • Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129 , 569585.

    • Search Google Scholar
    • Export Citation
  • Chen, F., and Coauthors, 2007: Description of evaluation of the characteristics of the NCAR High-Resolution Land Data Assimilation System during IHOP-02. J. Appl. Meteor. Climatol., 46 , 694713.

    • Search Google Scholar
    • Export Citation
  • Clowes, E. S., 1917: Sea breeze on eastern Long Island. Mon. Wea. Rev., 45 , 345346.

  • Crosman, E. T., and J. D. Horel, 2009: MODIS-derived surface temperature of the Great Salt Lake. Remote Sens. Environ., 113 , 7381.

  • Donlon, C., P. J. Minnett, C. Gentemann, T. J. Nightingale, I. J. Barton, B. Ward, and M. J. Murray, 2002: Toward improved validation of satellite sea surface skin temperature measurements for climate research. J. Climate, 15 , 353369.

    • Search Google Scholar
    • Export Citation
  • Donlon, C., and Coauthors, 2007: The Global Ocean Data Assimilation Experiment High-Resolution Sea Surface Temperature Pilot Project. Bull. Amer. Meteor. Soc., 88 , 11971213.

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

    • Search Google Scholar
    • Export Citation
  • EOS Project Science Office, 2000: Surface temperature of land and oceans, fire occurrence, and volcanic effects—An overview. EOL Data Products Handbook, C. L. Parkinson and R. Greenstone, Eds. NASA, 152–154. [Available from EOS Project Science Office, Code 900, Goddard Space Flight Center, Greenbelt, MD 20771; also available online at eos.nasa.gov].

    • Search Google Scholar
    • Export Citation
  • Esaias, W. E., and Coauthors, 1998: An overview of MODIS capabilities for ocean science observations. IEEE Trans. Geosci. Remote, 36 , 12501265.

    • Search Google Scholar
    • Export Citation
  • Flament, P., J. Firing, M. Sawyer, and C. Trefois, 1994: Amplitude and horizontal structure of a large sea surface warming event during the Coastal Ocean Dynamics Experiment. J. Phys. Oceanogr., 24 , 124139.

    • Search Google Scholar
    • Export Citation
  • Franz, B., cited. 2009: Implementation of SST processing within the OBPG. [Available online at http://oceancolor.gsfc.nasa.gov/DOCS/modis_sst/].

    • Search Google Scholar
    • Export Citation
  • Gemmill, W., B. Katz, and X. Li, 2007: Daily real-time global sea surface temperature–high resolution analysis at NOAA/NCEP. NOAA/NWS/NCEP/MMAB Office Note 260, NOAA/NCEP, 39 pp. [Available online at http://polar.ncep.noaa.gov/sst/].

    • Search Google Scholar
    • Export Citation
  • Haines, S. L., G. J. Jedlovec, and S. M. Lazarus, 2007: A MODIS sea surface temperature composite for regional applications. IEEE Trans. Geosci. Remote, 45 , 29192927.

    • Search Google Scholar
    • Export Citation
  • Hong, S-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134 , 23182341.

    • Search Google Scholar
    • Export Citation
  • Janjić, Z. I., 1996: The surface layer in the NCEP Eta Model. Preprints, 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 354–355.

    • Search Google Scholar
    • Export Citation
  • Janjić, Z. I., 2002: Nonsingular implementation of the Mellor–Yamada level 2.5 scheme in the NCEP Meso Model. NCEP Office Note 437, 61 pp.

    • Search Google Scholar
    • Export Citation
  • Jeffreys, H., 1922: On the dynamics of wind. Quart. J. Roy. Meteor. Soc., 48 , 2946.

  • Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43 , 170181.

  • Kawai, Y., and H. Kawamura, 2002: Evaluation of the diurnal warming of sea surface temperature using satellite-derived marine meteorological data. J. Oceanogr., 58 , 805814.

    • Search Google Scholar
    • Export Citation
  • Knievel, J. C., G. H. Bryan, and J. P. Hacker, 2007: Explicit numerical diffusion in the WRF model. Mon. Wea. Rev., 135 , 38083824.

  • LaCasse, K. M., M. E. Splitt, S. M. Lazarus, and W. M. Lapenta, 2008: The impact of high-resolution sea surface temperatures on the simulated nocturnal Florida marine boundary layer. Mon. Wea. Rev., 136 , 13491372.

    • Search Google Scholar
    • Export Citation
  • Lazarus, S. M., C. G. Calvert, M. E. Splitt, P. Santos, D. W. Sharp, P. F. Blottman, and S. M. Spratt, 2007: Real-time, high-resolution, space–time analysis of sea surface temperatures from multiple platforms. Mon. Wea. Rev., 135 , 31583173.

    • Search Google Scholar
    • Export Citation
  • Li, J., X. Gao, R. A. Maddox, and S. Sorooshian, 2005: Sensitivity of North American monsoon rainfall to multisource sea surface temperatures in MM5. Mon. Wea. Rev., 133 , 29222939.

    • Search Google Scholar
    • Export Citation
  • Liang, X-Z., H. I. Choi, K. E. Kunkel, Y. Dai, E. Joseph, J. X. L. Wang, and P. Kumar, 2005: Surface boundary conditions for mesoscale regional climate models. Earth Interact., 9 .[Available online at http://EarthInteractions.org].

    • Search Google Scholar
    • Export Citation
  • Lin, Y-L., R. D. Farley, and H. D. Orville, 1983: Bulk parameterization of the snow field in a cloud model. J. Climate Appl. Meteor., 22 , 10651092.

    • Search Google Scholar
    • Export Citation
  • Liu, Y., and Coauthors, 2008: The operational mesogamma-scale analysis and forecast system of the U.S. Army Test and Evaluation Command. Part I: Overview of the modeling system, the forecast products, and how the products are used. J. Appl. Meteor. Climatol., 47 , 10771092.

    • Search Google Scholar
    • Export Citation
  • Minnet, P. J., 2003: Radiometric measurements of the sea-surface skin temperature: The competing roles of the diurnal thermocline and the cool skin. Int. J. Remote Sens., 24 , 50335047.

    • Search Google Scholar
    • Export Citation
  • Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A. Clough, 1997: Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102 , 1666316682.

    • Search Google Scholar
    • Export Citation
  • NASA, cited. 2009: MODIS level 3 binned data. [Available online at http://oceancolor.gsfc.nasa.gov/PRODUCTS/modis_binned.html].

  • Neumann, J., and Y. Mahrer, 1971: A theoretical study of the land and sea breeze circulation. J. Atmos. Sci., 28 , 532542.

  • NOAA, cited. 2010: NDBC: What are the sensors’ reporting, sampling, and accuracy readings? [Available online at http://www.ndbc.noaa.gov/rsa.shtml].

    • Search Google Scholar
    • Export Citation
  • Onton, D. J., and W. J. Steenburgh, 2001: Diagnostic and sensitivity studies of the 7 December 1998 Great Salt Lake–effect snowstorm. Mon. Wea. Rev., 129 , 13181338.

    • Search Google Scholar
    • Export Citation
  • Pullen, J., T. Holt, A. F. Blumberg, and R. D. Bornstein, 2007: Atmospheric response to local upwelling in the vicinity of New York–New Jersey Harbor. J. Appl. Meteor. Climatol., 46 , 10311052.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate, 7 , 929948.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., N. A. Rayner, T. M. Smith, D. C. Stokes, and W. Wang, 2002: An improved in situ and satellite SST analysis for climate. J. Climate, 15 , 16091625.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., T. M. Smith, C. Liu, D. B. Chelton, K. S. Casey, and M. G. Schlax, 2007: Daily high-resolution-blended analysis for sea surface temperature. J. Climate, 20 , 54735496.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., C. L. Gentemann, and G. K. Corlett, 2010: Evaluation of AATSR and TMI satellite SST data. J. Climate, 23 , 152165.

  • Rotunno, R., 1983: On the linear theory of the land and sea breeze. J. Atmos. Sci., 40 , 19992009.

  • Schluessel, P., W. J. Emery, H. Grassl, and T. Mammen, 1990: On the bulk-skin temperature difference and its impact on satellite remote sensing of sea surface temperature. J. Geophys. Res., 95 , (C8). 1334113356.

    • Search Google Scholar
    • Export Citation
  • Schmidt, F. H., 1947: An elementary theory of the land- and sea-breeze circulation. J. Meteor., 4 , 915.

  • Skamarock, W. C., 2004: Evaluating mesoscale NWP models using kinetic energy spectra. Mon. Wea. Rev., 132 , 30193032.

  • Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF version 3. NCAR/TN-475+STR, 113 pp. [Available at http://www.mmm.ucar.edu/wrf/users/docs/arw_v3.pdf].

    • Search Google Scholar
    • Export Citation
  • Song, Q., D. B. Chelton, S. K. Esbensen, N. Thum, and L. W. O’Neill, 2009: Coupling between sea surface temperature and low-level winds in mesoscale numerical models. J. Climate, 22 , 146164.

    • Search Google Scholar
    • Export Citation
  • Stauffer, D. R., and N. L. Seaman, 1994: Multiscale four-dimensional data assimilation. J. Appl. Meteor., 33 , 416434.

  • Stowe, L. L., P. A. Davis, and E. P. McClain, 1999: Scientific basis and initial evaluation of the CLAVR-1 global clear/cloud classification algorithm for the Advanced Very High Resolution Radiometer. J. Atmos. Oceanic Technol., 16 , 656681.

    • Search Google Scholar
    • Export Citation
  • Stramma, L., P. Cornillon, R. A. Weller, J. F. Price, and M. G. Briscoe, 1986: Large diurnal sea surface temperature variability: Satellite and in situ measurements. J. Phys. Oceanogr., 16 , 827837.

    • Search Google Scholar
    • Export Citation
  • Stuart-Menteth, A. C., I. S. Robinson, and P. G. Challenor, 2003: A global study of diurnal warming using satellite-derived sea surface temperature. J. Geophys. Res., 108 , (C5). 31553167.

    • Search Google Scholar
    • Export Citation
  • Thiébaux, J., E. Rogers, W. Wang, and B. Katz, 2003: A new high-resolution blended global sea surface temperature analysis. Bull. Amer. Meteor. Soc., 84 , 645656.

    • Search Google Scholar
    • Export Citation
  • Warner, T. T., M. N. Lakhtakia, J. D. Doyle, and R. A. Pearson, 1990: Marine atmospheric boundary layer circulations forced by Gulf Stream sea surface temperature gradients. Mon. Wea. Rev., 118 , 309323.

    • Search Google Scholar
    • Export Citation
  • Webster, P. J., C. A. Clayson, and J. A. Curry, 1996: Clouds, radiation, and the diurnal cycle of sea surface temperature in the tropical western Pacific. J. Climate, 9 , 17121730.

    • Search Google Scholar
    • Export Citation
  • Wentz, F. J., C. Gentemann, D. Smith, and D. Chelton, 2000: Satellite measurements of sea surface temperature through clouds. Science, 288 , 847850.

    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 2005: Statistical Methods in the Atmospheric Sciences. 2nd ed. Academic Press, 648 pp.

  • Zhong, S., and E. S. Takle, 1992: An observational study of sea- and land-breeze circulation in an area of complex coastal heating. J. Appl. Meteor., 31 , 14261438.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Sensor platforms used in this study and operated by the NDBC.

  • View in gallery

    The four computational domains used for the numerical simulations. The domains’ respective grid intervals are 40.5, 13.5, 4.5, and 1.5 km. Because of space, the interval of domain 3 is not labeled. Black dots on the finest domain mark the locations of stations mentioned in the text.

  • View in gallery

    Percentage of cells on the MODIS level-3 grid within the outermost computational domain (Fig. 2) that are filled (i.e., not missing) as a function of the number of days of data that compose the temporal composite. Colored lines apply to periods of time ending on the dates listed in the key. Values for daytime retrievals (SST) are solid and for nighttime (NSST) are dashed.

  • View in gallery

    Regions used for calculation of the autocorrelations of SST in Fig. 5: 1) Great Lakes, 2) mid-Atlantic, 3) Gulf Stream, 4) Gemini, and 5) Florida. (This plot is on an arbitrary subdomain of the MOD 28 dataset and does not correspond to any computational domain used for simulations)

  • View in gallery

    Autocorrelation of daily SST on the MODIS level-3 grid as a function of the number of days of data that compose the temporal composite. Colors refer to the regions in Fig. 4. Values for daytime retrievals are solid and for nighttime they are dashed.

  • View in gallery

    Histogram of 24-h changes in daytime SST (°C) from the MOD 28 dataset in the outermost computational domain (Fig. 2). This example is for the period ending 26 Apr 2006. The isolated values near 20°C and the two lobes centered near ±10°C in the tails of the main distribution are treated as erroneous retrievals. The heavy black contour outlines the histogram calculated after an additional layer of quality control is applied, which removes SSTs that differ by ≥6°C from the values on neighboring days.

  • View in gallery

    The 11 May 2007 (a) single-day SST (°C) from Aqua and Terra and (b) the difference (°C) between the mean SST from the 12 days ending on 11 May and the single-day value. The vast regions without data in the bottom panel represent the union of missing data from the single-day field and from the composite field.

  • View in gallery

    Annual cycle of mean monthly SST (°C) averaged over region 3 (Gulf Stream) in Fig. 4. Data are from NASA’s 9-km product from Aqua during 2008.

  • View in gallery

    Graphical depiction of how RTG and MODIS data are blended to define the WRF model’s lower boundary condition. The RTG analysis includes temperatures over land but the MODIS dataset does not. Shown are the (a) RTG analysis used as the background field, (b) 12-day means of MODIS data from Aqua and Terra, (c) combined field in which holes in the MODIS data are filled with RTG analysis, and (d) the combined field mapped to the coarsest of the WRF model’s four computational grids (grid interval of 40.5 km; Fig. 2), with data over land excluded. This example is from a simulation initialized at 1200 UTC 12 May 2007. The 12-day composite is based on MODIS data from 30 Apr through 11 May 2007, the same period from which Fig. 7b was constructed.

  • View in gallery

    Spectra of SSTs from the 12-day MODIS-based composite (reds), from the 0.5° RTG analysis (greens), and from the 0.083° RTG analysis (blues), calculated over regions 3 (Gulf Stream, light shades), 4 (Gemini, medium shades), and 2 (mid-Atlantic, dark shades) as defined in Fig. 4. Spectra are calculated south to north (dashed) and west to east (solid) after averaging in x and y, respectively, for 4–15 Oct 2008 (composite) and 9 Oct 2008 (RTG analyses).

  • View in gallery

    Histograms of gradients in SSTs [°C (100 km)−1] from the 12-day MODIS-based composite (gray bars), the single-day 0.083° RTG analysis (solid black line), and the single-day 0.5° RTG analysis (dashed black line). Gradients are averages over a region equal to the outermost computational domain of the model (Fig. 2) calculated for 4–15 Oct 2008 (composite) and 9 Oct 2008 (RTG analyses).

  • View in gallery

    The MAD measured against in situ observations from NDBC platforms, of the 0.083° RTG analysis and MODIS-based composite SSTs (°C) as a function of distance (km) from the nearest mainland. Circles on the positive side of the dashed 0°C line mark instances when the MODIS-based composite more closely matches the NDBC observations; circles on the negative side mark instances when the RTG analysis more closely matches the NDBC observations. The farther a circle is from the 0°C line, the greater is the difference between the datasets’ MADs. The dashed vertical line marks a distance of 20 km from the nearest mainland.

  • View in gallery

    Two-year time series of in situ observations and gridded analyses at (a) NDBC buoy 44025 and (b) sdhn4. The top four series in each panel are SSTs (°C) from the buoys (obs, black), MODIS-based composite (MODIS, dark green), 0.5° RTG analysis [RTG (L), dark blue)], and 0.083° RTG analysis [RTG (H), dark red]. The bottom three series in each panel are the differences between the buoy data and the gridded analyses (lighter shades of their respective hues). Listed at the bottom of each panel are the 2-yr means of the absolute differences (°C) between the gridded analyses and the buoy observations. Gaps in some series correspond to gaps in the datasets’ archives as available from their sources.

  • View in gallery

    Difference in SST (°C) between the 12-day MODIS-based composite and the daily 0.083° RTG analysis (latter subtracted from the former) for 12 May 2007 on computational domain 2 of 4.

  • View in gallery

    Difference in wind at 10 m AGL (vectors) and 2-m temperature (colors) between simulations based on the 12-day MODIS-based composite and on the daily 0.083° RTG analysis (latter subtracted from the former) at 1800 UTC 12 May 2007.

  • View in gallery

    Time series of 2-m temperature (°C) at Central Park (NYC) from a forecast initialized at 1200 UTC 12 May 2007. Observations are in black dots and simulations are in the dark gray (MODIS-based composite) and light gray (0.083° RTG analysis) lines.

  • View in gallery

    Time series of the (a) meridional and (b) zonal components of the 10-m (AGL) wind speed (m s−1) at Central Park (NYC) from a forecast initialized at 1200 UTC 12 May 2007. Observations are in black dots and simulations are in the dark gray (MODIS-based composite) and light gray (0.083° RTG analysis) lines. Gaps in the observations are because of missing data.

  • View in gallery

    Difference in anomaly correlation (i.e., simulations correlated with observations) of (a) 2-m temperature (°C) and (b) 10-m meridional wind (m s−1) between simulations based on the 12-day MODIS-based composite and on the daily 0.083° RTG analysis. Simulations were initialized at 1200 UTC 12 May 2007. Positive values (reds) indicate where the MODIS-based forecasts have a higher anomaly correlation, i.e., are more correlated with the observations.

  • View in gallery

    Difference in RMSE of (a) 2-m temperature (°C) and (b) 10-m meridional wind (m s−1) between simulations based on the 12-day MODIS-based composite and on the daily 0.083° RTG analysis. Simulations were initialized at 1200 UTC 12 May 2007. Negative values (blues) indicate where the MODIS-based forecasts have a lower RMSE.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 67 42 3
PDF Downloads 41 32 1

A Simple Technique for Creating Regional Composites of Sea Surface Temperature from MODIS for Use in Operational Mesoscale NWP

View More View Less
  • 1 National Center for Atmospheric Research, Boulder, Colorado
© Get Permissions
Full access

Abstract

This paper describes a simple technique for creating regional, high-resolution, daytime and nighttime composites of sea surface temperature (SST) for use in operational numerical weather prediction (NWP). The composites are based on observations from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) aboard Aqua and Terra. The data used typically are available nearly in real time, are applicable anywhere on the globe, and are capable of roughly representing the diurnal cycle in SST. The composites’ resolution is much higher than that of many other standard SST products used for operational NWP, including the low- and high-resolution Real-Time Global (RTG) analyses. The difference in resolution is key because several studies have shown that highly resolved SSTs are important for driving the air–sea interactions that shape patterns of static stability, vertical and horizontal wind shear, and divergence in the planetary boundary layer. The MODIS-based composites are compared to in situ observations from buoys and other platforms operated by the National Data Buoy Center (NDBC) off the coasts of New England, the mid-Atlantic, and Florida. Mean differences, mean absolute differences, and root-mean-square differences between the composites and the NDBC observations are all within tenths of a degree of those calculated between RTG analyses and the NDBC observations. This is true whether or not one accounts for the mean offset between the skin temperatures of the MODIS dataset and the bulk temperatures of the NDBC observations and RTG analyses. Near the coast, the MODIS-based composites tend to agree more with NDBC observations than do the RTG analyses. The opposite is true away from the coast. All of these differences in point-wise comparisons among the SST datasets are small compared to the ±1.0°C accuracy of the NDBC SST sensors. Because skin-temperature variations from land to water so strongly affect the development and life cycle of the sea breeze, this phenomenon was chosen for demonstrating the use of the MODIS-based composite in an NWP model. A simulated sea breeze in the vicinity of New York City and Long Island shows a small, net, but far from universal improvement when MODIS-based composites are used in place of RTG analyses. The timing of the sea breeze’s arrival is more accurate at some stations, and the near-surface temperature, wind, and humidity within the breeze are more realistic.

* Current affiliation: Risø National Laboratory for Sustainable Energy, Technical University of Denmark, Roskilde, Denmark

+ Current affiliation: Department of Meteorology, Naval Postgraduate School, Monterey, California

Corresponding author address: Dr. Jason Knievel, NCAR, 3450 Mitchell Ln., Boulder, CO 80301. Email: knievel@ucar.edu

Abstract

This paper describes a simple technique for creating regional, high-resolution, daytime and nighttime composites of sea surface temperature (SST) for use in operational numerical weather prediction (NWP). The composites are based on observations from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) aboard Aqua and Terra. The data used typically are available nearly in real time, are applicable anywhere on the globe, and are capable of roughly representing the diurnal cycle in SST. The composites’ resolution is much higher than that of many other standard SST products used for operational NWP, including the low- and high-resolution Real-Time Global (RTG) analyses. The difference in resolution is key because several studies have shown that highly resolved SSTs are important for driving the air–sea interactions that shape patterns of static stability, vertical and horizontal wind shear, and divergence in the planetary boundary layer. The MODIS-based composites are compared to in situ observations from buoys and other platforms operated by the National Data Buoy Center (NDBC) off the coasts of New England, the mid-Atlantic, and Florida. Mean differences, mean absolute differences, and root-mean-square differences between the composites and the NDBC observations are all within tenths of a degree of those calculated between RTG analyses and the NDBC observations. This is true whether or not one accounts for the mean offset between the skin temperatures of the MODIS dataset and the bulk temperatures of the NDBC observations and RTG analyses. Near the coast, the MODIS-based composites tend to agree more with NDBC observations than do the RTG analyses. The opposite is true away from the coast. All of these differences in point-wise comparisons among the SST datasets are small compared to the ±1.0°C accuracy of the NDBC SST sensors. Because skin-temperature variations from land to water so strongly affect the development and life cycle of the sea breeze, this phenomenon was chosen for demonstrating the use of the MODIS-based composite in an NWP model. A simulated sea breeze in the vicinity of New York City and Long Island shows a small, net, but far from universal improvement when MODIS-based composites are used in place of RTG analyses. The timing of the sea breeze’s arrival is more accurate at some stations, and the near-surface temperature, wind, and humidity within the breeze are more realistic.

* Current affiliation: Risø National Laboratory for Sustainable Energy, Technical University of Denmark, Roskilde, Denmark

+ Current affiliation: Department of Meteorology, Naval Postgraduate School, Monterey, California

Corresponding author address: Dr. Jason Knievel, NCAR, 3450 Mitchell Ln., Boulder, CO 80301. Email: knievel@ucar.edu

1. Introduction

Numerical simulations of the atmosphere above and near large bodies of water (e.g., oceans, bays, and large inland lakes) are sensitive to how the water’s skin temperature is specified (e.g., Onton and Steenburgh 2001; Thiébaux et al. 2003; Li et al. 2005; Pullen et al. 2007; LaCasse et al. 2008; Song et al. 2009). There are various methods of specifying skin temperature, or something that approximates skin temperature. Each involves trade-offs in qualities such as spatial and temporal resolution, timeliness, ease of access and processing, precision, accuracy, global coverage, and lifetime of the dataset. The relative importance of each quality depends on the purpose of the numerical simulations. For example, studies of global climate benefit from datasets that cover large areas stably over long periods of time without artifacts, such as biases from instruments, that might corrupt physical signals (e.g., Reynolds et al. 2002; Donlon et al. 2007). On the other hand, for operational mesoscale numerical weather prediction (NWP), this paper’s emphasis, the most important qualities in a dataset of skin temperature include ease and efficiency of access and processing, temporal and spatial resolution (e.g., Chelton 2005; Song et al. 2009), and timeliness (e.g., Lazarus et al. 2007). For an operational, limited-area NWP system that needs to be relocatable, applicability anywhere on the globe is also advantageous.

Most computational domains and many bodies of water are too large for in situ observations at a few fixed locations to capture the spatial heterogeneity that can potentially be specified at a model’s lower boundary (e.g., Donlon et al. 2007; Crosman and Horel 2009). Even the spatial coverage from individual scans by satellite-borne instruments is often inadequate; swaths are narrow, orbits are unevenly timed and located, and clouds and precipitation obscure and skew the views of many instruments (Liang et al. 2005). A common solution to these shortcomings is the creation of a composite field of temperature that combines data from multiple times, multiple instruments, and often multiple types of sources, such as remotely sensed observations, in situ observations, and model analyses (e.g., Reynolds and Smith 1994; Reynolds et al. 2002; Donlon et al. 2007; Haines et al. 2007; Lazarus et al. 2007; Reynolds et al. 2007). The temporal and spatial smoothing in composites can be problematic, however. In particular, it is important that strong horizontal gradients in temperature be adequately resolved because they drive air–water interactions that shape patterns of static stability, vertical and horizontal wind shear, and divergence in the planetary boundary layer (Warner et al. 1990; Chelton et al. 2004). Increasing the resolution of the skin-temperature field used by NWP models can improve simulations (e.g., Chelton 2005; Song et al. 2009).

The goals of this paper are to describe and demonstrate a simple method of creating regional composites of sea surface temperature (SST) that 1) preserve spatial gradients on scales commensurate with the typical operational mesoscale NWP model and 2) when measured against buoys, have an accuracy similar to that of the smoother SST datasets that have traditionally been used for operational NWP. Improved resolution is our emphasis, not improved point-wise verification scores.

Our composites of SSTs are based primarily on datasets derived from the Moderate Resolution Imaging Spectroradiometers (MODIS) aboard the polar-orbiting satellites Aqua and Terra, and distributed by the National Aeronautics and Space Administration (NASA). In keeping with our application and the applications of many others involved in operational NWP, the composites are constructed from analyses that are available nearly in real time for any region of the globe; are capable of representing, at least roughly, the diurnal cycle in skin temperature; and comprise as few datasets as are needed to cover typical mesoscale NWP domains at high resolution. We keep the process simple and minimize vulnerability from unreliable sources of data. To demonstrate how the MODIS-based composites are created and how they can be used in NWP, we focus on New England and the mid-Atlantic, and apply the Weather Research and Forecasting (WRF) model to simulate the sea breeze, a phenomenon that is sensitive to SST (e.g., Clowes 1917; Jeffreys 1922; Schmidt 1947; Neumann and Mahrer 1971; Rotunno 1983; Zhong and Takle 1992). We chose this region for our demonstration because 1) the coastline has a complex, challenging geometry; 2) the proximity of the Gulf Stream, with its large temperature gradients, suggests that high resolution in SSTs might be important; 3) there is a body of literature on the sea breeze over Manhattan, Long Island, and nearby locations; and 4) one sponsor of this work is most interested in improving numerical simulations of lower-tropospheric wind over coastal cities of the eastern United States.

2. Data

a. MODIS

Swaths observed by the MODIS aboard each of the polar-orbiting satellites Aqua and Terra cover a given location on the globe roughly twice per day per satellite. From the observed radiances, NASA produces a variety of SST products (Esaias et al. 1998). For this work we use daily level-3 daytime SSTs and nighttime SSTs (NASA product MOD 28). NASA’s level-3 datasets are georeferenced, two-dimensional arrays on a global grid with cells of 4.6 km × 4.6 km at the equator (NASA 2009). For our purposes, the level-3 datasets offer several advantages over level-2 datasets—more extensive quality control and georeferencing, in particular.

Each file of NASA’s daily level-3 daytime SST comprises the arithmetic mean of skin temperatures observed in a 24-h period along the parts of satellite overpasses made during local day. The daily nighttime files comprise the corresponding temperatures observed during local night. For this paper’s specific region of interest, the timing of satellite overflights yields daytime observations from approximately 1000 to 1400 local solar time (LST) and nighttime observations from approximately 2200 to 0200 LST. NASA computes skin temperature with two algorithms, one based on bands of shortwave brightness temperature centered at 3.959 and 4.050 μm, and another based on bands of longwave brightness temperature centered at 11.0172 and 12.0324 μm (Franz 2009). The shortwave algorithm is used only for nighttime SST because sun glint during the day corrupts the retrievals. The longwave algorithm is valid during both day and night. We use the longwave SST products to construct our composites.

NASA’s MODIS-based SST datasets (MOD 28) are an estimate of skin temperature, not bulk temperature (EOS Project Science Office 2000). This is an important distinction when assessing how well level-3 data compare with other gridded SST datasets and with in situ measurements at depths of ∼1 m, which we do in section 5b. The topic of skin versus bulk temperature is covered in more detail in section 5b(1).

b. Real-time global SST

For our composites we also use the Real-Time Global (RTG) analyses from the Marine Modeling and Analysis Branch (MMAB) of the National Centers for Environmental Prediction (NCEP). Daily analyses are created from a two-dimensional variational analysis of data from buoys, ships, and satellites over the preceding 24 h (Thiébaux et al. 2003; Gemmill et al. 2007). Operational models such as the North American Model (NAM) and the global forecast model at the European Center for Medium-Range Weather Forecasts (ECMWF) use the RTG analysis for lower boundary conditions over water. Since January 2001, the analysis has been available daily on a grid with pixel sizes of 0.5° in latitude and longitude (also known as the ½°, or low resolution, analysis). In September 2005, a 0.083° analysis became available (also known as the , or high resolution, analysis). Unless otherwise stated, this paper’s subsequent references to the RTG analysis pertain to the 0.083° product.

c. Buoys and other in situ platforms

Buoy data for the verification described in section 5b are from the National Data Buoy Center (NDBC; information online at http://www.ndbc.noaa.gov). We used observations made during 2007–08 from 31 sites (Fig. 1) that include moored buoys and other instrumented platforms, such as lighthouses and piers, within the WRF model’s second largest computational domain (section 3 and Fig. 2), omitting sites within the landmask of the MODIS dataset. To calculate the mean absolute difference (MAD), root-mean-square difference (RMSD), and mean difference (MD) between NDBC observations and MODIS-based composites, we used the mean SSTs from each NDBC platform over the intervals 1500–1900 UTC (daytime) and 0300–0700 UTC (nighttime), which are the windows during which Aqua and Terra observed the area. Each mean in situ observation was then compared to the MODIS pixel collocated with the NDBC platform.

We also used observations taken during May 2004 from the same five buoys used by Lazarus et al. (2007) in their study of SSTs off the Florida coast (see their Fig. 1 for the locations of the buoys). To compare our results with the results from that study, we calculated mean buoy observations over larger windows of 1100–2300 UTC (daytime) and 0000–1000 UTC (nighttime) to match those authors’ technique (S. M. Lazarus 2009, personal communication). Sensitivity tests show little effect from changing the window length by several hours. For example, extending the nighttime window from 4 to 10 h produces changes in MD and RMSD of merely hundredths of a degree.

Moored buoys and other NDBC platforms record bulk temperature. SSTs from the moored buoys are accurate to ±1.0°C (Lazarus et al. 2007; NOAA 2010); the accuracies of the other data are not apparent from the NDBC documents to which we have access. [In comparisons between satellite-based gridded SSTs and in situ observations, one might expect the term mean absolute error to be used instead of mean absolute difference, and likewise for the other metrics. Within this context, “difference” seems more appropriate, given the large errors in the buoy data and the challenges of reconciling skin temperature and bulk temperature, as described in section 5b(1). Later in the paper we do use “error” when verifying the NWP simulations against observation of near-surface atmospheric variables.]

3. Numerical model

For our numerical simulations, we used the Advanced Research core of the Weather Research and Forecasting (WRF) model, version 3.0.1.1 (Skamarock et al. 2008) and the Real-Time Four-Dimensional Data Assimilation (RTFDDA) system (Liu et al. 2008). The latter applies Newtonian relaxation to introduce observations into the model in a computationally efficient manner that preserves their temporal dimension (Stauffer and Seaman 1994).

The four one-way nested computational domains are shown in Fig. 2. Their respective horizontal grid intervals are 40.5, 13.5, 4.5, and 1.5 km. There are 50 vertically stretched levels, with thicknesses ranging from ∼70 m in the lower troposphere to ∼600 m near the model’s top. Physical parameterizations include the Yonsei University (YSU) planetary boundary layer (PBL) scheme (Hong et al. 2006), the Noah land surface model (Chen and Dudhia 2001; Chen et al. 2007), the Monin–Obhukov surface-layer scheme (Janjić 1996, 2002), the new Kain–Fritsch cumulus scheme (Kain 2004), the Lin microphysics scheme (Lin et al. 1983), Dudhia shortwave radiation (Dudhia 1989), and Rapid Radiative Transfer Model (RRTM) longwave radiation (Mlawer et al. 1997). Explicit sixth-order diffusion (Knievel et al. 2007) is weakly applied, with the monotonic constraint.

Initial and lateral boundary conditions are derived from the Global Forecast System (GFS) operated by NCEP. The WRF model with RTFDDA assimilates data available from the Meteorological Assimilation Data Ingest System (MADIS), which includes standard near-surface observations in aviation routine weather report (METAR) format and from a variety of mesonetworks; soundings from balloons and profilers; in situ observations from aircraft, ships, and buoys; and satellite-derived wind from geostationary satellites. Quick Scatterometer (QuikSCAT) sea surface wind is also assimilated into the model.

4. Creation of the SST composites

Creation of the MODIS-based composites involves five steps. Each is designed to retain properties important to NWP and to overcome, or at least mitigate, inherent inadequacies in the individual daily level-3 files, such as missing data, erroneous retrievals that escape quality control at NASA, and asynchronous satellite overflights. During these steps, daytime and nighttime data are treated independently and identically, and are processed in parallel to produce two composite SST fields for each date. Which of the two independent composite fields is introduced into the WRF model during assimilation, or how they are blended together, depends on whether it is day or night on the finest computational domain when lower boundary conditions are specified at the start of a forecast cycle. When conducting case studies of past weather, one has the luxury of using archives of observations to update SSTs during a simulation, not just at its start. In real-time forecasting, observation-based updates of SSTs are impossible during the prediction of future conditions; only predicted SSTs (e.g., from an ocean model or an assumption of persistence) can be introduced at midsimulation. Because this paper’s focus is on the utility of MODIS-based composites for real-time forecasting, in the example simulations presented in section 6, SSTs are not updated during the short 30-h forecast cycle. Because the simulations are focused on the sea breeze, it is the daytime SSTs that are applied. For the sake of brevity, the following five steps are described in terms of daytime SST, but each is applied to nighttime SST as well.

a. Merger of daily files from Aqua and Terra

In the first step of creating our MODIS-based composites, each day’s level-3 data from Aqua and Terra are merged on the level-3 grid into a two-satellite, daily array of skin temperature. Where data from both satellites are available for a given location, the average is used; otherwise, an available datum from one of the two satellites is used. It is at this stage that the location and size of the array are defined according to what is necessary for the regional NWP computational domains. For this paper, the domains are focused on the mid-Atlantic seaboard.

b. Averaging over multiple days

Second, the two-satellite daily files from the prior N days are combined into a multiday composite of average values. This step is necessary because daily files of IR-based retrievals suffer missing data over large areas where clouds obscure instruments’ views of the water surface (e.g., Li et al. 2005). In choosing N, one must balance two competing objectives. The first is to make N large enough to fill most gaps due to missing pixels in the original level-3 dataset; NWP models cannot have an incomplete field as a lower boundary condition. The second objective is to make N small enough to retain as much of the physical signal as possible in the SST field—that is, to prevent unnecessarily large smoothing and loss of horizontal temperature gradients on the scale of NWP models.

Although balancing these two objectives does not point to an exact, ideal value of N, for the part of the Atlantic Ocean of interest to us, N = 12 days is appropriate. This window is long enough to capture most recent valid satellite retrievals (Fig. 3), yet short enough not to unduly smear the physical structure in the distribution of the ocean’s skin temperature, as represented by the autocorrelation of temperature on the MODIS grid (Figs. 4 and 5). At shorter window lengths, not only are the number of missing pixels greater, but to the left of N = 12 days in Fig. 3, increasing the window length in 1-day increments produces notable gains in valid pixels (i.e., many of the curves in Fig. 3 are still rising sharply as a function of N). To the right of N = 12 days, the gains approach an asymptote. Meanwhile, according to Fig. 5, the autocorrelation at N = 12 days differs little from the autocorrelations for windows of N = 3–11 days. The autocorrelations at N = 1 or 2 days are higher, but for windows this short, large voids would still exist in the SST field. Beyond roughly day 3, the slow seasonal signal appears to be the dominant trend in the slow rolloff of autocorrelation. Provided it is longer than 1 day and less than a few weeks, the lag can be varied without producing much change in the autocorrelation. Because the number of missing pixels in the level-3 data is strongly a function of time of year (Fig. 3), making N larger in winter and smaller in summer might be useful.

c. Removal of unphysical 24-h changes in SST

In the third step, we apply one additional layer of quality control beyond the layers that are part of NASA’s level-3 processing. Detection of clouds in IR data is imperfect (Cayula and Cornillon 1996; Stowe et al. 1999; Chelton and Wentz 2005), particularly near clouds’ edges and where clouds are thin (Lazarus et al. 2007) or low over water at night (Ackerman et al. 1998), so even the heavily processed level-3 data are occasionally corrupted by clouds—particularly cirri and low stratocumuli. This corruption can produce 24-h changes in SST that are unphysically large. Observations and simulations with simple models demonstrate that the skin temperature over open ocean can change fractions of a degree to several degrees in less than 24 h (e.g., Stramma et al. 1986; Webster et al. 1996; Kawai and Kawamura 2002; Minnet 2003; Stuart-Menteth et al. 2003). However, changes of greater than 6°C occur only in extreme and rare cases (e.g., Flament et al. 1994). Samples of the level-3 MODIS data that we examined point to a similar conclusion. For example, for some days when satellite images suggest that cloud contamination might be problematic, histograms of 24-h change in SST comprise not only a main distribution, but also two dubious sidelobes and scattered outliers (Fig. 6). Study of Fig. 6 alone, even without knowledge of the papers cited above, suggests a separation between plausible and implausible data at ±6°–9°C. In the end, we decided to define as missing any SSTs responsible for a 24-h change of magnitude 6°C or greater, when compared to both the previous and next days’ temperatures. These data are set to missing while the N = 12 days of daily merged data are being combined. When data are not available on the day immediately before or immediately after the SST being checked, only a single neighbor is used in the evaluation. This is always the situation in the case of the most recent day of the 12 used for the composite. This quality control algorithm targets single-day spikes or troughs in temperature, while leaving untouched any sudden rises or falls that simply are transitions between consecutive, multiday periods of more steady temperature. For those latter cases, we did not find an easy, automated way to determine whether the erroneous data are those before or those after a transition. Multiday steadiness in temperature on both sides of such transitions, in fact, argues against characterizing either group of values as erroneous. The questionable data that we find tend to be isolated in time, not part of several consecutive days of errors. The precise threshold of 6°C is slightly arbitrary and probably not critical; one could possibly justify a choice of 5° or 8°C. As with most efforts at quality control, erring on the side of retaining data that are probably erroneous must be balanced with erring on the side of rejecting data that are probably physical.

d. Removal of seasonal lag

In the fourth step, we compensate for the fact that using N = 12 days of daily merged files (i.e., the most recently available merged file—typically comprising “yesterday’s” retrievals, in an operational setting—plus the files from the previous 11 days) means that each composite will lag the seasonal fluctuations in SST by nominally 5.5 days—half of the past 11 days (Fig. 7). Even such a relatively short lag can equate to approximately 1°C or more during times of the year when SSTs change most rapidly (Fig. 8). To compensate for this lag, we (i) calculate on the largest computational domain the spatial and temporal mean SST from a 12-day composite of the RTG analysis, (ii) subtract that value from the most recent single-day spatial mean of RTG analysis over the same region, then (iii) add the difference between the results of steps i and ii, which is the negative of the mean lag, to every pixel of the 12-day composite MODIS SST field. The lag is best calculated from the RTG analysis, not from the MODIS data themselves, because the former has no missing pixels. Missing pixels in the latter can bias the calculation—for example, when most of the missing pixels are over the cold water of the North Atlantic. This approach assumes that, to first order, the seasonal lag can be treated as uniform across a computational domain. For most regional NWP approaches, this is a reasonable assumption, but for mesoscale models run over exceptionally large areas, this step might need to be modified so that the correction is a function of location.

e. Application of background field

After implementing the first four steps, there are inevitably still gaps in the SST field owing to persistent cloudiness and the removal of pixels with unphysical 24-h changes in SST (Figs. 3, 7b, and 9b). Accordingly, in the fifth and final step, holes in the MODIS-based composite are filled with the 12-day mean of the RTG analysis, after removal of any systematic difference between the domain-wide means of the RTG and MODIS data (Fig. 9). Removing any systematic difference prevents the background field from introducing artificial extrema when it is used to fill in the MODIS data’s holes. Figure 9d shows the daytime SST field once all five steps are completed and the resultant composite is mapped to the model’s computation domains.

f. Representing the diurnal cycle

Once the daily daytime and nighttime composites are created via the five steps described above, diurnal changes in SST can be approximated in various ways. A simple approach is to use the daytime SST composite as the lower boundary condition when it is day in simulations, and to use the nighttime composite when it is night. A more sophisticated approach is to blend the two in proportions that depend on local time of day or night. Blending can also compensate for domain-wide mean errors that are high during the day or low during the night, as explained in section 5b. For the example simulations presented in section 6, we use the simple approach.

5. Properties of the MODIS-based composites

a. Spatial resolution

Power spectra of SST (Fig. 10) and histograms of the average horizontal SST gradient (Fig. 11) demonstrate that the resolution of the MODIS-based composites is quite high, commensurate with the grid intervals of typical mesoscale NWP models, of order 1 km. The RTG analyses are markedly more coarse and smooth.

The spectra were computed on each dataset’s native grid using a one-dimensional spectral decomposition. The decomposition was done along south-to-north (dashed) and west-to-east (solid) horizontal grid lines spanning three of the regions defined in Fig. 4: region 2, the mid-Atlantic; region 3, the Gulf Stream; and region 4, Gemini. The energy densities for all south-to-north grid lines within each region were then spatially averaged east to west, and the west-to-east densities were averaged south to north. This filters out noise that arises from sampling variations for individual spectra and provides a more representative estimate of the true energy densities (Wilks 2005). Consistent with Skamarock (2004), the differences due to the direction over which the averages were computed are minor relative to the differences among regions and among SST datasets. These three regions were chosen for plotting in the figure because their spectra define the envelope of spectra from all five regions.

In Fig. 10, spectra of the 0.083° RTG analysis begin to deviate noticeably from the spectra of the MODIS-based composites at wavelengths of 100–400 km. In the vicinity of the Gulf Stream, where gradients in SST are high, the powers in the RTG analyses’ spectra are all more than an order of magnitude lower than the power in the MODIS-based composite’s spectrum at wavelengths of a few tens of kilometers. At wavelengths shorter than ∼15 km, only the MODIS-based composite has much information. Away from the Gulf Stream and the coast, in region 4 (Gemini), all spectra reflect a more homogeneous SST field. Perhaps because of fewer in situ observations in this region, the RTG analyses are particularly coarse compared to the MODIS-based composite; the former resolve the same heterogeneity as the latter only at wavelengths longer than ∼400 km.

Blended SST products, such as the coarse and fine RTG analyses, are often smoother than their pixel size or data granularity suggests, owing to the objective analysis schemes used to create them. For example, Chelton and Wentz (2005) found that the spatial variance of Reynolds SST data (a dataset that at the time incorporated measurements from the Advanced Very High Resolution Radiometer and in situ sensors), which are on a grid with intervals of 1° latitude × 1° longitude, falls off dramatically at wavelengths of ∼500 km, approximately 5 times the grid interval. In our study, the MODIS-based SSTs were not spatially smoothed through an objective analysis scheme, so there was no decrease in variance throughout the resolved scales.

Histograms of SST gradient (Fig. 11) corroborate the spectra. The distribution of gradients in the MODIS-based composite is broad, with most values in the range of 1°–4°C (100 km)−1 and some values as high as 10°–19°C (100 km)−1. In contrast, SST gradients in the two RTG analyses usually do not exceed 1°C (100 km)−1. Consistent with the spectra in Fig. 10, these histograms demonstrate that there is far less difference between the fine and coarse RTG analyses than their stated resolutions suggest.

b. Verification

Because the level-3 MODIS data that we use are skin temperatures, and not bulk temperatures (section 2a), verification of the MODIS-based composite against data from the NDBC, and comparisons against the RTG analyses, are not entirely clean. Nevertheless, such verification and comparisons still offer subjective value.

1) Skin versus bulk temperature

The temperature profile in the top few meters of the ocean is complex and variable, depending on the time of day, weather, and other conditions (e.g., Schluessel et al. 1990; Donlon et al. 2007). Examining a column of the ocean from the top downward, uppermost is the hypothetical interface between the atmosphere and the ocean. Beneath that lies the ocean’s skin, at depths of tens of microns, which is the layer sensed by the MODIS instruments. In the absence of mixing from strong wind, the skin is typically cooler than the water immediately below it, owing to a net loss of longwave radiation and to the transfer of latent and sensible heat from the water to the air (Schluessel et al. 1990). Beneath the skin is the subskin, and beneath that is the layer to which the term bulk is generally applied. In practice, bulk depths are centimeters to meters below the surface.

The offset between skin measurements and bulk measurements can sometimes make comparisons between retrievals from radiometers and observations from buoys problematic. Fortunately, for our purposes the skin–bulk offset is not critical. Our goal is to construct high-resolution regional composites of SST fields that, when measured against buoys and similar platforms, produce verification scores similar to those of other operational SST datasets. Achieving greatly superior verification scores is not our goal.

Rigorously reconciling a single skin temperature with a single bulk temperature for one comparison is extremely challenging outside of highly controlled and instrumented environments. It might require, for example, that each temperature observation be accompanied by metadata such as sensor depths and heights relative to the sea surface, and by complementary observations such as cloud cover, low-level wind speed, and water vapor mixing ratios of the sea surface and the air just above it (e.g., Schluessel et al. 1990). Not all of these are available through the standard NDBC datasets. A much simpler—and, in our case, viable—approach is to apply mean corrections to entire collections of statistics, based on results from representative field studies. From ship cruises in the North Atlantic Ocean, Schluessel et al. (1990) compared measurements of skin temperature made with a radiometer to measurements at bulk depths made with platinum resistance thermometers. The mean skin–bulk offset was −0.11°C during the day and −0.30°C during the night (skin cooler). Based on multiple cruises in multiple oceans, Donlon et al. (2002) concluded that at wind speeds above 6 m s−1, day and nighttime skin temperatures are lower than bulk temperatures (depths of 2–5 m) by merely 0.17°C ± 0.07° RMSD. For the one cruise in their collection during which bulk measurements were made at a depth of 1 m (closest to the depths in the NDBC dataset that we used), the skin–bulk offsets were −0.11°C during the day and −0.18°C during the night (skin cooler). Based on all cruises, Donlon et al. (2002) found the largest offset to be −0.44°C for calm conditions at night, as approximated through an optimal least squares fit to the observations.

Even this extreme value of −0.44°C is small relative to the 2°C error bars in the NDBC observations. Because the offset between the skin and bulk temperatures is relatively small in certain contexts, and because of the challenges of rigorously calculating the precise offset for any single pair of observations, some authors simply do not attempt any correction for the offset (e.g., Wentz et al. 2000; Haines et al. 2007; LaCasse et al. 2008; Reynolds et al. 2010). We chose to apply the simple corrections of Schluessel et al. (1990) and Donlon et al. (2002), while recognizing that doing so has no significant bearing on our results, given the small mean differences among these datasets and the inaccuracies in the NDBC observations.

2) Comparison with in situ observations

When measured against 31 moored buoys and other instrument platforms within the second largest computational domain (Fig. 1; the grid interval of domain 2 is 13.5 km), the MODIS-based daytime composite has lower MAD than that of the high- and low-resolution RTG analyses (Table 1). The extent of the MODIS-based composite’s improvement is generally a function of distance away from the mainland (Fig. 12). At all NDBC sites at least 20 km from shore, the high-resolution RTG analysis is better by a small margin. Closer to shore, the MODIS-based composite is generally better, although not without exception. A common message from Table 1 and Fig. 12 is that for all NDBC sites, the MD, MAD, and RMSD are generally very similar among the three gridded SST datasets.

High positive mean biases during the day and high negative mean biases during the night in a gridded dataset are amenable to correction if the daytime and nighttime composites are blended together to approximate the diurnal cycle. For example, by combining the daytime and nighttime composites in respective percentages of 84.6% and 15.4%, the nighttime mean difference of −0.25°C in Table 1 (without correcting for the skin–bulk offset) is entirely removed from the 2-yr dataset. Of course, using mean differences from the past to correct unknown mean differences in the present works only to the extent that mean differences persist or change very slowly over time.

Figure 13 shows representative time series of SST and MAD at station 44025 in relatively deep water, well away from land, and at station sdhn4 in very shallow water, along a pier or dock very close to land (see station locations in Figs. 1 and 2). The time series in both figures comprise sharp, fast fluctuations with periods of order 1 day, and more slowly varying seasonal fluctuations. As one would expect, seasonal extrema and rates of seasonal changes are greater at sdhn4 at least partly because of its proximity to land. These rapid seasonal changes translate into similar patterns of seasonally oscillating envelopes of MAD in the two RTG analyses and in the MODIS-based composite. At 44025, the low-resolution RTG analysis performs best among the three datasets (i.e., lower MAD). At sdhn4, the MODIS-based composite performs best. The levels of performance at both locations among all datasets are very similar, though, consistent with Table 1. The MADs of all three datasets are higher at sdhn4 than at 44025, where the differences exhibit little seasonal oscillation.

We have attempted to mitigate seasonal lags in the MODIS-based composites (section 4d), but Fig. 13 suggests that the mitigation is probably inadequate for stations whose observations are representative of areas much smaller than the area of a MODIS pixel. For example, according to Fig. 13b, during the observed jump from ∼11° to ∼24°C at sdhn4 in spring of 2007, the SST in our composites is persistently too low by 2°–5°C. The two RTG analyses have the same problem. In each of these three datasets, the pixel collocated with sdhn4 applies not only to water at the pier or dock, but also to water several kilometers offshore, where SST responds more slowly to seasonal forcing.

Our technique of combining data over the last N days makes it difficult for the composite to adjust rapidly to dramatic changes in SST. Yet it is near the coast, ironically, where seasonal changes in SST can be most rapid, that the MODIS-based composite offers the greatest improvement over the RTG analyses (Fig. 12). The explanation, we speculate, is that the high horizontal gradients in SST near the coast are problematic for the spatial smoothing inherent in the interpolation algorithms used to create the RTG analyses. In addition, the more dense spacing of the NDBC platforms near the coast might also make it more difficult for the algorithms simultaneously to fit each of many adjacent observations. This limited verification against the admittedly problematic NBDC platforms suggests that the beneficial effects of high spatial variability outweigh the detrimental effects from the lag introduced by the temporal averaging.

When interpreting these verification scores, it is important to consider that the RTG analyses incorporate the very NDBC observations used herein for verification. Although the NDBC observations do also influence the final MODIS-based composites—through the use of the RTG analyses as a background field to fill missing pixels—that influence is excluded from these statistics because the verification is done before the background field is applied. Therefore, one would expect the RTG analyses to have a distinct advantage over the MODIS-based composites in the statistics in Table 1 and Figs. 12 and 13.

c. Comparison with alternative approaches

Several other papers have described alternative approaches to using MODIS data to specify SSTs for NWP. For example, Haines et al. (2007) used level-2 data to create composite analyses of SSTs in the Gulf of Mexico and Atlantic Ocean off the Florida coast. Cells of the level-2 data are 1.0 km wide, which suggests a potential advantage over the 4.6-km level-3 data that we use, provided that 1) the processing used to create a level-2 composite does not subsequently produce too much smoothing and 2) the composite is used for a purpose for which the added resolution is important. Haines et al. (2007) did not include in their paper any spectral analyses similar to our Fig. 10, so we cannot compare variances as a function of scale. We can compare verification scores. When measured against in situ observations collected by five buoys during May 2004, the SST composite of Haines et al. (2007) produced average daytime (nighttime) RMSD of 0.63°C (0.61°C) and MD of −0.19°C (−0.31°C); measured against the same data, our technique produces RMSD of 0.67°C (0.88°C) and MD of −0.37°C (−0.74°C).

One of the main differences between our approach and that of Haines et al. (2007) is in how each defines a temporal window over which to average prior observations in order to fill missing pixels in the latest daily MODIS file. Our window is 12 days. The period is flexible, but once chosen, it is fixed and applied to the whole computational domain. Haines et al. (2007) instead defined at each point a period long enough to collect the most recent three valid SST retrievals. The lowest value is excluded, in case it was contaminated by clouds, and the other two are averaged. This algorithm was later modified by LaCasse et al. (2008) to retain the highest three of the last five retrievals. The advantage to such an approach is that the temporal window for any given location is no larger than necessary to produce a value that is unlikely to be influenced by clouds. The automatic exclusion of the lowest SSTs among the last several, whether or not an SST is suspicious, also indirectly compensates for any tendency for MODIS-based composites to be biased low when compared to bulk SSTs from buoys. This probably explains the authors’ reduced negative mean differences. One disadvantage of a variable, pixel-based composite window is that temporal changes in SST can lead to artificial and unpredictable spatial gradients and extrema. Also, averaging over fewer data points raises the influence of any given datum, which makes the technique more sensitive to the occasional erroneous value that escapes quality control (e.g., Fig. 6).

In another paper whose focus is also the waters off Florida, Lazarus et al. (2007) applied a sophisticated data-assimilation approach to create a multiplatform SST composite that included data from MODIS. In its most basic, or “standard” form (their term), the approach of Lazarus et al. (2007) yielded average daytime (nighttime) RMSD of 1.01°C (0.46°C) for the same buoys and time period used by Haines et al. (2007). However, Lazarus et al. (2007) then optimized their composite by 1) adding an additional quality control step to reject outlying data, 2) correcting biases, 3) compensating for the diurnal cycle, and 4) correcting for seasonal trends in the multiplatform data. Those four steps reduced RMSD to 0.63°C during the day and 0.46°C at night (right-most column in their Table 2). This is slightly better than what we obtained from our much simpler approach. Interestingly, Lazarus et al. (2007) concluded that not all steps of their full approach should be applied at all times, but rather that some steps should be excluded, depending on the time of day. In the end, the differences in buoy-based statistics between our study and those of Lazarus et al. (2007) and Haines et al. (2007) are small relative to the error bars on the buoys’ observations.

6. Example simulations

A thorough exploration of how composite SSTs from MODIS data affect simulations of coastal atmospheric circulations is beyond the scope of this paper. Even so, it is useful briefly to present results from an arbitrarily chosen case from 2007 that exemplifies how the MODIS-based composites are applied, and what differences they can produce in comparison with simulations based on other SST datasets.

Figure 14 shows from 12 May 2007 the difference between SSTs from the 12-day MODIS-based composite and those available in the daily 0.083° RTG analysis (the latter subtracted from the former). Over the depicted region, differences between the two datasets range from −7°C to +4°C, which is much larger than the 2-yr aggregate differences between the MODIS-based composite and the NDBC observations used for verification in section 5b. The MODIS-based SSTs are distinctly lower off the coast of Long Island and eastward. Off the Delmarva Peninsula, it is the RTG analysis that has lower SSTs, except for in the Chesapeake and Delaware Bays. We speculate that the sizes and locations of the extrema in Fig. 14 are the results of many factors, such as the differences in the effective resolutions of the two datasets (Fig. 10), the locations of the in situ data used for the RTG analysis and its spatial weighting functions, characteristics of the remote sensors, differences in the timing of satellite overpasses, and physical processes, such as shifts in ocean currents and changes in winds along the ocean surface during the 12 days used for the creation of the composite.

Not surprisingly, several test simulations of boundary layer winds have proven sensitive to these differences in SST. For example, by 1800 UTC on 12 May (Fig. 15), in a simulation initialized at 1200 UTC on 12 May with the MODIS-based composite, the sea breeze is farther inland by 10–30 km in many parts of northern New Jersey and Long Island when compared with the simulation initialized with the RTG analysis. This difference in the progress of the sea breeze is the principal change in the near-surface wind, but there are mesoscale differences elsewhere in the domain as well. The comparatively cooler water to the north and south of western Long Island leads to a stronger easterly component of the near-surface wind in western Long Island Sound and southwestern Connecticut. Farther inland and ahead of the sea breeze, in northern New Jersey and southern New York, the differences in near-surface wind are much smaller.

At Central Park, New York City (NYC) and at a variety of other locations near Manhattan and on Long Island, the MODIS-based composite slightly improves predictions of the sea breeze, as manifest in the temperature (Fig. 16), wind (Fig. 17), and humidity (not shown). Timing of the sea breeze’s arrival between 1700 and 1800 UTC at NYC is improved, as is the simulation of the breeze’s influence on near-surface conditions during the rest of the day. In particular, the model’s warm bias is consistently reduced during virtually the entire simulation. Improvements in the 10-m wind are less distinct and less consistent.

Generally similar improvements in the forecast of near-surface conditions are realized at a majority of stations in the computational domains, particularly in the vicinity of New York City, in northeastern New Jersey, and over western Long Island (Figs. 18 and 19). Figure 18 shows the anomaly correlation of the RTG-based simulation subtracted from the anomaly correlation of the MODIS-based simulation for the 2-m (AGL) temperature and the south–north component of 10-m (AGL) wind. Reds indicate where the MODIS-based forecasts have a higher anomaly correlation—that is, are more correlated with the observations. (An anomaly correlation, also known as a pattern correlation, is the correlation between a pair of datasets after each has had its mean subtracted. In this case, the first dataset of a pair is the simulation and the second is the observations. Anomaly correlations and means were calculated over the length of a 24-h forecast.) Figure 19 shows for the same two variables the RMSE of the RTG-based simulation subtracted from the RMSE of the MODIS-based simulation. Blues indicate where the MODIS-based forecasts have a lower RMSE.

At a minority—but quite a large minority—of stations, the MODIS-based composite degrades the forecast in this case. Over eastern Long Island, for example, errors in temperature are higher and the anomaly correlation lower in the MODIS-based simulation. The forecast for coastal New Jersey is also degraded in many locations.

These mixed results are not surprising and are consistent with myriad verification studies of mesoscale NWP, in which upgrades or improvements to a forecast system produce a net benefit, but not a universal benefit. More extensive verification is planned to assess whether these example results can be generalized in some way, and to explore how the results might be sensitive to the model’s configuration.

7. Summary and commentary

This paper describes a simple method of creating regional composites of sea surface temperature based on observations from the Moderate Resolution Imaging Spectroradiometer aboard each of NASA’s polar-orbiting satellites Aqua and Terra. The composites are constructed from data typically available nearly in real time, are applicable anywhere on the globe, and are capable of representing, at least roughly, the diurnal cycle in the skin temperature.

Most important, the MODIS-based composites have a much higher resolution than do many other standard SST products used for operational numerical weather prediction (NWP), such as NCEP’s Real-Time Global (RTG) analyses and the Reynolds SST dataset. When verified against in situ data, the composite’s performance is similar to that of the low- and high-resolution RTG analyses and of other recently demonstrated, high-resolution MODIS-based composites, especially considering the ±1.0°C accuracy of SST observations from the buoys used for evaluation.

Each daytime and nighttime composite is created by 1) merging daily files from Aqua and Terra; 2) combining daily files into an N-day composite, for which N = 12 was appropriate for our study; 3) removing SSTs responsible for unphysical 24-h changes in the composite; 4) removing any seasonal lag; and 5) superposing the MODIS-based composite onto a background field to eliminate any remaining holes in the data. Daytime and nighttime composites can be used together to approximate the diurnal cycle in SST.

Simulations of an arbitrary example case of a sea breeze in the vicinity of New York City and Long Island are generally improved when initialized with the MODIS-based composite instead of with RTG analyses, but the improvements are small and far from universal. At a minority of locations the forecasts are degraded, according to the simple metrics of RMSE, MAE, mean error, and anomaly correlation.

Our emphasis in defining the steps to create the MODIS-based composites is on operational, globally relocatable, mesoscale NWP. Different users of SST datasets (e.g., climatologists or oceanographers) with different goals might require a composite with more or less restrictive attributes than we require, in which case some steps might need to be modified or excluded, and additional steps might need to be added.

For the sake of brevity and focus, our emphasis has been on SST, but accurately specifying lake-surface temperatures is also important for NWP. In our experience, the global models and large, limited-area models often used for the initial and boundary conditions of operational mesoscale models can produce quite deficient lake-surface temperatures. Accordingly, we are working on a follow-up paper that will include numerical simulations based on MODIS-based composites for the Great Salt Lake in Utah. That paper will complement the recent work by Crosman and Horel (2009), who demonstrated the utility of using MODIS data for constructing climatographies of lake-surface temperature.

Acknowledgments

This work was funded by the National Aeronautics and Space Administration through Grant NNS06AA58G, and by the U.S. Army Test and Evaluation Command through an interagency agreement with the National Science Foundation. Special thanks are given to J. Pace for his support; W. Lapenta, R. Grumbine, W. Gemmill, and B. Katz helped to provide high-resolution RTG data from earlier than 2008, and P. Minnett provided valuable expertise on MODIS retrievals and on the relationships between skin temperature and bulk temperature. Thanks also are given to the following for fruitful discussions and assistance: S. Ambrose, E. Crosman, C. Davis, S. Lazarus, Y. Liu, J. Steenburgh, T. Warner, and G. Wiener. Two anonymous reviewers greatly helped us improve the manuscript. NASA’s Giovani tool (information online at http://giovanni.gsfc.nasa.gov) was used to create the plot on which Fig. 8 is based.

REFERENCES

  • Ackerman, S. A., K. I. Strabala, W. P. Menzel, R. A. Frey, C. C. Moeller, and L. E. Gumley, 1998: Discriminating clear sky from clouds with MODIS. J. Geophys. Res., 103 , 3214132157.

    • Search Google Scholar
    • Export Citation
  • Cayula, J-F., and P. Cornillon, 1996: Cloud detection from a sequence of SST images. Remote Sens. Environ., 55 , 8088.

  • Chelton, D. B., 2005: The impact of SST specification on ECMWF surface wind stress fields in the eastern tropical Pacific. J. Climate, 18 , 530550.

    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., and F. J. Wentz, 2005: Global microwave satellite observations of sea surface temperature for numerical weather prediction and climate research. Bull. Amer. Meteor. Soc., 86 , 10971115.

    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., M. G. Schlax, M. H. Freilich, and R. F. Milliff, 2004: Satellite measurements reveal persistent small-scale features in ocean winds. Science, 303 , 978983.

    • Search Google Scholar
    • Export Citation
  • Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129 , 569585.

    • Search Google Scholar
    • Export Citation
  • Chen, F., and Coauthors, 2007: Description of evaluation of the characteristics of the NCAR High-Resolution Land Data Assimilation System during IHOP-02. J. Appl. Meteor. Climatol., 46 , 694713.

    • Search Google Scholar
    • Export Citation
  • Clowes, E. S., 1917: Sea breeze on eastern Long Island. Mon. Wea. Rev., 45 , 345346.

  • Crosman, E. T., and J. D. Horel, 2009: MODIS-derived surface temperature of the Great Salt Lake. Remote Sens. Environ., 113 , 7381.

  • Donlon, C., P. J. Minnett, C. Gentemann, T. J. Nightingale, I. J. Barton, B. Ward, and M. J. Murray, 2002: Toward improved validation of satellite sea surface skin temperature measurements for climate research. J. Climate, 15 , 353369.

    • Search Google Scholar
    • Export Citation
  • Donlon, C., and Coauthors, 2007: The Global Ocean Data Assimilation Experiment High-Resolution Sea Surface Temperature Pilot Project. Bull. Amer. Meteor. Soc., 88 , 11971213.

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

    • Search Google Scholar
    • Export Citation
  • EOS Project Science Office, 2000: Surface temperature of land and oceans, fire occurrence, and volcanic effects—An overview. EOL Data Products Handbook, C. L. Parkinson and R. Greenstone, Eds. NASA, 152–154. [Available from EOS Project Science Office, Code 900, Goddard Space Flight Center, Greenbelt, MD 20771; also available online at eos.nasa.gov].

    • Search Google Scholar
    • Export Citation
  • Esaias, W. E., and Coauthors, 1998: An overview of MODIS capabilities for ocean science observations. IEEE Trans. Geosci. Remote, 36 , 12501265.

    • Search Google Scholar
    • Export Citation
  • Flament, P., J. Firing, M. Sawyer, and C. Trefois, 1994: Amplitude and horizontal structure of a large sea surface warming event during the Coastal Ocean Dynamics Experiment. J. Phys. Oceanogr., 24 , 124139.

    • Search Google Scholar
    • Export Citation
  • Franz, B., cited. 2009: Implementation of SST processing within the OBPG. [Available online at http://oceancolor.gsfc.nasa.gov/DOCS/modis_sst/].

    • Search Google Scholar
    • Export Citation
  • Gemmill, W., B. Katz, and X. Li, 2007: Daily real-time global sea surface temperature–high resolution analysis at NOAA/NCEP. NOAA/NWS/NCEP/MMAB Office Note 260, NOAA/NCEP, 39 pp. [Available online at http://polar.ncep.noaa.gov/sst/].

    • Search Google Scholar
    • Export Citation
  • Haines, S. L., G. J. Jedlovec, and S. M. Lazarus, 2007: A MODIS sea surface temperature composite for regional applications. IEEE Trans. Geosci. Remote, 45 , 29192927.

    • Search Google Scholar
    • Export Citation
  • Hong, S-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134 , 23182341.

    • Search Google Scholar
    • Export Citation
  • Janjić, Z. I., 1996: The surface layer in the NCEP Eta Model. Preprints, 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 354–355.

    • Search Google Scholar
    • Export Citation
  • Janjić, Z. I., 2002: Nonsingular implementation of the Mellor–Yamada level 2.5 scheme in the NCEP Meso Model. NCEP Office Note 437, 61 pp.

    • Search Google Scholar
    • Export Citation
  • Jeffreys, H., 1922: On the dynamics of wind. Quart. J. Roy. Meteor. Soc., 48 , 2946.

  • Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43 , 170181.

  • Kawai, Y., and H. Kawamura, 2002: Evaluation of the diurnal warming of sea surface temperature using satellite-derived marine meteorological data. J. Oceanogr., 58 , 805814.

    • Search Google Scholar
    • Export Citation
  • Knievel, J. C., G. H. Bryan, and J. P. Hacker, 2007: Explicit numerical diffusion in the WRF model. Mon. Wea. Rev., 135 , 38083824.

  • LaCasse, K. M., M. E. Splitt, S. M. Lazarus, and W. M. Lapenta, 2008: The impact of high-resolution sea surface temperatures on the simulated nocturnal Florida marine boundary layer. Mon. Wea. Rev., 136 , 13491372.

    • Search Google Scholar
    • Export Citation
  • Lazarus, S. M., C. G. Calvert, M. E. Splitt, P. Santos, D. W. Sharp, P. F. Blottman, and S. M. Spratt, 2007: Real-time, high-resolution, space–time analysis of sea surface temperatures from multiple platforms. Mon. Wea. Rev., 135 , 31583173.

    • Search Google Scholar
    • Export Citation
  • Li, J., X. Gao, R. A. Maddox, and S. Sorooshian, 2005: Sensitivity of North American monsoon rainfall to multisource sea surface temperatures in MM5. Mon. Wea. Rev., 133 , 29222939.

    • Search Google Scholar
    • Export Citation
  • Liang, X-Z., H. I. Choi, K. E. Kunkel, Y. Dai, E. Joseph, J. X. L. Wang, and P. Kumar, 2005: Surface boundary conditions for mesoscale regional climate models. Earth Interact., 9 .[Available online at http://EarthInteractions.org].

    • Search Google Scholar
    • Export Citation
  • Lin, Y-L., R. D. Farley, and H. D. Orville, 1983: Bulk parameterization of the snow field in a cloud model. J. Climate Appl. Meteor., 22 , 10651092.

    • Search Google Scholar
    • Export Citation
  • Liu, Y., and Coauthors, 2008: The operational mesogamma-scale analysis and forecast system of the U.S. Army Test and Evaluation Command. Part I: Overview of the modeling system, the forecast products, and how the products are used. J. Appl. Meteor. Climatol., 47 , 10771092.

    • Search Google Scholar
    • Export Citation
  • Minnet, P. J., 2003: Radiometric measurements of the sea-surface skin temperature: The competing roles of the diurnal thermocline and the cool skin. Int. J. Remote Sens., 24 , 50335047.

    • Search Google Scholar
    • Export Citation
  • Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A. Clough, 1997: Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102 , 1666316682.

    • Search Google Scholar
    • Export Citation
  • NASA, cited. 2009: MODIS level 3 binned data. [Available online at http://oceancolor.gsfc.nasa.gov/PRODUCTS/modis_binned.html].

  • Neumann, J., and Y. Mahrer, 1971: A theoretical study of the land and sea breeze circulation. J. Atmos. Sci., 28 , 532542.

  • NOAA, cited. 2010: NDBC: What are the sensors’ reporting, sampling, and accuracy readings? [Available online at http://www.ndbc.noaa.gov/rsa.shtml].

    • Search Google Scholar
    • Export Citation
  • Onton, D. J., and W. J. Steenburgh, 2001: Diagnostic and sensitivity studies of the 7 December 1998 Great Salt Lake–effect snowstorm. Mon. Wea. Rev., 129 , 13181338.

    • Search Google Scholar
    • Export Citation
  • Pullen, J., T. Holt, A. F. Blumberg, and R. D. Bornstein, 2007: Atmospheric response to local upwelling in the vicinity of New York–New Jersey Harbor. J. Appl. Meteor. Climatol., 46 , 10311052.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate, 7 , 929948.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., N. A. Rayner, T. M. Smith, D. C. Stokes, and W. Wang, 2002: An improved in situ and satellite SST analysis for climate. J. Climate, 15 , 16091625.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., T. M. Smith, C. Liu, D. B. Chelton, K. S. Casey, and M. G. Schlax, 2007: Daily high-resolution-blended analysis for sea surface temperature. J. Climate, 20 , 54735496.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., C. L. Gentemann, and G. K. Corlett, 2010: Evaluation of AATSR and TMI satellite SST data. J. Climate, 23 , 152165.

  • Rotunno, R., 1983: On the linear theory of the land and sea breeze. J. Atmos. Sci., 40 , 19992009.

  • Schluessel, P., W. J. Emery, H. Grassl, and T. Mammen, 1990: On the bulk-skin temperature difference and its impact on satellite remote sensing of sea surface temperature. J. Geophys. Res., 95 , (C8). 1334113356.

    • Search Google Scholar
    • Export Citation
  • Schmidt, F. H., 1947: An elementary theory of the land- and sea-breeze circulation. J. Meteor., 4 , 915.

  • Skamarock, W. C., 2004: Evaluating mesoscale NWP models using kinetic energy spectra. Mon. Wea. Rev., 132 , 30193032.

  • Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF version 3. NCAR/TN-475+STR, 113 pp. [Available at http://www.mmm.ucar.edu/wrf/users/docs/arw_v3.pdf].

    • Search Google Scholar
    • Export Citation
  • Song, Q., D. B. Chelton, S. K. Esbensen, N. Thum, and L. W. O’Neill, 2009: Coupling between sea surface temperature and low-level winds in mesoscale numerical models. J. Climate, 22 , 146164.

    • Search Google Scholar
    • Export Citation
  • Stauffer, D. R., and N. L. Seaman, 1994: Multiscale four-dimensional data assimilation. J. Appl. Meteor., 33 , 416434.

  • Stowe, L. L., P. A. Davis, and E. P. McClain, 1999: Scientific basis and initial evaluation of the CLAVR-1 global clear/cloud classification algorithm for the Advanced Very High Resolution Radiometer. J. Atmos. Oceanic Technol., 16 , 656681.

    • Search Google Scholar
    • Export Citation
  • Stramma, L., P. Cornillon, R. A. Weller, J. F. Price, and M. G. Briscoe, 1986: Large diurnal sea surface temperature variability: Satellite and in situ measurements. J. Phys. Oceanogr., 16 , 827837.

    • Search Google Scholar
    • Export Citation
  • Stuart-Menteth, A. C., I. S. Robinson, and P. G. Challenor, 2003: A global study of diurnal warming using satellite-derived sea surface temperature. J. Geophys. Res., 108 , (C5). 31553167.

    • Search Google Scholar
    • Export Citation
  • Thiébaux, J., E. Rogers, W. Wang, and B. Katz, 2003: A new high-resolution blended global sea surface temperature analysis. Bull. Amer. Meteor. Soc., 84 , 645656.

    • Search Google Scholar
    • Export Citation
  • Warner, T. T., M. N. Lakhtakia, J. D. Doyle, and R. A. Pearson, 1990: Marine atmospheric boundary layer circulations forced by Gulf Stream sea surface temperature gradients. Mon. Wea. Rev., 118 , 309323.

    • Search Google Scholar
    • Export Citation
  • Webster, P. J., C. A. Clayson, and J. A. Curry, 1996: Clouds, radiation, and the diurnal cycle of sea surface temperature in the tropical western Pacific. J. Climate, 9 , 17121730.

    • Search Google Scholar
    • Export Citation
  • Wentz, F. J., C. Gentemann, D. Smith, and D. Chelton, 2000: Satellite measurements of sea surface temperature through clouds. Science, 288 , 847850.

    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 2005: Statistical Methods in the Atmospheric Sciences. 2nd ed. Academic Press, 648 pp.

  • Zhong, S., and E. S. Takle, 1992: An observational study of sea- and land-breeze circulation in an area of complex coastal heating. J. Appl. Meteor., 31 , 14261438.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Sensor platforms used in this study and operated by the NDBC.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 2.
Fig. 2.

The four computational domains used for the numerical simulations. The domains’ respective grid intervals are 40.5, 13.5, 4.5, and 1.5 km. Because of space, the interval of domain 3 is not labeled. Black dots on the finest domain mark the locations of stations mentioned in the text.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 3.
Fig. 3.

Percentage of cells on the MODIS level-3 grid within the outermost computational domain (Fig. 2) that are filled (i.e., not missing) as a function of the number of days of data that compose the temporal composite. Colored lines apply to periods of time ending on the dates listed in the key. Values for daytime retrievals (SST) are solid and for nighttime (NSST) are dashed.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 4.
Fig. 4.

Regions used for calculation of the autocorrelations of SST in Fig. 5: 1) Great Lakes, 2) mid-Atlantic, 3) Gulf Stream, 4) Gemini, and 5) Florida. (This plot is on an arbitrary subdomain of the MOD 28 dataset and does not correspond to any computational domain used for simulations)

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 5.
Fig. 5.

Autocorrelation of daily SST on the MODIS level-3 grid as a function of the number of days of data that compose the temporal composite. Colors refer to the regions in Fig. 4. Values for daytime retrievals are solid and for nighttime they are dashed.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 6.
Fig. 6.

Histogram of 24-h changes in daytime SST (°C) from the MOD 28 dataset in the outermost computational domain (Fig. 2). This example is for the period ending 26 Apr 2006. The isolated values near 20°C and the two lobes centered near ±10°C in the tails of the main distribution are treated as erroneous retrievals. The heavy black contour outlines the histogram calculated after an additional layer of quality control is applied, which removes SSTs that differ by ≥6°C from the values on neighboring days.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 7.
Fig. 7.

The 11 May 2007 (a) single-day SST (°C) from Aqua and Terra and (b) the difference (°C) between the mean SST from the 12 days ending on 11 May and the single-day value. The vast regions without data in the bottom panel represent the union of missing data from the single-day field and from the composite field.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 8.
Fig. 8.

Annual cycle of mean monthly SST (°C) averaged over region 3 (Gulf Stream) in Fig. 4. Data are from NASA’s 9-km product from Aqua during 2008.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 9.
Fig. 9.

Graphical depiction of how RTG and MODIS data are blended to define the WRF model’s lower boundary condition. The RTG analysis includes temperatures over land but the MODIS dataset does not. Shown are the (a) RTG analysis used as the background field, (b) 12-day means of MODIS data from Aqua and Terra, (c) combined field in which holes in the MODIS data are filled with RTG analysis, and (d) the combined field mapped to the coarsest of the WRF model’s four computational grids (grid interval of 40.5 km; Fig. 2), with data over land excluded. This example is from a simulation initialized at 1200 UTC 12 May 2007. The 12-day composite is based on MODIS data from 30 Apr through 11 May 2007, the same period from which Fig. 7b was constructed.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 10.
Fig. 10.

Spectra of SSTs from the 12-day MODIS-based composite (reds), from the 0.5° RTG analysis (greens), and from the 0.083° RTG analysis (blues), calculated over regions 3 (Gulf Stream, light shades), 4 (Gemini, medium shades), and 2 (mid-Atlantic, dark shades) as defined in Fig. 4. Spectra are calculated south to north (dashed) and west to east (solid) after averaging in x and y, respectively, for 4–15 Oct 2008 (composite) and 9 Oct 2008 (RTG analyses).

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 11.
Fig. 11.

Histograms of gradients in SSTs [°C (100 km)−1] from the 12-day MODIS-based composite (gray bars), the single-day 0.083° RTG analysis (solid black line), and the single-day 0.5° RTG analysis (dashed black line). Gradients are averages over a region equal to the outermost computational domain of the model (Fig. 2) calculated for 4–15 Oct 2008 (composite) and 9 Oct 2008 (RTG analyses).

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 12.
Fig. 12.

The MAD measured against in situ observations from NDBC platforms, of the 0.083° RTG analysis and MODIS-based composite SSTs (°C) as a function of distance (km) from the nearest mainland. Circles on the positive side of the dashed 0°C line mark instances when the MODIS-based composite more closely matches the NDBC observations; circles on the negative side mark instances when the RTG analysis more closely matches the NDBC observations. The farther a circle is from the 0°C line, the greater is the difference between the datasets’ MADs. The dashed vertical line marks a distance of 20 km from the nearest mainland.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 13.
Fig. 13.

Two-year time series of in situ observations and gridded analyses at (a) NDBC buoy 44025 and (b) sdhn4. The top four series in each panel are SSTs (°C) from the buoys (obs, black), MODIS-based composite (MODIS, dark green), 0.5° RTG analysis [RTG (L), dark blue)], and 0.083° RTG analysis [RTG (H), dark red]. The bottom three series in each panel are the differences between the buoy data and the gridded analyses (lighter shades of their respective hues). Listed at the bottom of each panel are the 2-yr means of the absolute differences (°C) between the gridded analyses and the buoy observations. Gaps in some series correspond to gaps in the datasets’ archives as available from their sources.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 14.
Fig. 14.

Difference in SST (°C) between the 12-day MODIS-based composite and the daily 0.083° RTG analysis (latter subtracted from the former) for 12 May 2007 on computational domain 2 of 4.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 15.
Fig. 15.

Difference in wind at 10 m AGL (vectors) and 2-m temperature (colors) between simulations based on the 12-day MODIS-based composite and on the daily 0.083° RTG analysis (latter subtracted from the former) at 1800 UTC 12 May 2007.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 16.
Fig. 16.

Time series of 2-m temperature (°C) at Central Park (NYC) from a forecast initialized at 1200 UTC 12 May 2007. Observations are in black dots and simulations are in the dark gray (MODIS-based composite) and light gray (0.083° RTG analysis) lines.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 17.
Fig. 17.

Time series of the (a) meridional and (b) zonal components of the 10-m (AGL) wind speed (m s−1) at Central Park (NYC) from a forecast initialized at 1200 UTC 12 May 2007. Observations are in black dots and simulations are in the dark gray (MODIS-based composite) and light gray (0.083° RTG analysis) lines. Gaps in the observations are because of missing data.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 18.
Fig. 18.

Difference in anomaly correlation (i.e., simulations correlated with observations) of (a) 2-m temperature (°C) and (b) 10-m meridional wind (m s−1) between simulations based on the 12-day MODIS-based composite and on the daily 0.083° RTG analysis. Simulations were initialized at 1200 UTC 12 May 2007. Positive values (reds) indicate where the MODIS-based forecasts have a higher anomaly correlation, i.e., are more correlated with the observations.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Fig. 19.
Fig. 19.

Difference in RMSE of (a) 2-m temperature (°C) and (b) 10-m meridional wind (m s−1) between simulations based on the 12-day MODIS-based composite and on the daily 0.083° RTG analysis. Simulations were initialized at 1200 UTC 12 May 2007. Negative values (blues) indicate where the MODIS-based forecasts have a lower RMSE.

Citation: Journal of Applied Meteorology and Climatology 49, 11; 10.1175/2010JAMC2430.1

Table 1.

Averages of MD, MAD, and RMSD between three gridded SST analyses and in situ observations from 31 sites operated by the NDBC (Fig. 1). Each average is based on 36 020 individual comparisons between a datum from an NDBC site and a datum from the MODIS-based composite or from an RTG analysis made over 2007–08. Results are separated into daytime (SST; 18 053 comparisons) and nighttime (NSST; 17 967 comparisons). For the MODIS-based composite, the second, third, and fourth pairs of MD are adjusted for the offset between the skin and bulk SSTs, according to the findings of Schluessel et al. (1990) and Donlon et al. (2002). The respective adjustments based on the latter paper assume a near-surface wind speed >6 m s−1 in the first scenario, and calm near-surface winds at night in the second scenario, for which the maximum adjustment is applied. Please see the text for more details.

Table 1.

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

Save