Intercomparison of the NCEP/NCAR and the NASA/DAO
4.1 Zonal Mean Fields
The upper panels of Figs. 3, 4, 5 show latitude-time diagrams of the zonal mean zonal (u) and meridional (v) winds at 200-hPa and the zonal mean 500-hPa height field in the NCEP (for convenience we use NCEP rather than NCEP/NCAR in the discussion below) and the DAO reanalyses. Each figure is based on monthly data for the period March 1985-November 1993. The differences (NCEP-DAO) are shown in the bottom panels. The differences in the zonal wind are generally quite small, especially when compared to the magnitude of the u-wind itself. Throughout the period the differences tend to be negative along the equator (stronger easterlies in NCEP) and in the high latitudes of each hemisphere (weaker westerlies in NCEP) and positive in the subtropics of each hemisphere (stronger westerlies in NCEP). Differences near 40°S - 50°S are likely related to a zonal mean bias in the DAO. Differences in the zonal mean v are much larger in a relative sense. In many of the years the boreal winter Hadley cell is more vigorous in the NCEP (positive differences along and just north of the equator) while the boreal summer cell is more vigorous in the DAO (also contributes to positive differences along the equator). The differences in the height field are generally small except in the high latitudes of the SH.
Figures 6-25 show zonal mean cross-sections of various quantities for the two extreme seasons December-January-February (DJF) and June-July-August (JJA). The DJF figures are based on an eight-year (1985/86-1992/93) mean while the JJA figures are based on a nine-year (1985-1993) mean. The layout of the figures is such that the NCEP reanalysis is at the top, the DAO reanalysis is in the middle and the differences (NCEP-DAO) are in the bottom panel. In addition to the mean fields we also show the standard deviation (computed by removing the overall seasonal mean from the individual seasons); for the wind fields (u, v, omega) we also show comparisons to the Climate Diagnostics Data Base (CDDB) (see section 3).
Differences in the u-wind during DJF (Fig. 6) show a roughly symmetric pattern about the equator; the NCEP has stronger easterlies in the deep Tropics, stronger westerlies in the subtropics of each hemisphere (especially at upper levels) and weaker westerlies in middle and high latitudes. The variability in both NCEP and DAO compares quite well to CDDB (Fig. 7), though there is a tendency to underestimate the variability in the SH extratropics at upper levels. Similar results are found for JJA in the Tropics and SH (Fig. 8), but differences in the NH are quite small. The standard deviation for JJA (Fig. 9) shows considerable variability in the SH high latitudes in the CDDB which is not present in either of the reanalyses; this may be due to changes in the assimilation system.
The differences in the DJF Hadley cell mentioned earlier are evident in Fig. 10. In the tropical upper troposphere the NCEP zonal mean meridional (southerly) winds are stronger with the largest differences just south of the equator. In the tropical upper troposphere there is somewhat less variability in the reanalyses compared to the CDDB (Fig. 11). During JJA the tropical upper tropospheric meridional (northerly) winds are stronger in the DAO with the largest differences in the vicinity of the equator (Fig. 12). In addition, the NCEP upper-tropospheric cell extends a bit further into the SH contributing to negative differences near 15°S. Again we find less variability in the tropical upper troposphere in the reanalyses when compared to the CDDB (Fig. 13). The zonal mean temperature field (Figs. 16) show that the NCEP analyses are colder than the DAO throughout much of the troposphere (except at low levels) and warmer than the DAO above 100-hPa. The large differences near the surface are likely due to interpolation / extrapolation problems in regions of high topography. The patterns of variability (Figs. 17) are quite similar in both reanalysis products. The zonal mean vertical velocity patterns during DJF (Fig. 18) and JJA (Fig. 20) are consistent with the stronger boreal winter Hadley cell in the NCEP and stronger boreal summer cell in the DAO. In both seasons the variability in the reanalyses is smaller than in the CDDB (Figs. 21). The zonal mean specific humidity pattern during DJF (Fig. 22) shows that the NCEP is somewhat drier in the tropical upper troposphere and in the NH subtropics at low levels and somewhat moister in the SH subtropics at low levels. The difference pattern during JJA (Fig. 24) shows a more complicated structure with the NCEP generally drier in the NH Tropics and moister in the NH subtropics, the NH extratropics (below 850-hPa) and the SH subtropics (above 900-hPa). In the deep Tropics there is slightly less variability in the NCEP than in the DAO during DJF (Fig. 23) and JJA (Fig. 25).
4.2 Global Fields
Figures 26-67 show the global distribution of selected quantities from the NCEP and the DAO reanalyses along with difference maps. The layout of these figures is similar to that for the zonal means, with the NCEP results in the top panel, the DAO in the middle and the difference fields on the bottom. In addition, we provide comparisons of the standard deviation (computed by removing the overall seasonal mean from the individual seasons) in the reanalyses and in the CDDB and SSM/I (total precipitable water only); again the layout for these figures is similar to that for the zonal means.
The 200-hPa NCEP and DAO mean zonal winds are quite similar during both DJF (Fig. 26) and JJA (Fig. 30), with differences of less than 2 ms-1 in most places. The largest differences occur in the Tropics and at high latitudes in the SH during both seasons. In general, the NCEP shows stronger easterlies in the Tropics and weaker westerlies in the SH midlatitudes during both seasons; results for the NH are mixed. During DJF the variability is largest in the exit regions of the NH midlatitude jets (Fig. 27) while during JJA it is largest in the exit region of the South Pacific jet (Fig. 31). At 850-hPa the largest difference in the zonal wind occurs in the Tropics during DJF (Fig. 28) with stronger easterlies in the NCEP. The standard deviation patterns for DJF (Fig. 29) and JJA (Fig. 33) show many similarities to those at 200-hPa but with smaller amplitudes.
The meridional wind differences (NCEP-DAO) show more complicated patterns than the zonal wind. During DJF, the 200-hPa mean meridional winds in the Tropics (Fig. 34) show stronger southerlies over northern South America, Africa and the maritime continent in NCEP, consistent with a more vigorous Hadley circulation during northern winter. During JJA the Hadley circulation is slightly stronger in the DAO (Fig. 38), as evidenced by the stronger northerlies over the Tropics and subtropics of the SH. In both seasons the largest differences are in the mean meridional winds are in the high latitudes of the SH, and appear to be due to problems with the winds near the pole in the vicinity of high topography. At low levels (Figs. 36 and 40), the differences tend to be tied to low-level topographic features. For the most part, the variability found in the CDDB appears to be well captured in the reanalyses (Figs. 37 and 41).
The 500-hPa eddy height field shows the largest differences in the middle and high latitudes of the SH during both DJF (Fig. 42) and JJA (Fig. 44). In the DAO there is evidence of noise near the poles in both seasons; such noise also shows up in various other quantities including sea-level pressure (Figs. 58 and 60). The variability found in the CDDB is well reproduced in the reanalyses (Figs. 43 and 45).
Both the eddy streamfunction and velocity potential fields (Figs. 46-57) show the largest differences over the Tropics and subtropics (30°S - 30°N) except for the eastern North Pacific and western North America in the streamfunction field. Figure 50 shows clearly that the NCEP tends to produce stronger convection over the western and central tropical Pacific, consistent with a more vigorous Hadley cell, and over South America. Also, there is stronger subsidence in the eastern tropical oceans in the NCEP, consistent with a more intense tropical east - west (Walker) circulation in NCEP. The JJA velocity potential differences (Fig. 54) continue to suggest stronger convection over the maritime continent and South America in the NCEP, but differences in the Walker circulation have decreased.
The sea-level pressure differences (Figs. 58 and 60) largely follow the regions of high orography suggesting differences in the algorithm used to reduce the surface pressure to sea-level. There are also very large-scale (but small amplitude) differences with the NCEP sea-level pressure being lower than that from DAO over much of the Pacific, Atlantic and Indian Oceans during both seasons. The patterns of variability are quite similar in both reanalyses (Figs. 59 and 61), with the largest variability near the preferred locations of large amplitude persistent atmospheric flow anomalies.
During DJF the largest differences (NCEP-DAO) in the total precipitable water (Fig. 62) are found in the Tropics and subtropics with positive values in the eastern Pacific, eastern Atlantic and eastern Indian Oceans and with negative values over the western and central Pacific, maritime continent, south America and Africa. The Tropics and subtropics over the oceans are too dry compared with the vertically integrated moisture from SSM/I (Fig. 63). Overall, the horizontal moisture gradients between very moist and very dry regions are too weak. The variability in the reanalyses appears to be underestimated compared to SSM/I (Figs. 64 and 67) especially in the deep Tropics. During JJA the difference (NCEP-DAO) pattern is similar to that during DJF, but with larger amplitudes over the eastern Pacific (Fig. 65). The dry biases with respect to SSM/I show a clear shift towards the North Pacific (Fig. 66) consistent with the seasonal cycle.
4.3 Surface Energy Balance
This section presents the global distribution of the components of the surface energy balance from the NCEP, the DAO and COADS (ocean only) for boreal winter and boreal summer. In addition to the individual terms in the balance, we also show the standard deviation. The net surface energy balance can be expressed by the following equation:
Fsw - Fsw - (Flw - Flw ) - SH - LH = Qnet
where the Fsw terms are the upward and downward shortwave fluxes, the Flw terms are the upward and downward longwave fluxes, SH is the sensible heat flux, LH is the latent heat flux and Qnet is the net ground heat storage.
The latent heat flux depends chiefly on the surface wind magnitude, the air-sea moisture gradient, and the surface layer stability. Because of similarities in the techniques used to estimate evaporation in both reanalyses and in the COADS, any differences reflect differences in the near surface gradients of humidity or winds and surface roughness. During DJF (Fig. 68) there are maxima over the subtropical oceans in each hemisphere and over the SH continents. Fluxes in the vicinity of the Gulf Stream and Kuroshio appear to be slightly overestimated (underestimated) in the NCEP (DAO). From DJF to JJA (Fig. 70) there is a strong seasonal shift in the flux towards the winter hemisphere subtropical ocean basins and towards the summer hemisphere east coasts of the continents. During JJA the fluxes appear to be too large over the Bay of Bengal, the Arabian sea and in the subtropical North Atlantic (NCEP only). In both seasons the variability in the reanalyses is underestimated (Figs. 69 and 71), especially over the subtropical ocean basins and in the midlatitudes of the SH. Molod et al. (1995) provide an in depth discussion of results for the DAO, with similar comparisons to the COADS.
The flux of sensible heat at the surface is important for energy transfer from the surface to the atmosphere. The most important factors determining the sensible heat over the oceans are the air-sea temperature gradient and the ocean surface roughness. Oceanic sensible heat flux is generally from the ocean to the atmosphere, but can be downward over sea ice. The sensible heat flux during DJF is largest in the vicinity of the largest temperature gradients, i.e. over the Gulf Stream and Kuroshio currents (Fig. 72); intercomparison suggests that these fluxes are overestimated (underestimated) in the NCEP (DAO). Between 45°S and 60°S the sensible heat flux is negative (positive) in the NCEP (DAO). Over land, the largest fluxes occur over the SH desert regions. During JJA the largest fluxes shift to the NH desert regions (Fig. 74). During DJF the variability in the reanalyses appears to be overestimated over the north Pacific and North Atlantic (Fig. 73). During JJA the variability in the vicinity of Antarctica is greatly overestimated (Fig. 75).
The largest source term for the earth's surface is the net shortwave radiation at the surface, which provides energy for latent and sensible heat flux as well as surface heating. The major factors influencing the net shortwave at the surface are the incoming shortwave radiation at the top of the atmosphere, the surface albedo, and the absorption and reflection by clouds. The net shortwave radiation for the reanalyses and COADS is shown in Figs. 76 and 78 for DJF and JJA, respectively. During both seasons the largest values are found in the subtropics of the summer hemisphere, and in each case the DAO values are considerably larger. Local minima in the Tropics are found over the ITCZ and maritime continent where much solar radiation is reflected by clouds. During DJF the variability is larger in the DAO in the Tropics (especially ITCZ and SPCZ) and subtropics of the SH (Fig. 77). During JJA the variability is larger in the DAO in the Tropics and subtropics of each hemisphere (Fig. 79).
The longwave radiation at the surface is the difference between the upward surface emission and the downward emission by clouds and the atmosphere. The net surface longwave flux (upward minus downward flux) is smaller than the net surface shortwave and exhibits less seasonal variability. However, large seasonal variations are found over the extratropical storm track regions of both hemispheres. During DJF the net longwave appears to be too large in the DAO over the NH storm tracks and in the subtropical ocean basins of each hemisphere (Fig. 80). The SST values used by COADS and GEOS-DAS are essentially the same, so discrepancies must be due to an underestimation of the downward component, either due to clouds or to water vapor emissions. As noted by Molod et al. (1995), the low downward longwave emissions in the Tropics are probably related to the dry bias. During DJF the variability is larger along the NH ocean / continent boundaries in the DAO (Fig. 81) compared to the NCEP and COADS. During JJA the net longwave is larger in the DAO over the SH oceans and over the NH continents (especially western and desert portions) (Fig. 82).
The net surface heat flux represents the residual heating available to raise the ground temperature or the local SST. It may be considered as a bulk indicator of the interactions between the land-atmosphere or the ocean-atmosphere interfaces. Figures 84 and 86 show positive maxima in the summer hemisphere throughout the subtropics and extratropics, though again the values are considerably larger in the DAO. The positive values extend towards the equator over the eastern oceans corresponding to local minima in evaporative and longwave flux. During DJF the NH fluxes are dominated by the strong negative fluxes over the Kuroshio and Gulf Stream. During JJA the net heat flux into the oceans is largest over the North Pacific and North Atlantic, where again the values appear to be too large in the DAO. Molod et al. (1995) show that the largest errors in the oceanic net energy flux in the DAO are in the downward shortwave fluxes, demonstrating an excess flux in the extratropics and an inadequate flux in the tropics. They argue further that these errors are due to inadequate extratropical cloud forcing and excess tropical cloud forcing.
4.4 Precipitation and Surface Temperature
In Figs. 88-95 we present an intercomparison of the global precipitation for each season from the NCEP reanalysis, the DAO reanalysis, station observations (Schemm), Xie/Arkin (OLR only) and Xie/Arkin (merged); please refer to section 3 for a description of the comparison datasets. For the NCEP reanalysis, a second version of the precipitation field is included (referred to as "tuned") that accounts for excess snow at certain locations.
The reanalysis precipitation patterns show a clear annual cycle, with the heaviest tropical precipitation shifting with the sun. Notable features in the reanalyses include the boreal winter precipitation in the mid-latitude storm tracks, the boreal summer monsoon rainfall over India, and the summer increase in rainfall over land. The largest differences with the observations occur in the vicinity of the heavy tropical rainfall. During all seasons the tight gradients observed in the ITCZ and SPCZ are not well captured in the reanalyses. Also, there is a tendency for the ITCZ in the reanalyses to be shifted north of its observed position throughout the annual cycle. The DJF reanalysis precipitation appears to be somewhat weak over the southeastern United States, portions of Europe and in the North Atlantic storm track (DAO only). During DJF the NCEP and DAO rainfall over South America is somewhat heavier than observed, and the rainband extending southeast of the continent is too weak. The JJA NCEP and DAO precipitation is too heavy over southeast Asia, northern South America, the western tip of Africa, the southeastern United States and much of the NH middle and high latitude land masses. In general, the variability of the reanalysis precipitation is too small in each season, especially in the ITCZ, SPCZ and in other regions of heavy tropical rainfall.
The mean annual cycle of precipitation for the NCEP, DAO and Xie/Arkin (merged) precipitation for selected land regions (Fig. 96) is given in Figs. 97-98. The seasonal cycle is computed from monthly means averaged over the 9 year period 1985-1993. The observations show that the seasonal cycle of precipitation varies considerably from one region to the next; the general behavior in each of the regions is reproduced by the reanalyses. While regions such as East Asia and India show considerable amplitude, there is very little seasonal variation in precipitation over Europe and Africa. The out of phase relationship between North America and South America is clearly evident. During boreal summer the reanalysis values are often larger than the observed values (especially in the DAO) while in boreal winter the agreement is generally better.
An intercomparison of the mean (1985-1993) monthly precipitation over the conterminous United States is given in Figs. 99-101 for the NCEP, DAO and observations (based on hourly gauge data, Higgins et al. 1996), respectively. In agreement with the observations, the reanalyses show that, in an annual mean sense, the wettest parts of the country are the Pacific Northwest and the Gulf Coast States. The reanalyses capture the reduction of rainfall in the Pacific Northwest during spring, the very light precipitation in the west during summer and the heavy precipitation in the southeast during summer (though amplitudes are overestimated). A more detailed intercomparison of the mean annual cycle of precipitation in the reanalyses and in observations for selected regions (Fig. 102) in the United States shows some considerable differences among the regions (Fig. 103). For example, all products show that the precipitation distribution in the Pacific Northwest is out of phase with that in the Midwest and the Great Plains states, while the Rocky Mountain and Northeast regions have little seasonal variation. East of the Rocky Mountains both reanalyses overestimate rainfall rates during the spring and summer months (the amplitude of the diurnal cycle is too large). During the winter season the reanalyses underestimate the observations along the Gulf Coast and Southeast, but show reasonably good agreement elsewhere.
In Figs. 104-111 we present an intercomparison of the global surface temperature and its standard deviation for each season from the NCEP reanalysis, the DAO reanalysis, and station observations (Schemm). Over high (low) latitude land areas the reanalyses are generally colder (warmer) than the observations in all seasons. In addition, the reanalyses show a much stronger topographic signal than that found in the observations. The variability in the reanalyses appears to be underestimated somewhat, especially at high latitudes.
4.5 Concluding Remarks
The results presented in this document are intended to provide an early look at the quality of the NCEP/NCAR and the NASA/Data Assimilation Office (DAO) reanalyses for a common period (1985-1993). The results show that the climate mean and seasonal evolution of the basic prognostic fields appear to be well captured in the NCEP analysis. Differences with the DAO reanalysis over the NH land masses are generally small in the troposphere. The largest differences are over the Tropics, and the SH oceans, where observations are sparse and model bias plays an important role. The general patterns of tropical convection and their seasonal evolution are consistent with available observations, but details of local maxima are not. Other quantities linked to the hydrologic cycle (e.g. evaporation) show substantial differences; these fields are at best weakly constrained by the observations and appear to primarily reflect model bias.
One deficiency in the NCEP/NCAR reanalysis is tied to bias in the humidity and cloud fields. Moisture bias of the GCM is clearly playing a role, as well as deficiencies in how the available moisture observations (currently only rawinsondes) are assimilated. It is anticipated that substantial improvements to the moisture field are possible with the SSM/I observations. Additional improvements to the treatment of clouds, to the PBL and further tuning of the convective parameterization should also help to alleviate these problems. Key moisture problems include weak horizontal moisture gradients between very moist and very dry regions, and a dry Tropics and subtropics over the oceans compared with the vertically integrated moisture from SSM/I.
There are various problems with the precipitation, and near surface temperature and humidity fields. Over land, these include substantial errors in the diurnal cycle (e.g. Higgins et al. 1996b, 1996c; Schubert et al. 1995). Some of these appear to be tied to the convective parameterization and should be remedied with some of the planned changes to the operational GCM. Some of the known problems in the NCEP/NCAR reanalysis precipitation product are: (1) "spectral rain" over the middle and high latitude interiors of the continents, (2) summertime precipitation over eastern North America is overestimated; (3) the amplitude of the diurnal cycle of precipitation over the southeast U.S. is too large with little evidence of a nocturnal maximum over the Great Plains; (4) too little rainfall along the southern coast of the U.S. during wintertime, over East Asia during summertime, and over Indonesia and Australia throughout the year; (5) too much rainfall over Europe during summertime, over the Aleutians during all seasons except summer and over South America throughout the year. Over North America the assimilation underestimates the percentage of rainfree days and days with drizzle when compared to station observations; in addition the assimilation overestimates the percentage of days with daily mean rates in excess of 4 mm day-1.
A number of other problems with the reanalyses have been identified. For example, there is considerable uncertainty in the strength of the Hadley cell; the Hadley cell during NH winter is more vigorous in the NCEP than in the DAO while the opposite occurs during NH summer. Various quantities, including winds, temperature and sea level pressure show evidence of noise near the poles. Some of the noise in the polar winds may be due to the fact that NCEP smoothed the wind components near the pole, rather than smoothing vorticity and divergence and computing u and v from the smoothed fields.
There is also considerable uncertainty in zonal mean fields in the middle and high latitudes of the SH; this uncertainty is largely due to model bias introduced in data sparse regions. Some of this uncertainty may be linked to an error introduced in the use of SH surface pressure bogus data (i.e. PAOBS) in the NCEP reanalysis for the years 1979-1992. The impact of this error on the reanalysis has been examined in detail by the EMC and the CPC; results of these investigations are posted on the NCEP/NCAR Reanalysis Home Page at http://wesley.wwb.noaa.gov/paobs.html. A summary of the impact of this error on the reanalysis is as follows:
(1) the NH is unaffected by this error
(2) the SH middle and high latitudes (poleward of 40°S) are most affected by this error (the SH middle and high latitudes are the most sensitive region on the globe to any kind of change)
(3) the SH winter months are affected more than the SH summer months
(4) the largest differences are close to the surface and decrease rapidly with height
(5) the differences decrease rapidly as the timescale increases from synoptic to monthly
(6) the differences are negligible on the global scale but can be significant on the synoptic scale
(7) the differences are larger in July 1981 than in July 1979 for synopticscale case studies
(8) differences are largest over the southern oceans (45°S-60°S) and then decrease approaching the pole.
Though not discussed here, a primary strength of the NCEP/NCAR and NASA/DAO reanalyses lies in their ability to capture key climate variations associated with ENSO events, monsoons, droughts and other low-frequency variations. This is due, in part, to the ability of each GCM's physical parameterizations to respond realistically to variations in boundary forcing. Results on low-frequency variability are to be presented in a subsequent atlas.