• Arkin, P., , E. Kalnay, , J. Laver, , S. Schubert, , and K. Trenberth, 2003: Ongoing analysis of the climate system: A workshop report. Rep. 18020, UCAR, Boulder, CO. [Available online at http://www.joss.ucar.edu/joss_psg/meetings/climatesystem/FinalWorkshopReport.pdf.].

  • Barnston, A. G., , A. Leetmaa, , V. E. Kousky, , R. E. Livezey, , E. A. O'Lenic, , H. Van den Dool, , A. J. Wagner, , and D. A. Unger, 1999: NCEP Forecasts of the El Nino of 1997–98 and its U.S. impacts. Bull. Amer. Meteor. Soc, 80 , 18291852.

    • Search Google Scholar
    • Export Citation
  • Bengtsson, L., , K. I. Hodges, , and S. Hagemann, 2004: Sensitivity of the ERA40 reanalysis to the observing system: Determination of the global atmospheric circulation from reduced observations. Tellus, 56A , 456471.

    • Search Google Scholar
    • Export Citation
  • Chelliah, M., , and C. F. Ropelewski, 2000: Reanalyses-based tropospheric temperature estimates: Uncertainties in the context of global climate change detection. J. Climate, 13 , 31873205.

    • Search Google Scholar
    • Export Citation
  • Gates, W. L., and Coauthors, 1999: An overview of the results of the Atmospheric Model Intercomparison Project (AMIP I). Bull. Amer. Meteor. Soc, 80 , 2955.

    • Search Google Scholar
    • Export Citation
  • Gibson, J. K., , P. Kallberg, , S. Uppala, , A. Nomura, , A. Hernandez, , and E. Serrano, 1997: ERA description. ECMWF Re-Analysis Final Report Series 1, Shinfield Park, Reading, Berkshire, United Kingdom, 71 pp.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc, 77 , 437471.

  • Kanamitsu, M., and Coauthors, 2002a: NCEP Dynamical Seasonal Forecast System 2000. Bull. Amer. Meteor. Soc, 83 , 10191037.

  • Kanamitsu, M., , W. Ebisuzaki, , J. Woolen, , J. Potter, , and M. Fiorino, 2002b: NCEP/DOE AMIP-II reanalysis (R-2). Bull. Amer. Meteor. Soc, 83 , 16311643.

    • Search Google Scholar
    • Export Citation
  • Kistler, R., and Coauthors, 2001: The NCEP–NCAR 50-year reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc, 82 , 247268.

    • Search Google Scholar
    • Export Citation
  • Parrish, D. F., , and J. C. Derber, 1992: The National Meteorological Center's Spectral Statistical–Interpolation Analysis System. Mon. Wea. Rev, 120 , 17471763.

    • Search Google Scholar
    • Export Citation
  • Rayner, N. A., , D. E. Parker, , E. B. Horton, , C. K. Folland, , L. V. Alexander, , and D. P. Rowell, 2003: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res, 108 .4407, doi:10.1029/2002JD002670.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., 1988: A real-time global sea surface temperature analysis. J. Climate, 1 , 7586.

  • Reynolds, R. W., 1993: Impact of Mount Pinatubo aerosols on satellite-derived sea surface temperatures. J. Climate, 6 , 768774.

  • 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
  • Servain, J., , M. Seva, , and P. Rual, 1990: Climatology comparison and long-term variations of sea surface temperature over the tropical Atlantic Ocean. J. Geophys. Res, 95 , 94219431.

    • Search Google Scholar
    • Export Citation
  • Simmons, A. J., , and J. K. Gibson, 2000: The ERA-40 project plan. ERA-40 Project Report Series, ECMWF, Shinfield Park, Reading, Berkshire, United Kingdom, 62 pp.

  • Smith, T. M., , and R. W. Reynolds, 2004: Improved extended reconstruction of SST (1854–1997). J. Climate, 17 , 24662477.

  • Smith, T. M., , R. W. Reynolds, , R. E. Livezey, , and D. C. Stokes, 1996: Reconstruction of historical sea surface temperatures using empirical orthogonal functions. J. Climate, 9 , 14031420.

    • Search Google Scholar
    • Export Citation
  • Stammer, D., and Coauthors, 2002: Global ocean circulation during 1992–1997, estimated from ocean observations and a general circulation model. J. Geophys. Res, 107 , 127.

    • Search Google Scholar
    • Export Citation
  • Sturaro, G., 2003: A closer look at the climatological discontinuities present in the NCEP/NMCAR reanalysis temperature due to the introduction of satellite data. Climate Dyn, 21 , 309316.

    • Search Google Scholar
    • Export Citation
  • Whitaker, J. S., , G. P. Compo, , X. Wei, , and T. M. Hamill, 2004: Reanalysis without radiosondes using ensemble data assimilation. Mon. Wea. Rev, 132 , 11901200.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    The distribution of observation sites for surface pressure at 0000 UTC 7 Nov 1997

  • View in gallery

    Simulated Psfc data distribution of (left) land and (right) ocean extracted from surface observation data on 7 Nov 1997

  • View in gallery

    Seasonal mean SST anomaly for (top) 1993 and (bottom) 1997

  • View in gallery

    Time-averaged normalized rmsd of daily mean (top) sea level pressure and (bottom) T2M for (left) 1997 and (right) 1993. For each panel, top portion is over ocean, middle is over land, and bottom is ocean plus land. Rmsd is normalized by the daily temporal variance

  • View in gallery

    Time-averaged normalized rmsd of daily mean (top) 200- and (bottom) 500-hPa height, for (left) 1997 and (right) 1993 columns. For each panel, top portion is over ocean, middle is over land, and bottom is ocean plus land. Rmsd is normalized by the daily temporal variance

  • View in gallery

    Rmsd of seasonal average (top) MSLP and (bottom) T2M for (left) 1997 and (right) 1993

  • View in gallery

    Rmsd of (bottom) 500- and (top) 200-hPa height of seasonal mean for (left) 1997 and (right) 1993

  • View in gallery

    Difference of zonally averaged temperature (°C)

  • View in gallery

    Time-averaged normalized rmsd of daily mean (a) latent heat flux, (b) total cloudiness, (c) precipitation for 1997, and (d) precipitation for 1993. For each panel, top portion is over ocean, middle is over land, and bottom is ocean plus land. Rmsd is normalized by the daily temporal variance

  • View in gallery

    Rmsd of seasonal mean (a) latent heat flux, (b) cloudiness, (c) precipitation for 1997, and (d) precipitation for 1993

  • View in gallery

    Difference of seasonal mean MSLP

  • View in gallery

    Difference of seasonal mean 500-hPa height

  • View in gallery

    Difference of seasonal mean precipitation

  • View in gallery

    Fig. A1. Monthly distribution of (a) land surface station and (b) observation point at ocean in November 1915

  • View in gallery

    Fig. B1. Daily rmsd of 500-hPa height of individual members and ensemble mean (think soil line) from control analysis (m)

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 5 5 2
PDF Downloads 1 1 0

The Role of Sea Surface Temperature in Reanalysis

View More View Less
  • 1 Scripps Institution of Oceanography, University of California, San Diego, La Jolla, California
  • 2 Korea Meteorological Administration, Seoul, South Korea
© Get Permissions
Full access

Abstract

With the aim of understanding the role of SST in the reanalysis for the preradiosonde, presatellite, and satellite eras, a number of observation system experiments were performed using the NCEP/Department of Energy (DOE) reanalysis system. Five pairs of experiments were conducted using observed and climatological SSTs for cases 1) without any observation, 2) surface pressure observation only with the observation density of 1915, 3) surface pressure observation with the observation density of 1997, 4) surface observation and radiosondes, and 5) all observations, including satellite retrievals. The analyses were run for 4 months in 1997 (strong El Niño) and 1993 (near-normal SST). The impact of SST and the various observation systems on the analysis of near-surface parameters, the upper-level field, and several diagnostic fields were compared against the control analysis with observed SST and all available observations.

The most important finding of this study is that the impact of SST varies with the time scale of the analysis, which is most apparent in the surface-pressure-only observation experiments. The impact of SST is largest for the low-frequency (seasonal) analyses and smaller for the high-frequency (daily) analyses. This is particularly apparent for near-surface temperature and upper-level height field analyses. In the extreme case of the strong El Niño year, the simulation with observed SST without any observations [Atmospheric Model Intercomparison Project (AMIP)-type run] produced seasonal mean 2-m temperature (T2M) and 500-hPa height fields that agreed better with the control analysis than the analysis with surface pressure observation only with climatological SST. On the contrary, the impact of surface pressure observation is greater on higher-frequency analyses, and lower on low-frequency analyses.

Generally speaking, accurate analysis of SST is important when limited atmospheric observation is available. But even for the full atmospheric observation system, climatological SST produces inferior analysis over ocean as well as land.

The introduction of radiosonde data drastically reduces errors in the analyses and diagnostic fields; thus, radiosonde data are indispensable for accurate estimation of the atmospheric state in both short and long time scales.

Corresponding author address: Dr. Masao Kanamitsu, Mail Code 0224, CRD/SIO/UCSD, 9500 Gilman Drive, La Jolla, CA 92093-0224. Email: mkanamitsu@ucsd.edu

Abstract

With the aim of understanding the role of SST in the reanalysis for the preradiosonde, presatellite, and satellite eras, a number of observation system experiments were performed using the NCEP/Department of Energy (DOE) reanalysis system. Five pairs of experiments were conducted using observed and climatological SSTs for cases 1) without any observation, 2) surface pressure observation only with the observation density of 1915, 3) surface pressure observation with the observation density of 1997, 4) surface observation and radiosondes, and 5) all observations, including satellite retrievals. The analyses were run for 4 months in 1997 (strong El Niño) and 1993 (near-normal SST). The impact of SST and the various observation systems on the analysis of near-surface parameters, the upper-level field, and several diagnostic fields were compared against the control analysis with observed SST and all available observations.

The most important finding of this study is that the impact of SST varies with the time scale of the analysis, which is most apparent in the surface-pressure-only observation experiments. The impact of SST is largest for the low-frequency (seasonal) analyses and smaller for the high-frequency (daily) analyses. This is particularly apparent for near-surface temperature and upper-level height field analyses. In the extreme case of the strong El Niño year, the simulation with observed SST without any observations [Atmospheric Model Intercomparison Project (AMIP)-type run] produced seasonal mean 2-m temperature (T2M) and 500-hPa height fields that agreed better with the control analysis than the analysis with surface pressure observation only with climatological SST. On the contrary, the impact of surface pressure observation is greater on higher-frequency analyses, and lower on low-frequency analyses.

Generally speaking, accurate analysis of SST is important when limited atmospheric observation is available. But even for the full atmospheric observation system, climatological SST produces inferior analysis over ocean as well as land.

The introduction of radiosonde data drastically reduces errors in the analyses and diagnostic fields; thus, radiosonde data are indispensable for accurate estimation of the atmospheric state in both short and long time scales.

Corresponding author address: Dr. Masao Kanamitsu, Mail Code 0224, CRD/SIO/UCSD, 9500 Gilman Drive, La Jolla, CA 92093-0224. Email: mkanamitsu@ucsd.edu

1. Introduction

Long-term historical analysis of atmosphere, land, and ocean is essential in climate research. The usefulness of such analysis is demonstrated by nearly 3000 references currently made to the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996). Recognizing this importance, several additional reanalysis projects have been launched [European Centre for Medium-Range Weather Forecasts (ECMWF) 15-yr Re-Analysis (ERA-15; Gibson et al. 1997); ECMWF 40-yr Re-Analysis (ERA-40; Simmons and Gibson 2000); NCEP/Department of Energy (DOE; Kanamitsu et al. 2002b); Japanese reanalysis (http://www.jreap.org/indexe.html); and U.S. national reanalysis project (Arkin et al. 2003)]. Some are now complete, while others are ongoing.

These analyses utilize a high-level data assimilation method combining observation with guess field from a highly complex and accurate general circulation model. The greatest merit of this method is that the parameters in the resulting analyses are dynamically, physically, and hydrologically consistent with the general circulation model and the constraints included in the analysis scheme. This consistency is extremely important in climate research, since finding unknown relationships between various parameters and understanding the physical processes that connect them is a central concern.

In addition to the consistency of analyzed parameters, minimization of artificial change in the climate signal due to changes in the analysis procedure is critical. To minimize these changes, modification of the analysis system during the course of the reanalysis production needs to be avoided, even to the extent that some errors found during the process need to be left uncorrected (Kistler et al. 2001; Kanamitsu et al. 2002b).

With these careful system designs, artificial change in the climate signal should have been minimized. However, it soon became apparent that the climate signal was distorted by large changes in the observation systems that have occurred during the last 40 yr (Chelliah and Ropelewski 2000). The primary reasons for the artificial change in the analyses are the systematic bias error in the guess field, as well as systematic error in the observation. The latter is caused by changes in instruments, data processing, and even by changes in station locations. In particular, the introduction of satellite soundings in the late 1970s created distinct discontinuities in the analysis (Sturaro 2003). These biases are extremely difficult to eliminate, and minimization of the trends and discontinuities in climate signal caused by changes in the observation system is now one of the critical targets of future reanalyses (Arkin et al. 2003).

The observations used in the reanalyses are not limited to free-atmospheric observations of winds, temperature, pressure, and humidity. Additional important observations are sea surface temperature (SST) and sea ice cover. Biases and discontinuities in these analyses are also potential sources of false climate signal. Changes in vegetation types, coverage, urbanization, and irrigation over land can also play an important role in the accuracy of analysis, but no reanalyses have ever taken account of these details.

For all the atmospheric reanalyses performed to date, the SST analyses are made independently from atmospheric analysis. The SST analyses are created by objective smoothing (Reynolds 1988), statistical techniques (Reynolds and Smith 1994), and with more refined EOF fitting (Smith et al. 1996), which allows the analysis to run back to the presatellite period. Note that these analyses are based on statistical method. Combined ship and satellite observations are used for SST analysis. The accuracy of the analysis depends on the quality and density of these observations (Reynolds and Smith 1994; Reynolds 1993). Typical accuracy of the monthly average SST analysis is about 0.1° for recent years, but could be much larger due to the bias existing in satellite data, which is very difficult to estimate (Reynolds and Smith 1994). The accuracy of SST analysis is worse before the satellite era, of the order of 0.3°, and can locally exceed 1° (Smith et al. 1996). There is also a long-term trend in the SST in situ measurements, which might affect the long-term trend in SST analysis (Servain et al. 1990).

Surprisingly, all the data assimilation systems used in reanalyses assume that the SST analysis is error free. A major reason for this somewhat unrealistic assumption is that the SST is not explicitly used as observation to analyze the atmospheric parameters. Recent four-dimensional variational ocean data assimilation (Stammer et al. 2002) assigns a range of errors to the atmospheric forcing (namely, the surface fluxes), and the resulting analysis provides “corrected” surface forcing consistent with the ocean analysis. The long-term average of corrected surface forcing provides an estimate of mean errors of the atmospheric forcing. Their long-term ocean reanalysis showed there is a significant error in the atmospheric forcing field. Since the atmospheric forcing is a combination of surface winds, near-surface temperature, surface radiation fluxes, and SST, the errors in the forcing are not necessarily the results of error in the SST analysis. Many errors can be explained by problems in the model cloudiness, but some errors are related to uncertainties in the SST analysis used in the reanalysis.

The SST analysis affects the accuracy of the estimate of latent and sensible heat fluxes over ocean as mentioned above. The resulting surface fluxes influence model precipitation and eventually the large-scale flow fields in the guess field. Thus, the impact of the accuracy of SST to resulting atmospheric analysis may be much greater for analysis before the 1950s, for which very few atmospheric and surface observations are available. The impact will be even greater for pre-radiosonde years (before the early 1940s), for which the analysis must be performed using surface observation only.

The Atmospheric Model Intercomparison Project (AMIP) (Gates et al. 1999) demonstrated that current general circulation models are capable of simulating three-dimensional structures of seasonal mean global atmosphere and their interannual variations reasonably well with SST forcing alone, especially when the tropical SST anomaly is large. These simulations clearly indicated that SST can be a significant source of information for analyzing the seasonal mean three-dimensional structure of atmosphere.

The purpose of this paper is to study the importance of SST in reanalysis using observation system experiments. Major focus will be placed on the impact of SST on daily and seasonal mean fields of surface pressure, surface temperature, upper-air fields, latent heat flux, cloudiness, and precipitation. Recently, Bengtsson et al. (2004) performed observation system experiments using the ERA-40 reanalysis system to estimate the accuracy of analysis for the surface-based, terrestrial-based, and satellite-based systems, roughly corresponding to the pre-radiosonde, presatellite, and satellite eras. They concluded that the radiosonde observation is essential in obtaining reasonably accurate analysis useful for climate application. This paper also looks at some aspects of the importance of radiosonde and satellite observations.

The paper is organized as follows: section 2 presents data and the experimental analysis procedure, section 3 presents the impact of SST and surface pressure observation for daily analysis, and section 4 for seasonal mean analysis. Section 5 describes the impact of radiosonde observation on the analysis of both time scales. Section 6 presents the impact of SST and observations on diagnostic fields, and section 7 describes the geographical distribution of the seasonal mean analysis differences. Summary and conclusions are presented in section 8.

2. Data and experimental analysis procedures

We utilized the NCEP/DOE AMIP-II reanalysis-2 (R-2; Kanamitsu et al. 2002a) Climate Data Assimilation System (CDAS). The forecast model physics are taken from the NCEP Seasonal Forecast Model (Kanamitsu et al. 2002b), and differ from the ones used in R-2, but details are omitted here. It should be noted that this CDAS 3D variational assimilation is not a state-of-the-art system. For example, it does not utilize satellite radiance, but instead uses independently retrieved temperature. Over land, surface pressure observation is used, but the wind, temperature, and humidity observations were discarded. Over ocean, all the surface observations were used. In addition, the latest satellite data, such as Special Sensor Microwave Imager (SSM/I) observations, are not used in the analysis system.

We performed five pairs of experiments with observed and climatological SSTs, the first pair without any observation (AMIP and AMIPC), the second pair with surface pressure only with the observation density of 1915 (PRSX and PRSXC), the third pair with surface pressure only and observation density of 1997 (PRS and PRSC), the fourth pair with surface pressure and radiosonde (SONDE and SONDEC), and the fifth pair with all the observations (CNTL and CNTLC). The climatological SST is a 30-yr mean of the SST analyses used in this study. These experiments are summarized in Table 1. We assumed the results of the analysis with all the observations with observed SST (CNTL) to be true, and all other experimental analyses were compared against it. Note that one of the first pairs of experiments, observed SST without observation, is similar to the AMIP-type runs, in which the atmospheric model is forced only by SST. The run with climatological SST without observation (AMIPC) does not have interannual variability in the external forcing; thus the simulation represents only the atmospheric internal variability. We presume that the AMIPC deviates most from the control analysis, since no information that differentiates from one year to another is contributed to the system.

The quantity of surface pressure data for the PRS and PRSC experiments numbers about 10 000 during ±3 h from the analysis time over the whole globe. The distribution of full surface pressure observations is shown in Fig. 1. The observation distribution of early 1915 was obtained using the World Monthly Surface Station Climatology and Global Historical Climatology Network (GHCN) available from the National Climatic Data Center (NCDC) Web site (Fig. 2). Detailed procedures to simulate the distribution of observation used for this experiment are described in appendix A.

The majority of the assimilation experiments were performed for the period of 7 November 1997 to 28 February 1998 with the November analysis discarded to avoid the spinup problem. Note that the winter of 1997/98 is a strong El Niño year (Barnston et al. 1999). Additionally, experiments for the same season but for the near-normal SST year, 1993, were also performed. The definition of near-normal SST is based on the Climate Prediction Center Web page (http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.html). Initial guess data at the beginning of the analysis and land surface conditions are taken from R-2.

The choice of five observation systems is somewhat arbitrary; it is designed to examine the impact of observation coverage before the radiosonde era, before the satellite era, and during the satellite era. We made no attempt to change the forecast error statistics or the observational errors in the analysis system, as this complicates the interpretation of the results. Since forecast error will increase when a smaller number of observations are injected, this approximation will place more weight on model forecast guess, and the fit of analysis to observation will deteriorate.

The use of climatological SST will amplify the impact of SST, particularly for the 1997/98 case. The experiment for 1993 is additionally performed to examine the impact of SST in more normal years. The difference of observed and climatological SSTs for 1993 and 1997 are shown in Fig. 3. Apparently, this difference is larger than the SST analysis error noted earlier. As a relative measure of the magnitude of SST difference, we computed the root-mean-square difference (rmsd) of two independent SST analyses, one from the U.K. Met Office, Hadley Centre Ice and Sea Surface Temperature (HadISST), V1.1 (Rayner et al. 2003) and the other from NCEP, Extended Reconstructed Sea Surface Temperature (ERSST) V2 (Smith and Reynolds 2004), and compared it with the rmsd of SST anomaly in 1993 and 1997. Roughly speaking, the rmsd of SST anomaly was 2 times as large as the difference of two independent SST analyses during 1870–1940 in the Northern Hemisphere (NH) (1.4 versus 0.75 K), 3 times as large in the Tropics (1.6 versus 0.45 K), and 2.5 times as large in the Southern Hemisphere (SH) (1.3 versus 0.55 K). Thus, the impact of SST in our experiment is exaggerated. However, since we attempted to examine the effect of SST on the three-dimensional structure of the atmosphere as well as on physical processes diagnosed in the data assimilation, and the experiments are performed only for two winters, the exaggerated SST difference was intentional to obtain a stronger and more meaningful signal.

For the assimilations using surface-only observation, and for the runs without any data (but using specified SST) in 1997, simplified ensemble analysis/simulation was performed. Five-member ensemble runs were made by randomly perturbing the initial guess at the very first assimilation cycle analysis. The five separate analysis cycles were performed and the ensemble members were averaged to obtain the final analysis. The advantage of this simple ensemble data assimilation is briefly described in appendix B. Note that the additional experiment for the near-normal SST year, 1993, is performed in a single member mode to conserve computer resources.

When comparing analyses with different observation systems but with observed SST (or with climatological SST), we expect that the addition of an observation system (or increase in the number of observations) makes the analysis fit closer to the control. This can be used to check the reliability of our experiments. From the pairs of experiments using observed and climatological SST, we can determine that if the analysis with observed SST closely matches the analysis with climatological SST for the same observation system, the SST has a positive impact on the analysis. If the fit of the analysis that uses a reduced observation system but with observed SST is better than the increased observation system with climatological SST, we can conclude that the use of observed SST is more beneficial than the change in the observation system. However, it should be noted that since the difference between observed and climatological SST is large, particularly for the 1997 case, the impact of SST does not necessarily represent a real-world scenario. In this respect, the experiment for the normal SST year, 1993, may provide somewhat more reasonable measures of both the impact of SST and changes in the observation system. It is also important to mention that the choice of observation system is somewhat arbitrary, so that the comparison of the effect of SST and the change in observation system should be taken as simply a relative measure. To make more quantitative results, multiyear runs with more realistically perturbed SST (e.g., using SST analysis error) is required, but such experiments are beyond our capability at this time.

3. Impact of SST and surface pressure observation on daily analyses

For the sake of clarity, we will first focus on the impact of SST and surface pressure observation in the current and next sections. The impact of radiosonde observation will be described separately in section 5.

a. Impact on surface pressure analysis

Figure 4a compares the rmsd normalized by the temporal variance between nine experimental analyses with the control analysis of daily mean sea level pressure (MSLP) averaged for three months in 1997 (left column) and 1993 (right column). The temporal variance is computed from 3-month daily control analysis. The unnormalized rmsd for 1997 is shown in Table 2a. Except in the Tropics, the rmsds of MSLP analysis from CNTL decrease in the order of AMIPC, AMIP, PRSXC, PRSX, PRSC, PRS (from left to right in the figure). The unnormalized rmsd ranges from more than 10 hPa for AMIPC to 2.7 hPa for PRS in the extratropics. The rmsd is less than 3 hPa in the Tropics (Table 2a). Figure 4a shows that the decrease of MSLP rmsd between observed and climatological SST in the extratropics is smaller than the decrease in rmsd between analyses with different surface-pressure-observation density. This is seen in the pair of bars, with the right bar slightly lower than the left, and the next pair separated by a large step. Considering that the impact of SST is small even though the magnitude of the SST difference is exaggerated, the impact of SST on MSLP analysis is less than the impact of adding surface pressure observation.

In the Tropics, the impact of SST on MSLP analysis is as large as the impact of adding surface pressure observation. The impact of SST is much greater in 1997 than in 1993, due to the larger SST anomaly in 1997. Most of the results noted above hold for 1993, with the smaller impact of SST on surface pressure analysis as expected (Fig. 4c). The rmsd is larger for the AMIP, PRSX, and PRS experiments than those in 1997. In addition to the magnitude of the tropical SST anomaly, the use of the single-member analysis is contributing to this difference.

b. Impact on 2-m temperature analysis

Although the SST is not directly used by the analysis system, the SST information is passed to the seasonal mean 2-m temperature (T2M) through model guess. From Fig. 4b, we can see that the greatest impact of SST on T2M is in the tropical oceans where the impact is much greater than adding the surface pressure observation, as seen in the sawtooth pattern. The unnormalized rmsd for 1997 is also shown in Table 2a. The rmsd ranges from 6 to 1.0 K, slightly larger than the SST anomaly itself. The impact is smaller in 1993, but still noticeable (Fig. 4d). We also observe that over the tropical ocean, the surface pressure observations do not improve the T2M analysis, while they clearly do in the extratropics. Over tropical land, the impact of SST is still greater than that of surface pressure observation in 1997. In the extratropical oceans, use of observed SST reduces the T2M rmsd more than adding the surface pressure observations.

c. Impact on upper-air field analysis

The difference in the analysis near the surface should influence the analysis in the free atmosphere through hydrostatic relationship. The near-surface information also propagates upward during the data assimilation cycle, modifying the analysis at upper levels through guess field. These processes affect the free-atmospheric analysis especially when upper-air observation coverage is scarce. The impact of surface pressure observation and SST on free-atmospheric analysis is of particular interest in the pre-radiosonde era, when observation cannot directly influence the upper-level analysis.

In Table 2b, the rmsd of 500-hPa geopotential height is presented. Quantitatively, in the NH extratropics, the surface-pressure-only analysis (PRS) reduced the daily 500-hPa height (Z500) rmsd from 111 m in AMIP runs to just 49 m (Table 2b, highlighted in bold). The PRS analysis is as accurate as a 2–3-day forecast of a respectable numerical forecast model. This remarkable capability of the data assimilation system was first demonstrated by Whitaker et al. (2004). For the reduced number of surface pressure observations (PRSX), the rmsd of Z500 increased to 91 m, still better than AMIP, but considerably worse than PRS (Table 2b). The rmsd of PRSX (91 m) is worse than the 64 m from the similar experiment performed by Whitaker et al. (2004, shown in their Fig. 3). The degradation in our analysis is due to fewer surface pressure observations, a different analysis period, and other tunings on the analysis system not made in our study. It should be emphasized that Whitaker et al. (2004) utilized the ensemble Kalman filter technique to further improve the surface-pressure-observation-only analysis for the data coverage of 1915. With their method, the rmsd was reduced to 40 m, a considerable improvement. In our experiment, increasing the number of surface observations by several factors (as in the PRS experiment) in conventional data assimilation seems to mimic the ensemble Kalman filtering analysis technique.

The SST has a fairly significant impact on the extratropical upper-level analysis. For the PRS experiment, the use of climatological SST increases the rmsd in the NH extratropics by 14 m (from 49 to 63 m), and by 11 m (from 55 to 66 m) in the SH. The importance of SST in the Tropics is much more apparent, which can be seen in that the AMIP run rmsd is smaller than that of the PRSC run. This is more clearly demonstrated in the normalized rmsd (Fig. 5c). The use of climatological SST resulted in very large rmsd over the entire Tropics. Closer examination of the rmsd indicated that the bias accounted for more than 90% of the rmsd (not shown). This height bias is caused by the bias in temperature, to be discussed later. The impact of SST increases with height in the extratropics, as the impact of SST in extratropical latitudes is more apparent at 200 hPa than at 500 hPa (Fig. 5a). For 1993, the impact of SST in the NH extratropics is not as apparent, but is still significant in the Tropics (Figs. 5b,d).

4. Impact of surface pressure observation and SST on monthly and seasonal mean analyses

In this section characteristics of the impact of SST and surface pressure observation on seasonally averaged parameters are described. The impact of SST was found to be very different from daily time scale in the free atmosphere. This is one of the key findings of this paper.

a. Impact on surface pressure analysis

Figure 6 presents normalized rmsd of surface pressure for seasonal (3 month) average, shown in a similar manner as Fig. 4. The rmsd is normalized by the variance of seasonal mean computed from 20-yr reanalysis data. Many of these features are similar to the daily figures (Fig. 4a), except that the impact of SST is more enhanced in the seasonal figures, particularly over NH ocean and land. This enhancement is less for 1993. Clearly the analysis of surface pressure in the extratropics is responding to the tropical SST anomaly in this time scale.

b. Impact on 2-m temperature analysis

The normalized T2M rmsd (Fig. 6b) is drastically different from daily figures (Fig. 4b) in the extratropics. The difference increases as the time-average period increases from monthly to seasonal (not shown). Over the ocean in particular, the impact of SST is significant in the seasonal mean. At the same time, the difference between AMIP and analysis with observations decreases. Particularly noteworthy is that in 1997, the AMIP T2M is as accurate as the analysis with the full observation system with climatological SST. Apparently, the impact of observation is different for different time scales of the analysis. For the non-ENSO year (Figs. 6c,d), the impact of SST is generally smaller but still significant.

c. Impact on upper-level field analysis

The rmsd of 500-hPa height in the NH (Fig. 7b) does not decrease from the AMIP, PRSX, and PRS experiments, indicating that the surface pressure observation does not have a very strong influence on seasonal mean 500-hPa analysis. Alternatively, the SST drastically decreases the rmsd for the same observation system. This is in sharp contrast to the role of surface pressure observation and SST on daily 500-hPa analysis (Fig. 5b), where the surface pressure observation reduces the rmsd much more. Table 2b shows that the rmsd for daily analysis decreased from 111 for AMIP to 49 m for PRS, that is, a 62-m decrease using surface pressure observation, and the maximum decrease of rmsd is from 105 for PRSXC and 91 for PRSX, a 14-m decrease, using observed SST. In comparison, Table 3 shows that rmsd does not decrease from AMIP to PRS, but it decreases 20 m from PRSC and PRS. Although the rmsd of daily and seasonal averages cannot be compared directly, it is very clear that SST dominates the analysis of seasonal mean upper-level height field, while surface pressure observation dominates for daily analysis. Apparently, the high-frequency observations (synoptic surface pressure observation) provide the greatest impact on high-frequency analysis (daily), while low-frequency observations (monthly averaged SST interpolated to daily) provide the greatest impact on low-frequency analysis (seasonal).

It is important to ask why the data assimilation used in this study (3D variational analysis scheme) filters out low-frequency signal from surface pressure observation for upper-level analysis. It is also important to examine whether a different analysis system has different filtering characteristics. Since surface pressure is a total mass of the air column, and it is strongly influenced by the external gravity wave, the balance equation constraints built into the analysis system (Parrish and Derber 1992) may have some role in this filtering property. In this context, it is of great interest to examine the time scale of information included in various observation systems, such as radiosondes, satellite observation, aircraft observation, and conventional surface observation and how they are filtered out in the analysis forecast system, namely through the forecast model and the objective analysis scheme. In addition, whether this result holds for other analysis systems, such as the more advanced ensemble Kalman filter method, is not clear at this point. We hope such a study will be performed by institutions and operational centers.

One reason for the strong dependency of the time mean analysis to SST obtained in this study may be due to the very large SST anomaly in the Tropics. However, the experiments for the near-normal tropical SST anomaly year, 1993, still show reduced magnitude but significant impact. For this year, the accuracy of the AMIP experiment is similar to or slightly worse than the analysis with full surface pressure observation with climatological SST. Thus, even for near-normal SST years, SST is important in determining the low-frequency part of the upper-level analysis (Fig. 7c).

Similar results are obtained for 200-hPa height (Fig. 7a) and 850- and 300-hPa temperature (not shown). We see in Fig. 7 that the role of SST tends to amplify with height. In the Tropics and in the SH, SST is again important, but surface pressure observation is also as important.

d. Bias

The large reduction of rmsd by the use of observed SST in the Tropics is mostly due to the reduction in the systematic part of the error, which amounts to more than 90% of total rmsd. The vertical distribution of the effect of SST and surface pressure observation on the zonally averaged temperature is shown in Fig. 8. The use of climatological SST (left column) produces cold bias in the tropical upper troposphere, extending to lower levels in the NH subtropics. This bias is caused by the difference in precipitation in the climatological SST and observed SST runs (see Fig. 13 later). The surface pressure observation almost entirely eliminates temperature bias in the Tropics and NH troposphere. The large positive bias in the NH polar region is replaced by a small cold bias resulting from the introduction of surface pressure observation.

The differences between observed SST runs and control analyses (right column) show a large cold bias in the stratosphere. Note that the control analysis uses satellite data in the stratosphere. Thus the surface pressure observation does not have any impact on analysis in the stratosphere, and even radiosonde cannot eliminate the cold bias as also shown by Bengtsson et al. (2004).

5. Impact of radiosonde observation

The impact of radiosonde on analysis is generally significant and overwhelms the effect of SST as seen in all the rmsd figures (Figs. 4 –7). However, close examination of the results suggests that we can still detect the impact of SST. For example, there is a small degradation in the analysis over land (about 0.8 hPa) using all observations with climatological SST (CNTLC) compared with the analysis with full observation but with observed SST (CNTL) (see Table 2b). The SST signal propagates from ocean to land in the data assimilation system and thus the inaccurate information over ocean also propagates to land and contaminates the analysis over land even when full observations are used.

6. Impact of SST and observation on diagnosed fields

The change in observation system affects model-diagnosed fields through the change in the analysis variables. We here present rmsd of latent heat flux (LHF), cloudiness (CLD), and precipitation (PRCP).

a. Impact on daily diagnostic fields

1) Latent heat flux

The unnormalized rmsd of LHF ranges from the lower 80s to about 20 W m−2 (Table 2c) among the experiments. The rmsd is smaller over land, probably due to the memory of soil moisture. The bar graphs of normalized LHF (Fig. 9a) show that the rmsd decreases with the use of observed SST as well as with the improved observation system, almost equally over the globe. The exception is the impact of SST on the full observation system over ocean (CNTLC), which is less inaccurate than observed SST with full surface observation and radiosondes (SONDE), showing the importance of SST, which cannot be compensated by the conventional observations. Interestingly, the improvement of latent heat flux with the use of observed SST is also found over NH land regions. Surface pressure observation improves latent heat flux in the extratropics, but not so much in the Tropics. This is consistent with the impact of surface pressure observation on T2M (Fig. 4b).

2) Cloudiness

Consistent improvement in cloudiness due to SST is observed over the globe, but the magnitude is small, of the order of 1% to 2% (Fig. 9b; Table 2c). The observation system also contributes to the improvement of clouds. Of particular interest is the drastic reduction (nearly 10%) in cloudiness rmsd over land in the NH as soon as the upper-air observations are introduced. This is most likely due to the bias in the model relative humidity, which is corrected by radiosondes over land. It is noted that the cloudiness in the R-2 CDAS system is diagnosed from relative humidity.

3) Precipitation

The response of precipitation to SST is the weakest of all parameters we examined. Precipitation is more affected by the change in observation system, about a 10% decrease in rmsd as the observation system is increased (Table 2c). We only see a small improvement in precipitation over tropical ocean by SST in both 1997 (Fig. 9c) and in the near-normal SST year, 1993 (Fig. 9d). Closer examination of the figures reveals that the SST has a small positive impact on extratropical land areas, which is more pronounced in the near-normal SST year (Fig. 9d). The greatest contribution to the decrease in rmsd is the radiosonde observation. The impact is the largest over NH extratropical land, as seen in the abrupt decrease of rmsd when radiosonde observations are used (Fig. 9c).

b. Impact on seasonal-averaged diagnostic fields

1) Latent heat flux

The impact of observation and SST on seasonal mean latent heat flux is shown in Fig. 10a. Compared to the daily rmsd (Fig. 9), the impact of SST is much larger in the NH extratropics. The responses in the NH resemble those of T2M (Fig. 6b). Over the NH extratropical oceans, the rmsd does not change much with the introduction of surface pressure observations, until the radiosonde observation is introduced. The seasonal mean latent heat flux in the extratropical ocean is controlled by SST more than by surface pressure observation. The importance of surface pressure observation is a little stronger over extratropical land. Over the SH extratropics, the impact of SST and surface pressure is about equal. In the Tropics, the impact of SST is weak and only apparent for SONDE experiments. The large rmsd for CNTLC (rightmost bar) indicates the importance of SST for latent heat flux over the globe, even when all the observations are used.

2) Cloudiness

As to seasonal mean cloudiness (Fig. 10b), the impact of surface pressure is small. We observe some impact of SST over the tropical ocean. The radiosonde data are essential in bringing the cloudiness down to the control analysis, which is particularly significant over extratropical land.

3) Precipitation

For the seasonal mean precipitation over NH extratropical land and over tropical ocean, the impact of SST is clear, but its magnitude is not large (Fig. 10c). Note that SST had no impact on daily precipitation (Fig. 9c). The impact of radiosondes is very clear over the globe. The impact of SST on the near-normal year, 1993 (Fig. 10d), is somewhat greater than that in the strong El Niño year, 1997. This can simply be an artifact of the use of single member analysis.

7. Geographical patterns of differences in seasonal mean field

a. MSLP

Figure 11 shows the differences of seasonal mean MSLP between the experiments. The analysis with surface-pressure-observations-only (PRS) analysis shows a significant SST impact in the Pacific, roughly corresponding to the area of SST difference (Fig. 3). There is a clear anticorrelation between the difference in SST and that of the surface pressure analysis. The impact of SST can be seen even for CNTL analysis over the El Niño region, where the SST difference reaches 4°–5°C, and surface observation is scarce (Fig. 1). Comparing the analysis with control for observed SST experiments (right column), the differences are generally larger for AMIP and PRSX, particularly over ocean. There is a marked similarity in the patterns between AMIP and CNTL and between PRSX and CNTL. The difference of the PRS experiments is much smaller, simply indicating that the data assimilation system is properly incorporating the surface-pressure-observation information into the analysis, and the influence of SST is clear where the surface pressure observation is absent.

b. 500-hPa height

The geographical distributions of the difference of 500-hPa height between various analyses are shown in Fig. 12. The top-left panel shows the difference between AMIPC and AMIP, which is the response of the atmosphere to observed SST. The panels below are the response pattern modified by the injection of surface pressure observation. The use of all the surface pressure observations eliminated differences located at the west coast of the United States, northeastern North America, and Europe. Even the tropical dipole pattern is reduced. On the right column, we see that the difference of each observation system is independent from the error using climatological SST. The patterns of differences in the NH vary somewhat from the AMIP to PRS experiments. Note that some of the difference increases as more surface pressure observations are added. The areas of large difference in the PRS experiment roughly coincide with the areas of coarse surface pressure observations (Fig. 1). In the SH, the difference pattern does not change much, due to the small number of surface pressure observations available.

c. Precipitation

The geographical pattern of the impact of SST and observation on model-diagnosed precipitation is shown in Fig. 13. The difference is larger and extends to the extratropics for SST impact experiments (left column), while the impact of the observation system (right column) is smaller and more concentrated in equatorial latitudes. This is in sharp contrast to the rmsd of daily precipitation (not shown), for which the rmsd is larger and extends into the extratropics for observation system experiments.

8. Conclusions

The importance of SST in reanalysis is examined using the NCEP/DOE reanalysis system. The winter season in the strong El Niño year of 1997 as well as in the near-normal SST anomaly year of 1993 were chosen for the experiment. The analyses with climatological and observed SST were performed and compared for no observation, surface-pressure-only observation of 1915 observation distribution, surface-pressure-only observation of 1997 observation coverage, surface pressure and radiosonde observation, and complete observation for 1997. The impact of SST and observation systems was measured using the rmsd of analysis from control analysis with observed SST and the complete observation system.

The most important finding of this paper is that the effect of SST and the observation system is different for different time scales. It is found that low-frequency observation, such as SST, provides greatest impact on low-frequency analysis, such as the seasonal mean. On the other hand, high-frequency observation, such as synoptic surface pressure observation, provides the largest impact on high-frequency analysis, namely, daily analysis. The latter does not necessarily provide information on low-frequency analysis. This is most clearly demonstrated in T2M and upper-level height field analysis in the extratropics. Whether this finding is characteristic only to the data assimilation system used in this study is uncertain and needs to be examined using other data assimilation systems, including those using the latest techniques, such as the ensemble Kalman filter.

More concise conclusions of the impact of SST, surface pressure observation, and upper-air observations for the analysis in the NH extratropics are summarized in Table 4. It shows the qualitative importance of each observation system (top row) on the analysis of variables shown in the leftmost column, separated for daily (high frequency) and seasonal mean (low frequency) analyses. Note that the table is based on a somewhat subjective judgment and only shows results for the NH extratropics. It does not differentiate land from ocean. Thus, the table should be used for a rough estimate of the importance of SST, surface pressure, and radiosonde observations in the data-rich extratropics. The table shows that SST has an important effect on the analysis of all the parameters in daily and seasonal time scales, but its importance increases for seasonal mean. The SST is also very important in obtaining a good estimate of diagnostic quantities such as surface fluxes, cloudiness, and precipitation. Alternatively, surface pressure observations are more important for daily analysis, but have little or no effect on seasonal average. They are also not so important for the diagnostic quantities. The radiosonde observations have the greatest impact on all the parameters for both daily and seasonal mean analyses, and are the essential source of observation for accurate analysis of all the parameters.

In the Tropics, the impact of SST and the observation system is about equal for daily and seasonal mean analysis. This is probably due to less activity of high-frequency modes compared to the extratropics. When the SST observation is replaced with climatology, the impact of surface pressure observation becomes larger for many of the parameters examined in this study, but the improvement of analysis resulting from the pressure observation is always less than that from observed SST.

This study indicates that obtaining accurate analysis of all the parameters before the radiosonde era is difficult. The importance of SST observation increases considerably for the pre-radiosonde era, particularly for the analysis of low-frequency modes, such as the monthly and seasonal mean. In addition, improving the analysis of derived quantities such as cloudiness and precipitation is even more challenging, since these parameters are not strongly affected by surface pressure, or only weakly affected by SST observations. Considering the importance of SST in the analysis of low-frequency modes, it is strongly recommended that the SST be analyzed as part of the objective analysis scheme. Since surface pressure, near-surface temperature, moisture, and even winds are statistically (or dynamically and thermodynamically) correlated with SST, such a method may not be difficult to develop. We also hope that the use of advanced analysis technique, such as the ensemble Kalman filter, improves this situation. This study also suggests that the analysis of low-frequency modes may require special consideration and research.

Acknowledgments

We sincerely thank Professor Song-You Hong of Yonsei University for enabling one of the authors to visit SIO to perform this research. We also thank Jack Woollen and Wesley Ebisuzaki for assisting with the data assimilation experiments. Dr. Tosiyuki Nakaegawa provided information on the difference of two independent SST analyses. We also thank Dr. John Roads for his continuous encouragement throughout this study. Diane Boomer contributed in improving and making the paper much more readable. This project is supported by Cooperative Agreement NOAA-NA17RJ1231 and by the project, “Development of the technology for the improvement of medium-range forecast,” one of the Research and Development on Meteorology and Seismology funded by the Korea Meteorological Administration (KMA).

REFERENCES

  • Arkin, P., , E. Kalnay, , J. Laver, , S. Schubert, , and K. Trenberth, 2003: Ongoing analysis of the climate system: A workshop report. Rep. 18020, UCAR, Boulder, CO. [Available online at http://www.joss.ucar.edu/joss_psg/meetings/climatesystem/FinalWorkshopReport.pdf.].

  • Barnston, A. G., , A. Leetmaa, , V. E. Kousky, , R. E. Livezey, , E. A. O'Lenic, , H. Van den Dool, , A. J. Wagner, , and D. A. Unger, 1999: NCEP Forecasts of the El Nino of 1997–98 and its U.S. impacts. Bull. Amer. Meteor. Soc, 80 , 18291852.

    • Search Google Scholar
    • Export Citation
  • Bengtsson, L., , K. I. Hodges, , and S. Hagemann, 2004: Sensitivity of the ERA40 reanalysis to the observing system: Determination of the global atmospheric circulation from reduced observations. Tellus, 56A , 456471.

    • Search Google Scholar
    • Export Citation
  • Chelliah, M., , and C. F. Ropelewski, 2000: Reanalyses-based tropospheric temperature estimates: Uncertainties in the context of global climate change detection. J. Climate, 13 , 31873205.

    • Search Google Scholar
    • Export Citation
  • Gates, W. L., and Coauthors, 1999: An overview of the results of the Atmospheric Model Intercomparison Project (AMIP I). Bull. Amer. Meteor. Soc, 80 , 2955.

    • Search Google Scholar
    • Export Citation
  • Gibson, J. K., , P. Kallberg, , S. Uppala, , A. Nomura, , A. Hernandez, , and E. Serrano, 1997: ERA description. ECMWF Re-Analysis Final Report Series 1, Shinfield Park, Reading, Berkshire, United Kingdom, 71 pp.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc, 77 , 437471.

  • Kanamitsu, M., and Coauthors, 2002a: NCEP Dynamical Seasonal Forecast System 2000. Bull. Amer. Meteor. Soc, 83 , 10191037.

  • Kanamitsu, M., , W. Ebisuzaki, , J. Woolen, , J. Potter, , and M. Fiorino, 2002b: NCEP/DOE AMIP-II reanalysis (R-2). Bull. Amer. Meteor. Soc, 83 , 16311643.

    • Search Google Scholar
    • Export Citation
  • Kistler, R., and Coauthors, 2001: The NCEP–NCAR 50-year reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc, 82 , 247268.

    • Search Google Scholar
    • Export Citation
  • Parrish, D. F., , and J. C. Derber, 1992: The National Meteorological Center's Spectral Statistical–Interpolation Analysis System. Mon. Wea. Rev, 120 , 17471763.

    • Search Google Scholar
    • Export Citation
  • Rayner, N. A., , D. E. Parker, , E. B. Horton, , C. K. Folland, , L. V. Alexander, , and D. P. Rowell, 2003: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res, 108 .4407, doi:10.1029/2002JD002670.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., 1988: A real-time global sea surface temperature analysis. J. Climate, 1 , 7586.

  • Reynolds, R. W., 1993: Impact of Mount Pinatubo aerosols on satellite-derived sea surface temperatures. J. Climate, 6 , 768774.

  • 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
  • Servain, J., , M. Seva, , and P. Rual, 1990: Climatology comparison and long-term variations of sea surface temperature over the tropical Atlantic Ocean. J. Geophys. Res, 95 , 94219431.

    • Search Google Scholar
    • Export Citation
  • Simmons, A. J., , and J. K. Gibson, 2000: The ERA-40 project plan. ERA-40 Project Report Series, ECMWF, Shinfield Park, Reading, Berkshire, United Kingdom, 62 pp.

  • Smith, T. M., , and R. W. Reynolds, 2004: Improved extended reconstruction of SST (1854–1997). J. Climate, 17 , 24662477.

  • Smith, T. M., , R. W. Reynolds, , R. E. Livezey, , and D. C. Stokes, 1996: Reconstruction of historical sea surface temperatures using empirical orthogonal functions. J. Climate, 9 , 14031420.

    • Search Google Scholar
    • Export Citation
  • Stammer, D., and Coauthors, 2002: Global ocean circulation during 1992–1997, estimated from ocean observations and a general circulation model. J. Geophys. Res, 107 , 127.

    • Search Google Scholar
    • Export Citation
  • Sturaro, G., 2003: A closer look at the climatological discontinuities present in the NCEP/NMCAR reanalysis temperature due to the introduction of satellite data. Climate Dyn, 21 , 309316.

    • Search Google Scholar
    • Export Citation
  • Whitaker, J. S., , G. P. Compo, , X. Wei, , and T. M. Hamill, 2004: Reanalysis without radiosondes using ensemble data assimilation. Mon. Wea. Rev, 132 , 11901200.

    • Search Google Scholar
    • Export Citation

APPENDIX A

Observation Distribution for the 1915 Experiment

The total number of land stations in 1915 was around 380. Not all surface stations took observation at all times. The average number of ocean observations was 41 h−1. Using the observation density of 1000 day−1 during the early twentieth century, estimated by Whitaker et al. (2004), we assigned about 200 stations per each analysis time, 50 of which are located over ocean. Land observations are extracted from stations that coincide with land stations in Fig. 1. Oceans are divided into six areas and observations in each area are randomly selected according to the observation density shown in the right panel of Fig. 2. For example the Atlantic has 30% of global observation, while the Southern Indian Ocean has 5%. An example of simulated distribution (reduced 1997 network) used to run the 1915 experiment is shown in Fig. A1. Approximately 200 land stations are nearly fixed because they are selected from 380 predefined 1915 station networks. The number of both land and ocean observations are reduced to 170 ∼ 220 (average number 200) after the data quality control procedure.

APPENDIX B

Simple Ensemble Data Assimilation

Simple ensemble analysis is performed to minimize uncertainty in the analysis resulting from uncertainty in the model initial guess. The method is to extend the ensemble AMIP runs methodology to analysis by randomly perturbing the initial guess of the first analysis of the data assimilation cycle. We performed five-member ensemble analyses. The analysis for each cycle is obtained by the ensemble averages. Figure B1 shows daily rmsd of 500-hPa height for each analysis from control as well as the rmsd of the ensemble mean. Clearly, the ensemble mean reduces the rmsd by as much as 20 m using this simple method.

Fig. 1.
Fig. 1.

The distribution of observation sites for surface pressure at 0000 UTC 7 Nov 1997

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 2.
Fig. 2.

Simulated Psfc data distribution of (left) land and (right) ocean extracted from surface observation data on 7 Nov 1997

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 3.
Fig. 3.

Seasonal mean SST anomaly for (top) 1993 and (bottom) 1997

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 4.
Fig. 4.

Time-averaged normalized rmsd of daily mean (top) sea level pressure and (bottom) T2M for (left) 1997 and (right) 1993. For each panel, top portion is over ocean, middle is over land, and bottom is ocean plus land. Rmsd is normalized by the daily temporal variance

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 5.
Fig. 5.

Time-averaged normalized rmsd of daily mean (top) 200- and (bottom) 500-hPa height, for (left) 1997 and (right) 1993 columns. For each panel, top portion is over ocean, middle is over land, and bottom is ocean plus land. Rmsd is normalized by the daily temporal variance

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 6.
Fig. 6.

Rmsd of seasonal average (top) MSLP and (bottom) T2M for (left) 1997 and (right) 1993

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 7.
Fig. 7.

Rmsd of (bottom) 500- and (top) 200-hPa height of seasonal mean for (left) 1997 and (right) 1993

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 8.
Fig. 8.

Difference of zonally averaged temperature (°C)

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 9.
Fig. 9.

Time-averaged normalized rmsd of daily mean (a) latent heat flux, (b) total cloudiness, (c) precipitation for 1997, and (d) precipitation for 1993. For each panel, top portion is over ocean, middle is over land, and bottom is ocean plus land. Rmsd is normalized by the daily temporal variance

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 10.
Fig. 10.

Rmsd of seasonal mean (a) latent heat flux, (b) cloudiness, (c) precipitation for 1997, and (d) precipitation for 1993

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 11.
Fig. 11.

Difference of seasonal mean MSLP

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 12.
Fig. 12.

Difference of seasonal mean 500-hPa height

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Fig. 13.
Fig. 13.

Difference of seasonal mean precipitation

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

i1520-0493-134-2-532-fa01

Fig. A1. Monthly distribution of (a) land surface station and (b) observation point at ocean in November 1915

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

i1520-0493-134-2-532-fb01

Fig. B1. Daily rmsd of 500-hPa height of individual members and ensemble mean (think soil line) from control analysis (m)

Citation: Monthly Weather Review 134, 2; 10.1175/MWR3084.1

Table 1.

Summary of experiments configuration

Table 1.

Table 2a. Rmsd of daily analysis from control analysis averaged for three months (1997). NH stands for latitudinal band of 30°–90°N, EQ for 30°S–30°N, and SH for 90°–30°S

i1520-0493-134-2-532-t0201

Table 2b. Rmsd of daily analysis of 500-hPa height from control analysis averaged for three months (1997)

i1520-0493-134-2-532-t0202

Table 2c. Rmsd of daily analysis from control analysis averaged for three months (1997)

i1520-0493-134-2-532-t0203
Table 3.

Rmsd of seasonal mean analysis from control analysis (1997)

Table 3.
Table 4.

Impact of SST, surface pressure, and radiosonde observations on analysis in the Northern Hemisphere extratropics, where 0 is no impact, + is marginal impact, and ++ is strong impact

Table 4.
Save