Systematic Errors in the NCEP Global Operational Analysis/Forecast System

Glenn H. White, K. Campana, R. Kistler, S. Moorthi, H.-L. Pan and S. Saha

Global Modeling Branch, Environmental Modeling Center

National Centers for Environmental Prediction, National Weather Service

National Oceanic and Atmospheric Administration, U.S. Dept. of Commerce

Washington, D.C., USA

Glenn.White@noaa.gov

Introduction

NCEP has run a global operational analysis/forecast system for 25 years. The present system is described in Table 1. Currently it produces daily global forecasts out to 15 days; a version of the global model also produces seasonal forecasts out to 15 months for the Climate Prediction Center. Operational forecasters examine more model fields than ever before and rely on them for longer forecasts than ever before. Over the years operational forecasters have observed several systematic errors in the global system that can seriously degrade model forecasts; current examples are a warm bias at low levels over the western United States and a cold stratospheric bias over the North Atlantic. Systematic errors are also carefully and routinely examined in the Global Modeling Branch to investigate the performance of the operational model, to suggest needed improvements to the model and to examine the effect of proposed changes to the operational model. The actual decision to implement changes is guided much more by the effect of the changes on standard skill scores and by operational forecasters' assessment than by the effect of the changes on systematic errors. Useful NCEP Internet sites are listed in Table 2.

T170, 42 sigma layers (T62, 28 layers beyond day 7)

3D variational analysis in sigma, no initialization

(Parrish and Derber, 1992; Derber et al., 1991)

Use of satellite radiances

Shortwave radiation-Chou (1992)

Longwave radiation-Fels and Schwarzkopf (1975), Schwarzkopf and Fels (1992)

Cloud fractions diagnosed from relative humidity based on RTNEPH observations

Over ocean Charnock (1955) formula updates momentum roughness, thermal roughness is based on TOGA COARE observations (Zeng et al., 1998)

Bulk aerodynamic formulae for surface fluxes (Miyakoda and Sirutis, 1996)

Boundary layer turbulent mixing-bulk Richardson number based non-local mixing scheme (Troen and Mahrt, 1986; Hong and Pan, 1996)

Free atmosphere vertical diffusion -local mixing scheme.

Shallow convection-Tiedtke (1983)

Deep convection-simplified Arakawa-Schubert (Grell (1993), Pan and Wu (1995), Hong and Pan (1996))

All figures in this paper are analyses and forecasts from 000 GMT. Forecasts are verified against analyses that can have departures from reality themselves, especially for fields involving divergent flow. Figure 1 displays temperature errors for July 2000 over northern hemisphere mid-latitude continents and southern hemisphere mid-latitude oceans. Both regions show a cold stratospheric bias that continues to grow throughout the 15 day forecast. Forecasters preparing flight plans for jets crossing the North Atlantic have noted this cold bias. Since the forecasts also appear to underestimate day-to-day variability in stratospheric temperatures, the error is particularly noticeable when warm anomalies occur. In the lower troposphere the NCEP global model displays a warm bias over the Northern Hemisphere continents in summer and a cold bias over the oceans. Operational forecasters are concerned about the warm bias in summer; there appears to be too much downward short-wave radiation at the surface. Changes to the model's radiation, albedo, aerosol and vegetative index have been tested; they tend to reduce the warm bias but by less than half. The cold bias over the oceans may be associated in part with a model tendency to weaken low-level temperature inversions over the subtropical oceans and the model's lack of low-level stratus in the eastern subtropical oceans.

Fig. 2 displays the evolution of the zonal mean temperature error during March-May 2000. Day 1 forecasts display a warm bias in the upper tropical troposphere that decreases beyond day 1. As will be shown below, upward vertical motion at the equator is stronger at day 1 than in the analysis, but by day 3 is weaker than in the analysis. A cold bias appears over the southern hemisphere mid-latitudes and a warm bias expands upward with forecast length over the Arctic. The warm bias may be associated with too much low-level cloudiness and too little near-surface long-wave cooling. A prognostic cloud liquid water scheme is currently being tested and the resulting cloudiness is being evaluated. Day 15 is dominated by the stratospheric cold bias. Fig. 3 compares the temperature bias during Dec.-Feb. and June-Aug. The Northern Hemisphere mid-latitudes display a cold bias in its winter and a warm bias in its summer, while the Southern Hemisphere mid-latitude cold bias is stronger in its winter. Fig. 4 emphasizes the contrast between the model's low-level warm bias over the summertime continents and cold bias over the oceans nearly everywhere, enhancing the land-sea temperature contrast in the Northern Hemisphere extra-tropics. This error pattern has been present in the NCEP model in the Northern Hemisphere during its summer for many years, although the magnitude of the error has decreased (Caplan and White, 1989; White, 1988). The model's low-level cold bias over the oceans tends to be most intense in the eastern subtropical oceans, where low-level stratus clouds are much more abundant in nature than in the current model. Earlier versions of the model had too much low-level stratus and the parameterization of low-level cloud was changed.

Zonal wind

The forecasts accelerate the jets in the lower stratosphere, displaying an easterly bias near the equator and westerly biases in the subtropics above 200 hPa and decreasing the distinction between the stratospheric and tropospheric jets, as can be seen in Fig. 5. The parameterization of gravity wave drag in the NCEP model is being extensively revised, especially since an updated orography is being introduced. The forecasts also tend to accelerate the low-level tropical trade winds, intensifying the Somali jet (see Fig. 7). The equatorial upper level easterly bias is quite clear even in day 1 forecasts, as Fig. 6 shows. The forecasts greatly strengthen the upper level equatorial easterly jet over the Indian Ocean and move it westwards and strengthen the low-level trades over the Pacific. This may reflect

Fig. 5: (above) Zonal mean zonal wind from (top) analyses and (bottom) 15-day forecasts during June-Aug. 1999. Contour interval 5 m/s, easterlies shaded. (below) zonal mean errors in (top) 1-day, (middle) 5-day and (bottom) 15-day forecasts. Contours (top) 0.25, 0.5, 1, 1.5, 2, 3 … m/s (middle) .5, 1, 2, 3,4, 6, 8 … m/s, (bottom) 1, 2, 3, 4, 5, 6, 8 … m/s, negative values dashed.

the forecasts' failure to maintain the distribution of tropical convection and their tendency to weaken convection over Indonesia.

Divergent flow

During the first 24 hours the NCEP global model tends to "spin up" tropical convection, as can be seen in Fig. 8. By day 3 the model weakens rising motion near the equator and shifts it poleward and lower. By day 15 the forecasts have enhanced upper level convergence and sinking over South Asia (Fig. 9) and upper level divergence and rising motion over the oceans just off South Asia. The forecasts also enhance rising motion over Africa and weaken rising motion over Indonesia and the Inter-tropical Convergence Zones. While the analysis of divergent flow reflects the model physics to a great extent especially in the tropics, the analyzed pattern does resemble rather well independent indicators of the divergent flow such as satellite observations of top of the atmosphere outgoing long-wave radiation and satellite-based precipitation estimates.

During the Northern Hemisphere winter (fig. 10) the 15-day forecasts enhance convection over Africa and South America and decrease it over Indonesia. The failure to maintain the correct distribution of tropical convection is discouraging for longer-range seasonal forecasts, especially since one of the most predictable and most important long-range signals is El Nino/La Nina and is directly linked to tropical convection. Experiments allowing a random cloud top selection rather than the deepest cloud top in the simplified Arakawa-Schubert convection parameterization better maintained the pattern of tropical convection; however, forecasters found that the change led to poorer forecasts over tropical South America and the United States. Momentum mixing by convection is also currently being tested.

Fig. 8: (above) Zonal mean vertical motion from (top) analyses and (bottom) 15-day forecasts during June-Aug. 1999. Contour interval .01 Pa/s, values less than -0.01 Pa/s shaded. (below) Zonal mean errors in (top) 1-day, (middle) 5-day and (bottom) 15-day forecasts of vertical motion. Contours .005, .01, .015, .02, .03, .04, .06, .08 Pa/s, values less than -.005 Pa/s shaded.

Fig. 9: Horizontal divergence at 150 hPa during June-Aug. 1999 from (top) the analyses and (middle) 15-day forecasts. Contour interval 2 x 10^-6/s; values less than -2 x 10^-6/s shaded. (Bottom) Error in 15-day forecasts of horizontal divergence. Contour interval 2 x 10^-6/s, values less than -2 x 10^-6/s shaded.

Conclusions

The NCEP operational global model has the following systematic errors:

a) low-level warm bias over the northern hemisphere summertime continents,

b) low-level warm bias over the Arctic in winter,

c) low-level cold bias over the oceans in all seasons,

d) stratospheric cold bias that grows throughout the 15-day forecast,

e) easterly bias at the equator and westerly bias in the subtropics above 200 hPa,

f) failure to maintain convection over Indonesia, and

g) strengthening of the Somali jet in the Indian summer monsoon.

Many of these biases appear related to problems in radiation and especially in cloudiness and have proved difficult to remove. This indicates the importance of direct verification of model physics as an essential component of model diagnostics and development. Verification against observations, primarily radiosondes, has provided vital information about the NCEP global model in the last few years, but provides no information on model behavior over large areas of the globe.

Fig. 10: As in fig. 9, except for Dec. 1999-Feb. 2000.

References

Caplan, P.M., and G.H. White, 1989: Performance of the National Meteorological Center's medium-range model. Wea.and Fcsting, 4, 391-400.

Charnock, H., 1955: Wind stress on a water surface. Quart. J. Roy. Meteor.Soc., 61,639-640.

Chou, M.-D., 1992: A solar radiation model for use in climate studies. J. Atmos. Sci., 49, 762-777.

Derber, J.C., D.F. Parrish and S.J. Lord, 1991: The new global operational analysis system at the National Meteorological Center. Wea. Forecasting, 6, 538-547.

Fels, S.B., and M.D. Schwarzkopf, 1975: The simplified exchange approximation: A new method for radiative transfer calculations. J. Atmos. Sci., 37, 2265-2297.

Grell, G.A., 1993: Prognostic evaluation of assumptions used by cumulus parameterizations. Mon. Wea. Rev., 121, 764-787.

Hong, S.-Y., and H.-L. Pan, 1996: Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev., 124, 2322-2339.

Miyakoda, K., and J. Sirutis, 1996: Manual of the E-physics. Available from Geophysical Fluid Dynamics Laboratory, Princeton University, PO Box 308, Princeton, NJ 08542.

Pan, H.-L., and W.-S. Wu, 1995: Implementing a mass flux convection parameterization package for the NMC medium-range forecast model. NMC Office Note 409, 40 pp. Available from NCEP, 5200 Auth Rd., Washington, DC 20233.

Parrish, D.F., and J.C. Derber, 1992: The National Meteorological Center's spectral statistical interpolation analysis system. Mon. Wea. Rev., 120, 1747-1763.

Schwarzkopf, M.D., and S.B. Fels, 1991: The simplified exchange method revisited: An accurate rapid method for computation of infrared cooling rates and fluxes. J. Geophys. Res., 96, 9075-9096.

Teidtke, M., 1983: The sensitivity of the time-mean large-scale flow to cumulus convection in the ECMWF model. ECMWF Workshop on Convection in Large-Scale Models, 28 Nov.-1 Dec. 1983, ECMWF, Reading, UK, 297-316.

Troen, I., and L. Mahrt, 1986: A simple model of the atmospheric boundary layer: Sensitivity to surface evaporation. Bound.-Layer Meteor., 37, 129-148.

White, G.H., 1988: The NMC model: Systematic errors in medium and extended range forecasts. Workshop on Systematic Errors in Models of the Atmosphere, Toronto, Canada,19-23Sept.1988. WMO/TD-273,World Meteoro- logical Organization, Geneva, Switzerland, 10-18.

Zeng, X., M. Zhao, and R.E. Dickinson, 1998: Intercomparison of bulk aerodynamical algorithms for the computation of sea surface fluxes using TOGA COARE and TAO data. J. Climate, 11, 2628-2644.