• Andrews, D. G., J. R. Holton, and C. B. Leovy, 1987: Middle Atmosphere Dynamics. Academic Press, 489 pp.

  • Auligné, T., A. P. McNally, and D. P. Dee, 2007: Adaptive bias correction for satellite data in a numerical weather prediction system. Quart. J. Roy. Meteor. Soc., 133, 631642.

    • Search Google Scholar
    • Export Citation
  • Baldwin, M. P., and T. J. Dunkerton, 2001: Stratospheric harbingers of anomalous weather regimes. Science, 294, 581584, doi:10.1126/science.1063315.

    • Search Google Scholar
    • Export Citation
  • Baldwin, M. P., D. B. Stephenson, D. W. J. Thompson, T. J. Dunkerton, A. J. Charlton, and A. O’Neill, 2003: Stratospheric memory and skill of extended-range weather forecasts. Science, 301, 636640, doi:10.1126/science.1087143.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., M. Roch, A.-M. Leduc, P. A. Vaillancourt, S. Laroche, and J. Mailhot, 2009: Medium-range quantitative precipitation forecasts from Canada’s new 33-km deterministic global operational system. Wea. Forecasting, 24, 690708.

    • Search Google Scholar
    • Export Citation
  • Brasseur, G. P., and S. Solomon, 2005: Aeronomy of the Middle Atmosphere. 3rd ed. Springer, 658 pp.

  • Buehner, M., P. L. Houtekamer, C. Charette, H. L. Mitchell, and B. He, 2010: Intercomparison of variational data assimilation and the ensemble Kalman filter for global deterministic NWP. Part II: One-month experiments with real observations. Mon. Wea. Rev., 138, 15671586.

    • Search Google Scholar
    • Export Citation
  • Cardinali, C., S. Pezzulli, and E. Andersson, 2004: Influence-matrix diagnostic of a data assimilation system. Quart. J. Roy. Meteor. Soc., 130, 27672786, doi:10.1256/qj.03.205.

    • Search Google Scholar
    • Export Citation
  • Charlton, A. J., A. O’Neill, W. A. Lahoz, and A. C. Massacand, 2004: Sensitivity of tropospheric forecasts to stratospheric initial conditions. Quart. J. Roy. Meteor. Soc., 130, 17711792.

    • Search Google Scholar
    • Export Citation
  • Charlton, A. J., A. O’Neill, P. Berrisford, and W. A. Lahoz, 2005: Can the dynamical impact of the stratosphere on the troposphere be described by large-scale adjustment to the stratospheric PV distribution? Quart. J. Roy. Meteor. Soc., 131, 525543.

    • Search Google Scholar
    • Export Citation
  • Christiansen, B., 2001: Downward propagation of zonal mean zonal wind anomalies from the stratosphere to the troposphere: Model and reanalysis. J. Geophys. Res., 106 (D21), 27 30737 322.

    • Search Google Scholar
    • Export Citation
  • Christiansen, B., 2005: Downward propagation and statistical forecast of the near-surface weather. J. Geophys. Res., 110, D14104, doi:10.1029/2004JD005431.

    • Search Google Scholar
    • Export Citation
  • CMC, 2008: Addition of new data in the deterministic data assimilation system and the ensemble Kalman filter. Environment Canada’s Tech. Notes on Operational Implementations, 46 pp. [Available online at http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/lib/op_systems/doc_opchanges/technote_20080523_ajoutdonnees_e.pdf.]

  • CMC, 2009: A new stratosphere-troposphere assimilation and medium range forecast system for operational numerical weather prediction. Environment Canada’s Tech. Notes on Operational Implementations, 64 pp. [Available online at http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/lib/op_systems/doc_opchanges/technote_glbstrato_20090618_en.pdf.]

  • Côté, J., S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998a: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 13731395.

    • Search Google Scholar
    • Export Citation
  • Côté, J., J.-G. Desmarais, S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998b: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part II: Results. Mon. Wea. Rev., 126, 13971418.

    • Search Google Scholar
    • Export Citation
  • Dee, D. P., and S. Uppala, 2009: Variational bias correction of satellite radiance data in the ERA-Interim reanalysis. Quart. J. Roy. Meteor. Soc., 135, 18301841.

    • Search Google Scholar
    • Export Citation
  • Douville, H., 2009: Stratospheric polar vortex influence on Northern Hemisphere winter climate variability. Geophys. Res. Lett., 36, L18703, doi:10.1029/2009GL039334.

    • Search Google Scholar
    • Export Citation
  • Gauthier, P., M. Tanguay, S. Laroche, S. Pellerin, and J. Morneau, 2007: Extension of 3DVAR to 4DVAR: Implementation of 4DVAR at the Meteorological Service of Canada. Mon. Wea. Rev., 135, 23392354.

    • Search Google Scholar
    • Export Citation
  • Hines, C. O., 1997a: Doppler-spread parameterization of gravity-wave momentum deposition in the middle atmosphere. Part 1: Basic formulation. J. Atmos. Sol.-Terr. Phys., 59, 371386.

    • Search Google Scholar
    • Export Citation
  • Hines, C. O., 1997b: Doppler-spread parameterization of gravity-wave momentum deposition in the middle atmosphere. Part 2: Broad and quasi monochromatic spectra, and implementation. J. Atmos. Sol.-Terr. Phys., 59, 387400.

    • Search Google Scholar
    • Export Citation
  • Kita, K., and A. Sumi, 1986: Reference ozone models for middle atmosphere. University of Tokyo Geophysical Institute, 26 pp. [Available from Division of Meteorology, Geophysical Institute, University of Tokyo, Bunkyo, Tokyo 113, Japan.]

  • Koshyk, J., B. A. Boville, K. Hamilton, E. Manzini, and K. Shibata, 1999: Kinetic energy spectrum of horizontal motions in middle-atmosphere models. J. Geophys. Res., 104, 27 17727 190.

    • Search Google Scholar
    • Export Citation
  • Kursinski, E. R., G. A. Hajj, J. T. Schofield, R. P. Linfield, and K. R. Hardy, 1997: Observing Earth’s atmosphere with radio occultation measurements using the Global Positioning System. J. Geophys. Res., 102 (D19), 23 42923 465.

    • Search Google Scholar
    • Export Citation
  • Langland, R. H., and N. L. Baker, 2004: Estimation of observation impact using the NRL atmospheric variational data assimilation adjoint system. Tellus, 56, 189201.

    • Search Google Scholar
    • Export Citation
  • Laroche, S., P. Gauthier, M. Tanguay, S. Pellerin, and J. Morneau, 2007: Impact of the different components of 4DVAR on the global forecast system of the Meteorological Service of Canada. Mon. Wea. Rev., 135, 23552364.

    • Search Google Scholar
    • Export Citation
  • Li, J., and H. W. Barker, 2005: A radiation algorithm with correlated-k distribution. Part I: Local thermal equilibrium. J. Atmos. Sci., 62, 286309.

    • Search Google Scholar
    • Export Citation
  • Limpasuvan, V., D. W. Thompson, and D. L. Hartmann, 2004: The life cycle of the Northern Hemisphere sudden stratospheric warmings. J. Climate, 17, 25842596.

    • Search Google Scholar
    • Export Citation
  • Marshall, A. G., and A. A. Scaife, 2010: Improved predictability of stratospheric sudden warming events in an atmospheric general circulation model with enhanced stratospheric resolution. J. Geophys. Res., 115, D16114, doi:10.1029/2009JD012643.

    • Search Google Scholar
    • Export Citation
  • Nezlin, Y., Y. J. Rochon, and S. Polavarapu, 2009: Impact of tropospheric and stratospheric data assimilation on mesospheric prediction. Tellus, 61A, 154159.

    • 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
  • Paul, J., P. Fortuin, and H. Kelder, 1998: An ozone climatology based on ozonesonde and satellite measurements. J. Geophys. Res., 103, 31 70931 734.

    • Search Google Scholar
    • Export Citation
  • Polavarapu, S. M., T. G. Shepherd, S. Ren, and Y. J. Rochon, 2005: Some challenges of middle atmosphere data assimilation. Quart. J. Roy. Meteor. Soc., 131, 35133527.

    • Search Google Scholar
    • Export Citation
  • Ren, S., S. M. Polavarapu, and T. G. Shepherd, 2008: Vertical propagation of information in a middle atmosphere data assimilation system by gravity-wave drag feedbacks. Geophys. Res. Lett., 35, L06804, doi:10.1029/2007GL032699.

    • Search Google Scholar
    • Export Citation
  • Ren, S., S. M. Polavarapu, S. R. Beagley, Y. Nezlin, and Y. J. Rochon, 2011: The impact of gravity wave drag on mesospheric analyses of the 2006 stratospheric major warming. J. Geophys. Res., 116, D19116, doi:10.1029/2011JD015943.

    • Search Google Scholar
    • Export Citation
  • Sankey, D., S. Ren, S. Polavarapu, Y. J. Rochon, Y. Nezlin, and S. Beagley, 2007: Impact of data assimilation filtering methods on the mesosphere. J. Geophys. Res., 112, D24104, doi:10.1029/2007JD008885.

    • Search Google Scholar
    • Export Citation
  • Scinocca, J. F., N. A. McFarlane, M. Lazare, J. Li, and D. Plummer, 2008: Technical note: The CCCma third generation AGCM and its extension into the middle atmosphere. Atmos. Chem. Phys., 8, 70557074.

    • Search Google Scholar
    • Export Citation
  • Shaw, T. A., and T. G. Shepherd, 2007: Angular momentum conservation and gravity wave drag parameterization: Implications for climate models. J. Atmos. Sci., 64, 190203.

    • Search Google Scholar
    • Export Citation
  • Shepherd, T. G., K. Semeniuk, and J. N. Koshyk, 1996: Sponge layer feedbacks in middle-atmosphere models. J. Geophys. Res., 101, 23 44723 464.

    • Search Google Scholar
    • Export Citation
  • Simmons, A., 2000: Assimilation of satellite data for numerical weather prediction: Basic importance, concept and issues. Proc. Seminar on Exploitation of the New Generation of Satellite Instruments for Numerical Weather Prediction, Reading, United Kingdom, ECMWF, 21–46. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/seminar/2000/sem2000_simmons.pdf.]

  • Vallis, G. K., 2006: Atmospheric and Oceanic Fluid Dynamics. Cambridge University Press, 745 pp.

  • Von Storch, H., and F. W. Zwiers, 2003: Statistical Analysis in Climate Research. Cambridge University Press, 470 pp.

  • Waters, J. W., and Coauthors, 2006: The Earth Observing System Microwave Limb Sounder (EOS MLS) on the Aura satellite. IEEE Trans. Geosci. Remote Sens., 44, 10751092.

    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences: An Introduction. Academic Press, 467 pp.

  • WMO, 2010: Manual on the global data-processing and forecasting system. World Meteorological Organization Publ. WMO-485, 256 pp. [Available online at http://www.wmo.int/pages/prog/www/DPFS/Manual_GDPFS.html.]

  • Zadra, A., M. Roch, S. Laroche, and M. Charron, 2003: The subgrid-scale orographic parameterization of the GEM model. Atmos.–Ocean, 41, 155170.

    • Search Google Scholar
    • Export Citation
  • Zhou, S., A. J. Miller, J. Wang, and J. K. Angell, 2002: Downward-propagating temperature anomalies in the preconditioned polar stratosphere. J. Climate, 15, 781792.

    • Search Google Scholar
    • Export Citation
  • View in gallery
    Fig. 1.

    Vertical levels of the low-top model (left and dashed) compared to those of the high-top model (right and solid). The top level of the low-top model is 10 hPa while the high-top model lid is at 0.1 hPa. The vertical coordinate of the high-top model is a hybrid sigma-pressure coordinate. Vertical levels are terrain following near the surface but become pressure levels in the stratosphere.

  • View in gallery
    Fig. 2.

    Verification of northern extratropics for 5-day forecasts against radiosondes. The high-top system (thin) is shown against the low-top system (thick) for zonal wind U and temperature T. Standard deviations (biases) are shown as solid (dashed) lines. Scores are averaged over (a),(b) winter [December–February (DJF)] and (c),(d) summer [June–August (JJA)] tests. Boxes on the left (right) of the figures indicate statistical significance levels for biases (standard deviations). Gray (white) boxes mean the high-top (low top) system is better. No box means the null hypothesis that statistics of the two samples are the same cannot be rejected.

  • View in gallery
    Fig. 3.

    Log10 of temperature power spectra of mean temperature analysis increments averaged over 20 Dec 2006 to 26 Jan 2007 at (left) 500 and (right) 50 hPa.

  • View in gallery
    Fig. 4.

    Five-day geopotential height forecast anomaly correlation time series averaged for the northern extratropics. Anomaly correlation involves verification against own analyses. Time series are for (a) 100 and (b) 500 hPa. The dotted line is for the high-top model and the solid line is for the low-top model. The lines on the right of the graph indicate the average score of all the cases for both systems.

  • View in gallery
    Fig. 5.

    Difference of the 500-hPa geopotential height root-mean-square forecast error against radiosondes in the northern extratropics between the low- and high-top systems. The comparison period is 26 Mar–17 Jun 2009 (i.e., during the operational parallel runs before official implementation). Positive values show greater quality for the high-top system. The x axis is forecast lead time.

  • View in gallery
    Fig. 6.

    Twelve-month running mean of the 500-hPa geopotential height root-mean-square error for 5-day operational forecasts by Environment Canada over the northern extratropics. The error is calculated against radiosondes. Arrows point to significant operational implementations of system upgrades: “Blocking” refers to Zadra et al. (2003), “4DVAR” to Gauthier et al. (2007), “Meso-Global” to Bélair et al. (2009), and “New Obs.” to CMC (2008).

  • View in gallery
    Fig. 7.

    Standard deviations of radiosonde observations minus 5-day forecasts for geopotential height for the Northern Hemisphere (20°–90°N). The low-top model is compared against the high-top model using 4DVAR in the (a) winter (W4ORH: thin solid, W4NNL: thick solid, and W4NNH: dotted) and (c) summer (S4ORH: thin, and S4NNL: thick) periods. The high-top model is compared using 4DVAR and 3DVAR for the (b) winter (W4ORH: thin solid, W3ORH: thick solid, and W3NRH: dashed) and (d) summer (S4ORH: thin solid, and S3ORH: thick solid). Boxes on the right side of each panel indicate statistical significance levels. Gray boxes indicate that (a),(c) the stratospheric model is better or that (b),(d) the 4DVAR is better.

  • View in gallery
    Fig. 8.

    Difference in forecast-error standard deviation between the high- and low-top systems in (a),(b),(e),(f) winter and (c),(d),(g),(h) summer for zonal wind U and temperature T. Negative values mean the high-top system is performing better. Negative contours are dotted, the zero contour is dashed, and positive contours are solid. Superimposed in shaded areas are 95% (light gray) and 90% (dark gray) confidence levels. Experiments used are (a),(b),(g),(h) W4ORH and W4NNL and (c)–(f) S4ORH and S4NNL. Contour intervals are 2 m s−1 and 1 K for U and T, respectively, with two extra contours of half and quarter intervals near zero.

  • View in gallery
    Fig. 9.

    As in Fig. 8, except for absolute bias. Experiments used are (a),(b),(g),(h) W4ORH and W4NNL and (c)–(f) S4ORH and S4NNL. Contour intervals are 1 m s−1 and 0.5 K for U and T, respectively.

  • View in gallery
    Fig. 10.

    Difference in zonal wind U and temperature T forecast-error standard deviation due to (a)–(d) changes in the model (W4NNH and W4NNL) vs (e)–(h) changes in observing network (W4ORH and W4NRH) for the northern extratropical (a),(b),(e),(f) winter and (c),(d),(g),(h) summer. Negative values indicate improvement. Negative contours are dotted, the zero contour is dashed, and positive contours are solid. Contour intervals are (a),(c) 2 m s−1; (b),(d) 1 K; (e),(g) 0.4 m s−1; and (f),(h) 0.4 K. Two additional contours of half and quarter this value are added around 0.

  • View in gallery
    Fig. 11.

    Difference in forecast-error absolute bias due to (a)–(d) changes in the model (W4NNH and W4NNL) vs (e)–(h) changes in observing network (W4ORH and W4NRH) for the northern extratropical (a),(b),(e),(f) winter or (c),(d),(g),(h) summer for (a),(c),(e),(g) zonal wind in m s−1 and (b),(d),(f),(h) temperature in kelvin. Contour intervals are (a),(c) 1 m s−1; (b),(d) 0.5 K; and (e)–(h) 0.2 K or m s−1.

  • View in gallery
    Fig. 12.

    (a),(b) Percent reduction in geopotential height GZ and temperature T forecast-error standard deviation of low-top system due to all changes for the northern extratropical winter (W4ORH vs W4NNL). (c),(d) Percentage of total improvement that is due to model changes only. (W4NNH and W4NNL as a percentage of W4ORH and W4NNL differences.) Negative values (dotted contours) indicate improvement. The zero contour is dashed and positive contours are solid. (a),(b) Superimposed are 90% (dark gray) and 95% (light gray) statistical significances. Contour intervals are 10% for all panels with (a),(b) an extra contour of 5% added.

  • View in gallery
    Fig. 13.

    Observation minus 5-day forecast scores for radiosondes in the northern extratropics (20°–90°N) for (a) geopotential height and (b) temperature. Mean (solid) and standard deviations (dashed) of differences are shown for two experiments: the control (heavy lines) and the degraded model forecast (thin lines). Here, the model was degraded by reverting to the previous radiation scheme. Confidence level in the difference being statistically significant is given by the boxes on the right and left edges of the panel. Unshaded boxes indicate that the control experiment is better. Shaded boxes indicate the degraded model forecast is better. No boxes indicate no statistical difference.

  • View in gallery
    Fig. 14.

    Five-day forecast differences with radiosondes in the northern extratropics (20°–90°N) for (a) zonal wind and (b) geopotential height. Mean (solid) and standard deviations (dashed) of differences are shown for two experiments: the control (heavy lines) and the degraded model forecast (thin lines). Here, the model was degraded by applying a lid at 10 hPa. Confidence levels are indicated using boxes along the vertical edges, following the convention of Fig. 13.

  • View in gallery
    Fig. 15.

    OI per data element for assimilation experiment W4ORH (high-top system) for the northern extratropics (30–90°N) using a (a) stratospheric or (b) tropospheric mask. Results are for January 2007.

  • View in gallery
    Fig. 16.

    Mean total OI for AMSU-A channels for assimilation experiment W4ORH (high-top system) for the northern extratropics (30°–90°N) using a (a) stratospheric or (b) tropospheric mask. Results are for January 2007.

  • View in gallery
    Fig. 17.

    Observation O minus 6-h-forecast F statistics for GPSRO refractivity over the globe. The curves show the average and standard deviation of the normalized difference (OF)/F over the entire cycle. Bias curves are closer to the zero line, and standard deviation curves are farther to the right. Two experiments are shown: GPSRO refractivity vs 6-h forecasts with observations above 30 km used (solid), and GPSRO refractivity vs 6-h forecasts with no observations above 30 km assimilated (dotted) for (a) 4DVAR and (b) 3D-FGAT.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 227 122 8
PDF Downloads 143 66 3

The Stratospheric Extension of the Canadian Global Deterministic Medium-Range Weather Forecasting System and Its Impact on Tropospheric Forecasts

Martin Charron* Meteorological Research Division, Environment Canada, Dorval, Québec, Canada

Search for other papers by Martin Charron in
Current site
Google Scholar
PubMed
Close
,
Saroja Polavarapu Meteorological Research Division, Environment Canada, Toronto, Ontario, Canada

Search for other papers by Saroja Polavarapu in
Current site
Google Scholar
PubMed
Close
,
Mark Buehner* Meteorological Research Division, Environment Canada, Dorval, Québec, Canada

Search for other papers by Mark Buehner in
Current site
Google Scholar
PubMed
Close
,
P. A. Vaillancourt* Meteorological Research Division, Environment Canada, Dorval, Québec, Canada

Search for other papers by P. A. Vaillancourt in
Current site
Google Scholar
PubMed
Close
,
Cécilien Charette* Meteorological Research Division, Environment Canada, Dorval, Québec, Canada

Search for other papers by Cécilien Charette in
Current site
Google Scholar
PubMed
Close
,
Michel Roch* Meteorological Research Division, Environment Canada, Dorval, Québec, Canada

Search for other papers by Michel Roch in
Current site
Google Scholar
PubMed
Close
,
Josée Morneau Canadian Meteorological Centre, Environment Canada, Dorval, Québec, Canada

Search for other papers by Josée Morneau in
Current site
Google Scholar
PubMed
Close
,
Louis Garand* Meteorological Research Division, Environment Canada, Dorval, Québec, Canada

Search for other papers by Louis Garand in
Current site
Google Scholar
PubMed
Close
,
Josep M. Aparicio* Meteorological Research Division, Environment Canada, Dorval, Québec, Canada

Search for other papers by Josep M. Aparicio in
Current site
Google Scholar
PubMed
Close
,
Stephen MacPherson* Meteorological Research Division, Environment Canada, Dorval, Québec, Canada

Search for other papers by Stephen MacPherson in
Current site
Google Scholar
PubMed
Close
,
Simon Pellerin* Meteorological Research Division, Environment Canada, Dorval, Québec, Canada

Search for other papers by Simon Pellerin in
Current site
Google Scholar
PubMed
Close
,
Judy St-James Canadian Meteorological Centre, Environment Canada, Dorval, Québec, Canada

Search for other papers by Judy St-James in
Current site
Google Scholar
PubMed
Close
, and
Sylvain Heilliette* Meteorological Research Division, Environment Canada, Dorval, Québec, Canada

Search for other papers by Sylvain Heilliette in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

A new system that resolves the stratosphere was implemented for operational medium-range weather forecasts at the Canadian Meteorological Centre. The model lid was raised from 10 to 0.1 hPa, parameterization schemes for nonorographic gravity wave tendencies and methane oxidation were introduced, and a new radiation scheme was implemented. Because of the higher lid height of 0.1 hPa, new measurements between 10 and 0.1 hPa were also added. This new high-top system resulted not only in dramatically improved forecasts of the stratosphere, but also in large improvements in medium-range tropospheric forecast skill. Pairs of assimilation experiments reveal that most of the stratospheric and tropospheric forecast improvement is obtained without the extra observations in the upper stratosphere. However, these observations further improve forecasts in the winter hemisphere but not in the summer hemisphere. Pairs of forecast experiments were run in which initial conditions were the same for each experiment but the forecast model differed. The large improvements in stratospheric forecast skill are found to be due to the higher lid height of the new model. The new radiation scheme helps to improve tropospheric forecasts. However, the degree of improvement seen in tropospheric forecast skill could not be entirely explained with these purely forecast experiments. It is hypothesized that the cycling of a better model and assimilation provide improved initial conditions, which result in improved forecasts.

Corresponding author address: Saroja Polavarapu, Environment Canada, 4905 Dufferin Street, Downsview ON M3H 5T4, Canada. E-mail: saroja.polavarapu@ec.gc.ca

Abstract

A new system that resolves the stratosphere was implemented for operational medium-range weather forecasts at the Canadian Meteorological Centre. The model lid was raised from 10 to 0.1 hPa, parameterization schemes for nonorographic gravity wave tendencies and methane oxidation were introduced, and a new radiation scheme was implemented. Because of the higher lid height of 0.1 hPa, new measurements between 10 and 0.1 hPa were also added. This new high-top system resulted not only in dramatically improved forecasts of the stratosphere, but also in large improvements in medium-range tropospheric forecast skill. Pairs of assimilation experiments reveal that most of the stratospheric and tropospheric forecast improvement is obtained without the extra observations in the upper stratosphere. However, these observations further improve forecasts in the winter hemisphere but not in the summer hemisphere. Pairs of forecast experiments were run in which initial conditions were the same for each experiment but the forecast model differed. The large improvements in stratospheric forecast skill are found to be due to the higher lid height of the new model. The new radiation scheme helps to improve tropospheric forecasts. However, the degree of improvement seen in tropospheric forecast skill could not be entirely explained with these purely forecast experiments. It is hypothesized that the cycling of a better model and assimilation provide improved initial conditions, which result in improved forecasts.

Corresponding author address: Saroja Polavarapu, Environment Canada, 4905 Dufferin Street, Downsview ON M3H 5T4, Canada. E-mail: saroja.polavarapu@ec.gc.ca

1. Introduction

Over the past 15 or so years, there has been an increasing recognition of the role of the stratosphere in modulating tropospheric processes. Baldwin and Dunkerton (2001) noted that during winter events of anomalously strong or weak polar vortices, the stratospheric signal appeared roughly 10 days in advance of the tropospheric signal, which then persisted for about 2 months. Thus, the stratosphere has a potential role in improving seasonal forecasts (e.g., Marshall and Scaife 2010; Douville 2009). On the medium-range forecasting (or 2 week) time scale, knowledge of the stratospheric state can also improve the quality of tropospheric forecasts (Christiansen 2001, 2005; Baldwin et al. 2003; Zhou et al. 2002; Charlton et al. 2004). However, mechanisms to explain the stratospheric impact on the troposphere are still in dispute (Charlton et al. 2005).

At the same time that advances in understanding stratosphere–troposphere coupling were being made, new observations of the stratosphere were becoming available for operational assimilation by weather forecasting centers. Indeed, the Advanced Microwave Sounding Unit (AMSU-A) instrument—a microwave sounder that provides global coverage in 6 h because of its presence on several concurrent polar orbiting satellites—is now an important component of the current operational observing system (Cardinali et al. 2004; Langland and Baker 2004). Though its horizontal coverage is excellent, vertically, the contribution of temperature to radiation in the measured frequencies spans deep layers. Thus, channels with peak sensitivity in the stratosphere also sense the troposphere, possibly adding information to the troposphere. Similarly, the Atmospheric Infrared Sounder (AIRS) and Infrared Atmospheric Sounding Interferometer (IASI) have measurements in the infrared that sense both the stratosphere and the troposphere. Arguably then, raising a model lid in order to include additional observations targeting the upper stratosphere might also benefit tropospheric analysis quality through the tropospheric sensitivity of some stratospheric channels of some nadir sounders. Furthermore, since tropospheric peaking channels also sense the stratosphere, the higher lid height and improved background forecast might also result in an improved assimilation of these channels.

In recent years, forecast centers such as the European Centre for Medium-Range Weather Forecasts (ECMWF), the Met Office, the Global Modeling and Assimilation Office (GMAO), and the U.S. Naval Research Laboratory have moved their model tops to the middle mesosphere (80 km or 0.01 hPa). At Environment Canada, the model lid for the Global Deterministic Prediction Systems (GDPS) was recently raised from 10 hPa to a lower value of 0.1 hPa, but for the same reason as the other centers: to improve tropospheric forecasts. In this work, we show that this new high-top system produces significant improvement in forecast skill of not only the stratosphere but also the troposphere. This impact of the stratospheric forecasting system on tropospheric medium-range weather forecasts is maximized on the 5-day time scale, which is shorter than that seen in Baldwin et al. (2003) or Charlton et al. (2004). However, since the high-top system also included model improvements not directly related to the stratospheric extension, it is not clear that the better stratospheric depiction explains the tropospheric impact. Thus, the goal of this work is to document the impact of the new high-top system on stratospheric and tropospheric forecast skill, and to try to identify the reasons for the improvement.

The outline of the article is as follows. In section 2 we present the new operational high-top deterministic prediction system and in section 3 results from this system are presented. In section 4, we determine how much of the improvement in forecast scores is due to the model changes and how much is due to the additional observations in the upper stratosphere. Section 5 then determines which of the model changes were most important for improving forecast scores while section 6 considers which stratospheric observations were most useful. We conclude in section 7 with a summary and discussion of the results.

2. Description of the high-top system

The numerical model utilized for research, development, and operational forecasts at Environment Canada is the Global Environmental Multiscale (GEM) model (Côté et al. 1998a,b). For global medium-range deterministic forecasts, it is run with a horizontal grid spacing of 0.45° in longitude and 0.3° in latitude. The main changes between the previous operational configuration (Bélair et al. 2009; low-top system) and the present one (high-top system) are summarized in Table 1. A more detailed description of the high-top system and its verification prior to implementation are available in (Canadian Meteorological Centre) CMC (2009).

Table 1.

Comparison of parameters and attributes of the previous and new models.

Table 1.

a. Changes to the forecast model

A σ-pressure hybrid vertical coordinate rather than a normalized σ coordinate is used (see appendix). In the troposphere, this coordinate is terrain following, while in the stratosphere it becomes closer to a pressure coordinate (see Fig. 1). The number of vertical levels was increased from 58 to 80 and the model lid was raised from 10 hPa (about 32 km) to 0.1 hPa (about 63 km). A horizontal diffusion operator—a sponge layer—acts over six rather than four levels just below the lid and the sponge layer now acts only on departures from the zonal-mean flow, as opposed to the total flow, as advocated by Shaw and Shepherd (2007). An extra vertical diffusion within 30°S and 30°N just below the lid is now applied over eight levels (from 3 to 0.1 hPa) rather than over four (from 50 to 10 hPa) with a maximum diffusion coefficient reduced by a factor of 9. The radiation scheme was changed to a correlated-k distribution method for gaseous transmission (Li and Barker 2005). This scheme is also used in Environment Canada’s fourth-generation climate model (as noted in Scinocca et al. 2008).

Fig. 1.
Fig. 1.

Vertical levels of the low-top model (left and dashed) compared to those of the high-top model (right and solid). The top level of the low-top model is 10 hPa while the high-top model lid is at 0.1 hPa. The vertical coordinate of the high-top model is a hybrid sigma-pressure coordinate. Vertical levels are terrain following near the surface but become pressure levels in the stratosphere.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

In the high-top model, the parameterization of Hines (1997a,b) treats the effects of subgrid-scale gravity waves from nonorographic sources. In this implementation of the Hines scheme, it is assumed that the geographical distribution of sources is uniform and isotropic. Some characteristics of this implementation include a launching height at about 3 km above the surface, a gravity wave amplitude at launching height of 1 m s−1, and an equivalent horizontal wavelength of 100 km. The resulting forcing on winds near the model lid can be of the order of 50–100 m s−1 day−1.

Methane oxidation is a net source of humidity in the middle atmosphere (see Brasseur and Solomon 2005). In the absence of the chemistry to represent methane oxidation, a simple parameterization is necessary to humidify the stratosphere. The parameterization implemented is very close to that used by ECMWF. The resulting humidity generation/destruction time scale due to methane oxidation is specified to be of the order of 100 days near the stratopause and above, and greater than 2000 days near 10 hPa and below.

The latitude–height monthly ozone climatology of Kita and Sumi (1986) used in the low-top configuration has been replaced by the more recent two-dimensional monthly climatology of Paul et al. (1998) in the high-top configuration. The latter climatology extends up to 0.3 hPa. Above 0.3 hPa, data from the Halogen Occultation Experiment (HALOE) is used. Total ozone concentrations in Paul et al. (1998) are about 10% smaller than those in Kita and Sumi (1986), and are closer to the Total Ozone Mapping Spectrometer (TOMS) measurements.

b. Changes to the assimilation system

Since the high-top configuration has a much higher vertical domain than the low-top one, completely new forecast- (or background) error statistics for the stratosphere needed to be determined. As in the low-top system, the background-error covariances were computed using the “National Meteorological Center (NMC; now known as National Centers for Environmental Prediction) approach” based on 48- minus 24-h forecast difference (Parrish and Derber 1992). Below 50 hPa, the variance profiles were simply obtained from the error variances used in the low-top system. Above this level, temperature variances were obtained by blending the innovation variances obtained from the Microwave Limb Sounder (MLS) temperature retrievals (Waters et al. 2006), MLS observation-error variances, and the original variances from the NMC approach. Wind variances were obtained by applying the wind-temperature ratio calculated from the original forecast-difference variances to the new estimates of the temperature variances. Although vertical correlations in the upper stratosphere were relatively narrow, they were further localized to avoid spreading spurious information into the mesosphere (see Polavarapu et al. 2005). Background-error variances were artificially reduced, relative to objective estimates from innovation statistics, for the top three levels of the model domain to prevent unbounded growth of increments near the model lid.

The vertical extension of the model domain allowed for the assimilation of more data: AMSU-A channels 11–14 and GPS radio occultation (GPSRO) between 30 and 40 km. The GPSRO data are from the Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC); Meteorological Operation (MeTop)/Global Navigation Satellite System Receiver for Atmospheric Sounding (GRAS); and Gravity Recovery and Climate Experiment (GRACE) satellites and are assimilated as refractivity data. Data up to 32 (40) km are used in the low- (high) top system. A new bias correction scheme was implemented in the high-top system to allow for dynamic bias correction of some data types while keeping static bias corrections for other data types. The dynamic scheme updates coefficients used in the observation-error bias model based on the last 15 days of observation-minus-background residuals. However, it was necessary to use static corrections for AMSU-A channels 11–14 to avoid a feedback mechanism that resulted in growing biases with time (Auligné et al. 2007). The predictors used by the bias scheme for the low-top system are thicknesses of geopotential height from trial fields over 1000–300 and 200–50 hPa. For the high-top system, two additional predictors (thicknesses of geopotential height from trial fields for 50–5 and 10–1 hPa) were added.

Observation-error statistics were unchanged except for the specification of the new assimilated AMSU-A channels 11–14 and the reduction of error AMSU-A channels 9 and 10 that had previously been kept artificially high to compensate for modeling errors near the low-top model lid (10 hPa and just below).

3. Assimilation experiments

The final assimilation and forecast tests with the high-top system are verified for the periods 22 June–20 August 2006 (boreal summer) and 27 December 2006–18 February 2007 (boreal winter). These are evaluated against a control experiment consisting of assimilation and forecasts obtained from the low-top system. Both systems employ a four-dimensional variational data assimilation (4DVAR) procedure. Standard deviation and bias scores against radiosondes, at the 5-day forecast range, are presented for the boreal winter periods (Fig. 2) for the northern extratropics (20°–90°N) following World Meteorological Organization standard verification procedures (WMO 2010) and using the radiosonde list updated in 2008. The scores are computed as observations minus forecasts, meaning that a cold model bias translates into a positive value. Statistical significance levels are computed for hypothesis tests that two given experiments have the same mean or variance, using a Student’s t test or an F test (e.g., Wilks 1995), respectively. Forecast scores in boreal winter show a very significant improvement in the stratosphere with consistent improvement throughout the troposphere. This is mostly true for all regions and variables (not all shown). The large biases in the zonal component of the wind in the lower stratosphere are generally greatly reduced with the high-top system (Figs. 2a,c). A small increase in the tropospheric wind speed slow bias is also observed in the northern (Fig. 2a) and southern (not shown) winter extratropics with the high-top system. The bias for temperature is generally improved at all forecast ranges except in the middle troposphere, where the cold bias slightly increases (Fig. 2b). In the extratropical boreal summer stratosphere (Fig. 2d), the temperature bias appears to have deteriorated in the high-top system. This occurs because the version of the GEM model used for all the experiments in this paper does not employ vertical staggering of momentum and thermodynamic variables. There is now evidence that this model version generates an artificial cooling of the stratosphere because of numerical inaccuracies (C. Girard 2011, personal communication). As a result, an accurate parameterization of radiation should produce a cold bias in the stratosphere, considering the known numerical deficiencies of the employed model. The near-zero bias of the low-top system in the extratropical boreal summer stratosphere is indicative of an error compensation between the deficiencies of the model numerics and the low-top radiation scheme. More recent development versions of the GEM model utilize Charney–Phillips vertical staggering and produce a significantly reduced artificial stratospheric cooling as well as a reduced cold bias (not shown).

Fig. 2.
Fig. 2.

Verification of northern extratropics for 5-day forecasts against radiosondes. The high-top system (thin) is shown against the low-top system (thick) for zonal wind U and temperature T. Standard deviations (biases) are shown as solid (dashed) lines. Scores are averaged over (a),(b) winter [December–February (DJF)] and (c),(d) summer [June–August (JJA)] tests. Boxes on the left (right) of the figures indicate statistical significance levels for biases (standard deviations). Gray (white) boxes mean the high-top (low top) system is better. No box means the null hypothesis that statistics of the two samples are the same cannot be rejected.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

In a perfect system for which observation and model errors are unbiased, the mean of the analysis increments is zero. The spatial structure of the mean analysis increments of both model fields is examined in Fig. 3 using power spectra of temperature. At 50 hPa, the power of the mean increment of the high-top system is nearly an order of magnitude lower than that of the low-top system at the largest scales, with more than an order of magnitude reduction for wavenumber 0 (not shown). At 500 hPa, the improvement in bias of the high-top system is smaller but noticeable at all scales. Since the stratosphere is dominated by large scales (Andrews et al. 1987), the reduction in time mean analysis increments at large scales is evidence of the high-top system’s improved representation of the stratosphere. The source of the improvement in bias will be discussed below.

Fig. 3.
Fig. 3.

Log10 of temperature power spectra of mean temperature analysis increments averaged over 20 Dec 2006 to 26 Jan 2007 at (left) 500 and (right) 50 hPa.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

In the stratosphere, large improvements with the high-top system are associated with the elimination of some very poor forecasts (Fig. 4). At 100 hPa (Fig. 4a), the anomaly correlation sometimes drops to low values (29 December, 28 January, and 13 February) with the low-top system. There was a major stratospheric sudden warming at the end of February. Around 29 December and 28 January, rapid changes in zonal wind (polar vortex deformation) were noted. The high-top system maintains a consistently higher correlation at 100 hPa. At 500 hPa (Fig. 4b), the improvement in forecasts is relatively more modest than in the stratosphere, but is still very significant. Indeed, the root-mean-square error (RMSE), calculated using radiosondes, of the 500-hPa geopotential height for 5-day forecasts over the northern extratropics in boreal spring has been reduced by a few meters (Fig. 5). This is an indication that raising the system’s lid to the mesosphere might provide significant advantages for tropospheric forecasts even for the first week in lead time. Overall, the benefit of raising the model lid is comparable to that achieved by upgrading the assimilation scheme from 3DVAR to 4DVAR (Fig. 6) since both operational implementations (4DVAR and high top) generated mean reductions in 500-hPa geopotential height RMSE of over 3 m for 5-day forecasts.

Fig. 4.
Fig. 4.

Five-day geopotential height forecast anomaly correlation time series averaged for the northern extratropics. Anomaly correlation involves verification against own analyses. Time series are for (a) 100 and (b) 500 hPa. The dotted line is for the high-top model and the solid line is for the low-top model. The lines on the right of the graph indicate the average score of all the cases for both systems.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

Fig. 5.
Fig. 5.

Difference of the 500-hPa geopotential height root-mean-square forecast error against radiosondes in the northern extratropics between the low- and high-top systems. The comparison period is 26 Mar–17 Jun 2009 (i.e., during the operational parallel runs before official implementation). Positive values show greater quality for the high-top system. The x axis is forecast lead time.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

Fig. 6.
Fig. 6.

Twelve-month running mean of the 500-hPa geopotential height root-mean-square error for 5-day operational forecasts by Environment Canada over the northern extratropics. The error is calculated against radiosondes. Arrows point to significant operational implementations of system upgrades: “Blocking” refers to Zadra et al. (2003), “4DVAR” to Gauthier et al. (2007), “Meso-Global” to Bélair et al. (2009), and “New Obs.” to CMC (2008).

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

4. Is the improvement due to model or observation changes?

Because a number of changes were introduced in the high-top system, in this section, a series of data assimilation experiments are run in which the differences between the two systems are systematically eliminated. Then, by comparing forecast scores of various pairs of experiments, it is possible to determine the impact of the different types of system changes. The differences between the assimilation systems using the high- and low-top models are of three main types:

  1. Extra upper-stratospheric observations were added to the high-top system. Specifically, AMSU-A channels 11–14 and GPSRO between 30 and 40 km altitudes were added.

  2. Observation errors were reduced for AMSU-A channels 9 and 10. Because of the 10-hPa lid of the low-top model, these channels that have peak sensitivities in that model’s sponge layer were assigned unrealistically high errors in the low-top system. This was not necessary with the high-top system.

  3. The model was changed. In addition to the higher model lid height, different sponge layers, a new radiation scheme, and other more minor changes, the high-top system also eliminated the extrapolations of temperature and moisture profiles (from 10 to 0.1 hPa) prior to applications of the fast Radiative Transfer for the Advanced Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (RTTOV8.7) model that were needed by the low-top model.

The first subsection describes the assimilation experiments performed to isolate the impact of these three types of system changes. Following this, we present results from comparing forecast scores of assimilation experiments, two at a time.

a. Data assimilation experiments

A series of assimilation experiments were performed in which the differences between the two systems were systematically eliminated (see Table 2). The first experiment is the full high-top system (W4ORH) that was implemented operationally on 22 June 2009 by the Canadian Meteorological Centre. The moniker is an acronym for the features of the experiment. For example, W4ORH refers to a boreal winter experiment using 4DVAR, extra observations, reduced observation errors for AMSU-A channels 9 and 10, and a high model lid of 0.1 hPa. The second experiment in Table 2 (W4NRH) is identical to the first experiment except that the extra observations were removed. Thus, comparing W4ORH and W4NRH results reveals the impact of the extra observations. Similarly, comparing W4NRH and W4NNH identifies the impact of reducing the observation errors for AMSU-A channels 9 and 10 in the high-top model system. Finally, comparing results from W4NNH and W4NNL yields the impact of changes to the model only, since the observations are handled identically in both experiments (apart for the temperature profile extrapolation needed by W4NNL for AMSU assimilation). In Table 2, the first four experiments are run in 4DVAR. The 3D first guess at appropriate time (FGAT; see Simmons 2000) experiments revealed that the main results described below are also obtained without 4DVAR and thus not shown. However, this fact is important to note because it means that the impact of the stratosphere on tropospheric forecasts can be studied with a cheaper system. Since the impact of reducing observation errors for channels 9 and 10 (not shown) is small relative to the impact of adding new observations, or changing the model, we focus on the latter two types of changes.

Table 2.

Configuration of various assimilation experiments performed for December 2006–February 2007. Statistics were calculated for 0000 UTC 26 Dec 2006 until 0000 UTC 2 Feb 2007 (providing 77 forecast-error samples), using fields at 0000 and 1200 UTC, for the 4DVAR experiments. The 3DVAR experiments used data from 1200 UTC 20 Dec 2006 to 1200 UTC 26 Jan 2007 for a total of 75 samples.

Table 2.

In each pair of data assimilation experiments, even if observation sets are identical, each experiment is allowed to follow its own quality control decisions. However, when comparing results of different experiments, a common set of verification observations is used. Moreover, the same observation set is used for verification of all experiments in Table 2. This set corresponds to the assimilated observations of experiment W4ORH (the high-top model using 4DVAR). For the boreal summer experiments (Table 3), the common set of observations used for verification is the assimilated set from experiment S4ORH (also with the high-top system).

Table 3.

Configuration of various assimilation experiments performed for June–August 2007. Statistics were calculated from 0000 UTC 22 Jun to 1200 UTC 21 Aug 2006, providing 122 error samples.

Table 3.

To evaluate differences between pairs of assimilation experiments, forecast differences from radiosondes are computed. Significance levels are plotted at the 90% and 95% confidence levels for both the t and the F tests. Since the F test is not robust against departures from the hypotheses of independent, identically distributed, Gaussian samples a lower significance level is typically used (Von Storch and Zwiers 2003) such as 80% or 90%. For the longer forecast ranges, the number of samples used is insufficient to obtain robust results based on experience. That is, the pattern and magnitude of the differences in mean or standard deviation might be different if a different time period (and meteorological situation) had been chosen. Thus, the details of the results presented for forecast ranges beyond day 5 should not be considered as robust. To partially compensate for the lack of robustness, very high confidence intervals of the F test were used.

In the stratosphere, the summer and winter seasons are dynamically very different. In the winter when the westerly jet extends throughout the stratosphere, planetary waves generated in the troposphere can propagate up to the stratosphere and mesosphere. Eventually, they are absorbed at their critical levels, or they break nonlinearly when their amplitudes become too large, impacting the zonal-mean flow. On the other hand, the summer stratosphere is quiescent since the stratospheric jet is easterly and planetary waves cannot propagate upward (see textbooks by Andrews et al. 1987 or Vallis 2006). In the absence of wave forcing, the summer stratosphere is statically stable with temperatures close to those dictated by radiative balance and winds primarily zonal and large scale. In the troposphere, observations are most useful when the flow is rapidly changing as during baroclinic- or synoptic-scale wave development. Similarly, in the stratosphere, one can suspect that observations are most useful in the winter when waves can propagate up from the troposphere and the stratosphere is dynamically active. In the summer, one can expect that fewer observations are required to determine a large-scale, balanced flow. For this reason, the summer and winter hemispheres of the boreal winter experiments in Table 2 are examined separately. Since the observation network density also differs in the two hemispheres, the boreal summer experiments in Table 3 are also considered in an attempt to distinguish differences due to dynamics from differences due to the observing system.

b. Results of assimilation experiments

1) High- versus low-top system

The reduction in standard deviations of 5-day geopotential height forecast differences with radiosondes is huge in the stratosphere (over 100 m at 10 hPa) but significant at all heights (becoming around 3 m at 500 hPa) in winter (Fig. 7a). If the high-top system uses only the observations (and observation-error statistics) of the low-top system, the same huge improvement is seen in the stratosphere and in the troposphere (dotted curve). However, the large improvement in forecast scores in the stratosphere is confined to the wintertime (cf. Figs. 7a,c). The improvement seen in the troposphere is comparable to that seen when comparing 3D-FGAT and 4DVAR in the high-top system during winter (cf. Figs. 7a,b near 300 hPa) but is smaller in summer (cf. Figs. 7c,d). As with 4DVAR, if the extra stratospheric observations are removed from the 3D-FGAT cycle (W3NRH; dashed curve in Fig. 7b), the forecast deteriorates in the stratosphere but not much in the troposphere (cf. dashed and thick solid curves in Fig. 7b).

Fig. 7.
Fig. 7.

Standard deviations of radiosonde observations minus 5-day forecasts for geopotential height for the Northern Hemisphere (20°–90°N). The low-top model is compared against the high-top model using 4DVAR in the (a) winter (W4ORH: thin solid, W4NNL: thick solid, and W4NNH: dotted) and (c) summer (S4ORH: thin, and S4NNL: thick) periods. The high-top model is compared using 4DVAR and 3DVAR for the (b) winter (W4ORH: thin solid, W3ORH: thick solid, and W3NRH: dashed) and (d) summer (S4ORH: thin solid, and S3ORH: thick solid). Boxes on the right side of each panel indicate statistical significance levels. Gray boxes indicate that (a),(c) the stratospheric model is better or that (b),(d) the 4DVAR is better.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

Figure 8 quantifies the total impact of all changes (model and observation) made to obtain the high-top system by showing the difference in standard deviation of forecast scores of the high- and low-top systems as a function of height and forecast range. (In Figs. 811, dewpoint depression is not shown because there was no significant impact in any season or region. Also wind speed results were similar to those for zonal wind so they are also not shown.) It is clear that the high-top system results in lower standard deviation at all forecast ranges and all heights in the northern extratropical winter (Figs. 8a,b). There appears to be a downward spreading of skill with increasing forecast length. Similar magnitudes of improvement as well as the apparent downward propagation of skill are also seen in the southern extratropical winter (Fig. 8e,f). However, because of fewer radiosondes being available in the southern extratropics, the significance of the improvement appears in the troposphere only for longer forecast lengths. The maximum improvements are seen at 10 hPa in both winter and summer, with values of 9.5 m s−1 and 4.9 K in the northern extratropics and 8.0 m s−1 and 3.7 K in the southern extratropics for zonal wind and temperature, respectively. In the summertime, there is still an improvement in forecast-error standard deviation, but it is smaller than that seen in winter and remains largely confined to the stratosphere (Figs. 8c,d,g,h).

Fig. 8.
Fig. 8.

Difference in forecast-error standard deviation between the high- and low-top systems in (a),(b),(e),(f) winter and (c),(d),(g),(h) summer for zonal wind U and temperature T. Negative values mean the high-top system is performing better. Negative contours are dotted, the zero contour is dashed, and positive contours are solid. Superimposed in shaded areas are 95% (light gray) and 90% (dark gray) confidence levels. Experiments used are (a),(b),(g),(h) W4ORH and W4NNL and (c)–(f) S4ORH and S4NNL. Contour intervals are 2 m s−1 and 1 K for U and T, respectively, with two extra contours of half and quarter intervals near zero.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

Figure 9 shows the difference in absolute bias of forecast error.1 Absolute bias is lower with the high-top system for all variables in the wintertime, with the greatest improvement seen in the lower stratosphere. In the summer season, the zonal wind bias is improved in the lower stratosphere with the high-top system but the temperature biases are worsened (note the positive contours) (Figs. 9d,h). Thus, absolute bias is improved with the high-top system only in the wintertime. Note that the deterioration in bias in temperature in the summer is comparable in magnitude to the improvement in standard deviation in the summer (cf. Figs. 8d,h and 9d,h). The source of the worsened bias in temperature and geopotential height during austral summer will be shown (in the next section) to be due to the extra observations assimilated by the high-top system.

Fig. 9.
Fig. 9.

As in Fig. 8, except for absolute bias. Experiments used are (a),(b),(g),(h) W4ORH and W4NNL and (c)–(f) S4ORH and S4NNL. Contour intervals are 1 m s−1 and 0.5 K for U and T, respectively.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

Comparing Figs. 8 and 9 reveals that the improvement in standard deviation is generally much larger than the improvement in bias during winter. This indicates that it is transient events (such as waves) that are being better captured in the high-top system in winter.

It is worth remarking on the fact that the biggest improvement with the high-top system is seen in the northern extratropical winter. Typical assimilation system improvements result in larger impacts in the southern extratropics, where there are fewer in situ measurements (e.g., Gauthier et al. 2007; Buehner et al. 2010). However, in Figs. 8 and 9, the improved stratosphere and troposphere is related more to the season (when planetary waves can propagate to the stratosphere) rather than to the difference in distribution of observations in the two hemispheres. Furthermore, impacts on both bias and standard deviation are similar for both hemispheres in the same season.

2) Relative impact of model changes versus observing system changes

Since all three types of changes are present when comparing the full high- and low-top experiments, the impact of each individual change was obtained for the boreal winter experiments. The previous subsection demonstrated that the key factor is the season, not the hemisphere, so it is not necessary to repeat all the experiments for the July–August period of Table 3. The results in the summer are obtained from the southern extratropics.

Figures 10a,b,e,f show that both the model changes and extra observations in the upper stratosphere lead to reduced standard deviations for all variables for the high-top model in winter. A maximum improvement of 7.8 m s−1 and 4.3 K due to model changes is seen at 10 hPa in the winter (Figs. 10a,b) with improvements of 0.5 m s−1 and 0.25 K below 100 hPa for zonal wind and temperature. However, a maximum improvement of only 1.5 m s−1 and 1.6 K is seen at 10 hPa because of observation changes (Figs. 10e,f). Interestingly, the improvement in skill spreads downward with forecast range, even reaching the surface by day 10. In the summer, model changes result in improved standard deviations (Figs. 10c,d) but the improvement is smaller than that seen in winter (Figs. 10a,b). However, no significant improvement is seen in the summer by adding upper-stratospheric observations (Figs. 10g,h).

Fig. 10.
Fig. 10.

Difference in zonal wind U and temperature T forecast-error standard deviation due to (a)–(d) changes in the model (W4NNH and W4NNL) vs (e)–(h) changes in observing network (W4ORH and W4NRH) for the northern extratropical (a),(b),(e),(f) winter and (c),(d),(g),(h) summer. Negative values indicate improvement. Negative contours are dotted, the zero contour is dashed, and positive contours are solid. Contour intervals are (a),(c) 2 m s−1; (b),(d) 1 K; (e),(g) 0.4 m s−1; and (f),(h) 0.4 K. Two additional contours of half and quarter this value are added around 0.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

Figure 11 considers the impact of just the model change (top row) versus the impact of adding upper-stratospheric observations (bottom row) on absolute bias. Now it is clear that the winter improvement in bias obtained with the high-top system is due to the model changes. The extra observations in the upper stratosphere actually worsen the bias—particularly the temperature at 10 hPa (Figs. 11e,f). This is most likely due to the AMSU-A bias corrections being far from ideal for channels 11–14. [Figure 17 will show that the high-top system has improved bias against GPSRO observations and GPSRO observations have high accuracy and precision at these heights (Kursinski et al. 1997).] The bias correction of these channels was unsatisfactory, likely because of the invalid assumption of unbiased stratospheric forecasts made by the bias correction scheme. In the summer, the model change explains the improvement in zonal wind bias in the lower stratosphere (Fig. 11c). However, both model changes and extra stratospheric observations worsen the bias in temperature in austral summer at 10 hPa. The stratospheric deterioration in temperature bias for boreal summer with respect to the low-top model (Fig. 11d) was seen in Fig. 2 and explained earlier. Comparing the deterioration in bias (Figs. 11e,f) to the improvement in standard deviation (Figs. 10e,f) reveals that the overall impact of the extra observations in winter is positive even at 10 hPa. In austral summer, the temperature deterioration in bias is larger than the impact on standard deviation for both observation changes (cf. Figs. 10g,h and 11g,h) and model changes (cf. Figs. 10c,d and 11c,d).

Fig. 11.
Fig. 11.

Difference in forecast-error absolute bias due to (a)–(d) changes in the model (W4NNH and W4NNL) vs (e)–(h) changes in observing network (W4ORH and W4NRH) for the northern extratropical (a),(b),(e),(f) winter or (c),(d),(g),(h) summer for (a),(c),(e),(g) zonal wind in m s−1 and (b),(d),(f),(h) temperature in kelvin. Contour intervals are (a),(c) 1 m s−1; (b),(d) 0.5 K; and (e)–(h) 0.2 K or m s−1.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

Figure 12 (top row) shows the percentage improvement in standard deviation of the high-top over the low-top system (W4NNL versus W4ORH). A maximum reduction of 75% (69%) is seen at 10 hPa for geopotential height (temperature). While this may be a remarkable improvement, it may simply reflect on the poor quality of the low-top system stratospheric forecasts. It is worth noting that the improvement in the troposphere of 5%–10% in geopotential height is comparable to that seen with other major implementations, such as going from 3DVAR to 4DVAR (see, e.g., Laroche et al. 2007 or Fig. 6). For geopotential height, the model changes account for 80%–90% of the improvement (Fig. 12c). For temperature, the extra observations have more of an impact. The model changes account for only 40% of the improvement at 10 hPa at day 10 (Fig. 12d). The heavy dashed contours (100% contour) appear where the model accounts for over 100% of the improvement. This occurs where the extra observations have a negative impact. However, these sporadic regions occur in the lower troposphere and at longer forecast ranges where the impact of extra observations was not significant. Interestingly, in geopotential height, the extra observations have their greatest impact at the surface on day 9. While this result is not robust and needs to be confirmed by others or with larger sample sizes, Baldwin et al. (2003) and Christiansen (2005) find that the lower stratospheric information is a better predictor of the surface fields 10 days later than the surface fields themselves. We return to this point in section 7.

Fig. 12.
Fig. 12.

(a),(b) Percent reduction in geopotential height GZ and temperature T forecast-error standard deviation of low-top system due to all changes for the northern extratropical winter (W4ORH vs W4NNL). (c),(d) Percentage of total improvement that is due to model changes only. (W4NNH and W4NNL as a percentage of W4ORH and W4NNL differences.) Negative values (dotted contours) indicate improvement. The zero contour is dashed and positive contours are solid. (a),(b) Superimposed are 90% (dark gray) and 95% (light gray) statistical significances. Contour intervals are 10% for all panels with (a),(b) an extra contour of 5% added.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

3) Impact of reducing observation errors for AMSU-A channels 9 and 10

The impact of reducing observation errors for AMSU-A channels 9 and 10 (from 1.60 to 0.32 K for channel 9 and from 3.00 to 0.40 K for channel 10) generally results in improved standard deviations in the midstratosphere and improved temperature biases in the stratosphere (not shown). Thus, these changes partially offset the negative impact on bias due to the addition of AMSU-A channels 11–14 and GPSRO between 30 and 40 km altitudes seen in Figs. 11e–h.

5. Which model changes are most important?

In the previous section, we demonstrated that most of the improvement in forecast scores is obtained even without adding new observations in the upper stratosphere. However, the forecast improvements due to changes in the modeling system may be due to improved forecasts or to improved initial conditions (obtained through better assimilation of existing observations with a resolved stratosphere and a better background state due to an improved forecast model). To separate these two influences, in this section, we use initial conditions from the high-top system to drive a degraded model to see if the poorer forecast scores of the low-top assimilation experiments can be replicated. If they cannot, then the improved initial conditions of the high-top system relative to the low-top system explain the improvement seen in assimilation experiments. Thus, in this section we run 10-day forecasts only (no assimilation cycles). Analyses were obtained from a 4DVAR experiment with the high-top system that had been run for the more recent 2008/09 winter period. Additionally, by degrading the high-top model by removing one of the many model changes that had been introduced, we can assess not only whether the model changes explain the improvement in forecast scores but also which model changes are important to the improvement. Only the changes believed to have a plausible impact on stratospheric forecasts were considered and are as follows.

  1. Revert to old radiation scheme.

  2. Revert to old sponge.

  3. Reduce vertical resolution in the lower stratosphere.

  4. Lower the lid height to 10 hPa.

  5. Revert to the old terrain-following vertical coordinate.

Figure 13 shows the 5-day forecast scores against radiosondes in the northern extratropics for geopotential height (Fig. 13a) and temperature (Fig. 13b). Five-day forecasts were compared to radiosondes every 36 h from 15 December 2008 to 19 February 2009, for a total of 45 samples. By reverting to the old radiation scheme, temperature and geopotential height standard deviations are degraded in the stratosphere, but also in the troposphere. The bias is also degraded in the stratosphere. However, the temperature bias at 500 hPa is better with the old radiation scheme (as noted earlier) (shaded box on left edge of Fig. 13b), as is the geopotential height bias between 500 and 200 hPa. Figure 13 shows that the new radiation scheme played a role in improving stratospheric and tropospheric forecast-error standard deviations. However, the tropospheric improvement (Fig. 13a) is small compared to that seen in Fig. 2a.

Fig. 13.
Fig. 13.

Observation minus 5-day forecast scores for radiosondes in the northern extratropics (20°–90°N) for (a) geopotential height and (b) temperature. Mean (solid) and standard deviations (dashed) of differences are shown for two experiments: the control (heavy lines) and the degraded model forecast (thin lines). Here, the model was degraded by reverting to the previous radiation scheme. Confidence level in the difference being statistically significant is given by the boxes on the right and left edges of the panel. Unshaded boxes indicate that the control experiment is better. Shaded boxes indicate the degraded model forecast is better. No boxes indicate no statistical difference.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

The impact of a too-strong sponge layer was suspected to negatively impact stratospheric motions through excessive damping. The low-top model sponge was applied to full fields (not only departures from a zonal mean as in the high-top system) using a operator and extended down to 50 hPa. The negative impact of sponge layers applied to full fields was discussed by Shaw and Shepherd (2007) and Shepherd et al. (1996). These authors note that spurious downward influence can arise from damping zonal-mean fields or not conserving momentum in gravity wave drag parameterization schemes. Thus, to degrade the model, forecasts were run with the high-top system but using a sponge layer equivalent to that in the low-top system (in terms of magnitude and depth of penetration) and its application to full fields. Five-day forecasts were degraded in the northern extratropical winter but only above 70 hPa, and only by a small amount (not shown). Thus, the raising and weakening of the sponge layer did not contribute to the improvement in tropospheric forecast skill.

The third test involved removing seven vertical levels between 10 and 70 hPa to see whether increased vertical resolution in the lower stratosphere could explain the improvement seen in tropospheric forecasts. Vertical levels below 70 hPa were already comparable in resolution to those used in the low-top system. Given the importance of breaking waves and critical level filtering to the driving of the stratospheric circulation, it was hypothesized that improved vertical resolution of the lower stratosphere would be important for capturing the wave-mean flow interaction, which is important for downward propagation of zonal wind anomalies. Results for this experiment are not shown because no degradation in 5-day forecast-error standard deviation was seen below 70 hPa. Some degradation both in standard deviation and bias was seen in the lower stratosphere, however. A maximum reduction of 0.5 K in temperature-error standard deviation at 10 hPa was seen (which is small compared to the 4.9-K reduction seen in Fig. 2c). Thus, vertical resolution of the lower stratosphere is relevant to stratospheric (but not tropospheric) forecasts, at least in this experiment.

To test the impact of the lid height, in the degraded model the lid height was placed at 10 hPa and the old sponge was also revived to prevent spurious wave reflection from the lowered lid. The vertical resolution was kept comparable to that of the high-top system. The nonorographic gravity wave drag scheme was removed (as it would be ineffective below 10 hPa), as was the methane oxidation. The results from this experiment are shown in Fig. 14. The large degradation in standard deviation (and bias) of forecast error is comparable to that seen in Figs. 7a and 8a. Thus, the lid height being raised from 10 to 0.1 hPa explains almost all of the improvement in standard deviation seen in stratospheric forecasts. Improvement in upper tropospheric forecasts is also evident to 250 hPa (Fig. 14b, unshaded boxes on right). Improvement also occurs below 250 hPa but it is not statistically significant. Both the lid height and radiation scheme contribute to the improvement in stratospheric bias in geopotential height (Figs. 13a and 14b).

Fig. 14.
Fig. 14.

Five-day forecast differences with radiosondes in the northern extratropics (20°–90°N) for (a) zonal wind and (b) geopotential height. Mean (solid) and standard deviations (dashed) of differences are shown for two experiments: the control (heavy lines) and the degraded model forecast (thin lines). Here, the model was degraded by applying a lid at 10 hPa. Confidence levels are indicated using boxes along the vertical edges, following the convention of Fig. 13.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

In the final experiment, the fourth experiment was repeated (lowering the lid height to 10 hPa) but this time, the vertical coordinate was reverted back to the terrain-following sigma-type coordinate of the low-top system. The results are not shown because they were almost indistinguishable from those of the fourth experiment. Thus, the change in vertical coordinate does not explain the improvement seen in forecast skill.

In summary, the large improvement in forecast standard deviations in the stratosphere is almost entirely due to raising the model lid height from 10 to 0.1 hPa. Some of the improvement in tropospheric forecast skill can be explained by the new radiation scheme. Since none of the experiments that degrade the model forecasts could even remotely approach the degree of improvement in tropospheric forecast skill, we conclude that the improved initial conditions explain most of the improvement seen in the tropospheric forecasts.

6. Which of the new observations are most important?

In section 4, the extra observations added to the upper stratosphere were shown to positively impact forecasts in the lower stratosphere during winter, and possibly also those in the troposphere—even those for the surface (Figs. 10d–f). Here we consider which observations are relevant to stratospheric and tropospheric analyses. Observation sensitivity experiments following Langland and Baker (2004) were conducted using an objective function based on differences in 36- and 30-h forecasts and an energy norm. The assimilation experiment studied is that of the high-top system (W4ORH). The verification areas were limited to the stratosphere or the troposphere. Since the tropopause varies with latitude, longitude, and time, an approximate definition was needed to separate the stratosphere and troposphere since results are accumulated in time and space. Between 30°S and 30°N, a tropopause height of 100 hPa was assumed. Between the pole and 30° of each hemisphere, a linear interpolation in lnp was applied assuming a tropopause height of 300 hPa at the pole. This resulted in a tropopause height that decreases with distance from the equator. By separating the objective function by vertical region, it is possible to see which observations impact the stratosphere. Figure 15 shows the observation impact (OI) per data element for the stratosphere (Fig. 15a) and the troposphere (Fig. 15b) for the high-top system and for the northern extratropics. In the stratosphere, as expected, the main contributions are from AMSU-A, radiosondes, and GPSRO observations. The channels peaking in the mid- to upper stratosphere (channels 11–14) do not impact tropospheric forecasts on the 30-h time scale (Fig. 16b). However, channels 7–10 do impact the 30-h stratospheric forecast (Fig. 16a). Ideally, it would be useful to know if their impact is due to their improved assimilation from having a better stratospheric background or to improved 30-h forecasts from having a better model. However, this cannot be determined from Fig. 16.

Fig. 15.
Fig. 15.

OI per data element for assimilation experiment W4ORH (high-top system) for the northern extratropics (30–90°N) using a (a) stratospheric or (b) tropospheric mask. Results are for January 2007.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

Fig. 16.
Fig. 16.

Mean total OI for AMSU-A channels for assimilation experiment W4ORH (high-top system) for the northern extratropics (30°–90°N) using a (a) stratospheric or (b) tropospheric mask. Results are for January 2007.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

Figure 17a shows that the assimilation of the extra stratospheric observations has a positive impact on forecast bias between 30 and 40 km. This is important since a major challenge of stratospheric data assimilation is the separation of forecast and observation bias (Dee and Uppala 2009). Since observation bias correction schemes typically assume that forecasts are unbiased, when enough GPSRO data is assimilated, this assumption might become viable. Furthermore, the advantage of 4DVAR is seen in that the extra observations reduce the standard deviation as well as the bias of forecast error, but with 3D-FGAT only the bias is improved (Fig. 17b). Thus, even though we argue that much of the impact of the stratosphere on tropospheric forecasts can be studied with a 3D-FGAT system, it is clear that 4DVAR is advantageous in the stratosphere and the advantage increases with height (as divergent motions increasingly dominate the energy spectrum; Koshyk et al. 1999).

Fig. 17.
Fig. 17.

Observation O minus 6-h-forecast F statistics for GPSRO refractivity over the globe. The curves show the average and standard deviation of the normalized difference (OF)/F over the entire cycle. Bias curves are closer to the zero line, and standard deviation curves are farther to the right. Two experiments are shown: GPSRO refractivity vs 6-h forecasts with observations above 30 km used (solid), and GPSRO refractivity vs 6-h forecasts with no observations above 30 km assimilated (dotted) for (a) 4DVAR and (b) 3D-FGAT.

Citation: Monthly Weather Review 140, 6; 10.1175/MWR-D-11-00097.1

7. Summary and discussion

A new global deterministic weather prediction system was implemented at Environment Canada on 22 June 2009. The changes introduced with the high-top system include 1) raising the model lid from 10 to 0.1 hPa using a hybrid σ-pressure vertical coordinate, 2) including a nonorographic gravity wave drag scheme and methane oxidation, 3) updating the radiation scheme to a correlated-k distribution algorithm, 4) redesigning and adjusting the background-error statistics, 5) including additional observations (AMSU-A channels 11–14 and GPSRO between 30 and 40 km altitudes), and 6) using a dynamic bias correction algorithm for most satellite observations (except AMSU-A channels 11–14). This new high-top system resulted in very large improvements in stratospheric forecasts as well as significant improvement in tropospheric forecast skill on the medium range (Figs. 212). The improvement in tropospheric forecasts (Fig. 5) rivals that due to the replacement of 3DVAR with 4DVAR (Fig. 6). The greatest improvement was seen in the winter hemisphere (Figs. 79) and most of this improvement could be achieved without adding any new observations in the upper stratosphere (e.g., AMSU-A channels 11–14 and GPSRO between 30 and 40 km) (Figs. 7, 10, and 11). These two facts are consistent with the notion that information propagates vertically with waves that travel upward from the troposphere in winter. Such propagation of information in data assimilation systems has been documented by Nezlin et al. (2009) and can be due to resolved waves (Sankey et al. 2007) or parameterized gravity waves (Ren et al. 2008, 2011). In other words, in the winter, a well-resolved troposphere should result in an improved stratosphere and mesosphere even without adding more observations if the stratosphere is well modeled.

Although the motivation for the high-top system was an improved stratospheric representation, other model changes were introduced at the same time. Therefore, it was not clear that the improvement in tropospheric forecasts is mainly due to the improved representation of the stratosphere. To address this question, a series of forecast experiments were done in which initial conditions came from the “best” high-top system and the model was degraded in ways that mimic the low-top model in one aspect. The lowering of the lid height (Fig. 14) was found to produce the degree of stratospheric forecast degradation seen in the comparison of the full high- and low-top systems (Fig. 7). The old radiation scheme degraded tropospheric forecasts (Fig. 13) but not to the degree seen with the comparison of the full systems (Fig. 7). We conclude that the lid height explains almost all of the improvement seen in stratospheric forecast skill, and that the radiation scheme explains some of the improvement in tropospheric forecast skill. The remaining improvement in tropospheric forecast skill remains unexplained and, thus, may be due to the improved initial conditions. However, since different time periods were used for the forecast and assimilation experiments in sections 4 and 5, the forecast scores are not directly comparable. Moreover, the implicit assumption made in section 5 that model degradations have additive impacts on forecast skill is unlikely to be true in general. Thus, these experiments are not definitive with regard to the source of improvement in tropospheric forecast skill.

Improved tropospheric initial conditions may be due to (i) a better assimilation of stratospheric observations that also sense the troposphere, (ii) a better assimilation of tropospheric observations that also sense the stratosphere, and (iii) a better background, or a combination of these factors. However, AMSU-A channels 11–14 did not have much impact on 30-h tropospheric forecasts (Fig. 16b), suggesting that factor (i) is likely not important. Determining the importance of the remaining two factors will be very difficult, and is left as an open question. Preliminary results obtained with a tripling of mainly tropospheric observations in our high-top system revealed a further large improvement (equivalent to 6 h at 500 hPa for 5-day forecasts). If such improvements were not possible by adding the same new data in the low-top system, then this supports the notion that raising the model’s lid aids in improving the assimilation of those tropospheric measurements that require a good stratospheric representation.

The clear pattern of downward propagation of skill seen in the extratropical winter forecasts over forecast ranges from 1–10 days (in Figs. 8a,b,e,f and 10a,b,e,f) is reminiscent of the downward propagation of zonal wind anomalies seen in time series (e.g., Zhou et al. 2002; Christiansen 2001, 2005; Limpasuvan et al. 2004; etc.). The descent of the zonal wind anomalies is connected with the descent of the northern annular mode signal seen in Baldwin and Dunkerton (2001). Zonal wind forecasts from the high- and low-top systems become increasingly different at progressively lower heights as the forecasts get longer (not shown). Thus, the stratospheric jet structures in forecasts differ in the two systems and therefore the modulation of upward propagating waves will differ. Since the improvement in forecast-error standard deviations of the high-top system over the low-top system is much larger than the improvement in bias (cf. Figs. 8 and 9) in winter, transient events such as waves may be better captured. Thus the improvement in skill at progressively lower heights may be due to both the improved depiction of zonal-mean winds and the improved representation of waves modulated by these winds.

The fact that 3D-FGAT experiments were able to produce the same qualitative results as 4DVAR (concerning seasonality of the impact and impact on the troposphere) (Fig. 7) means that 3D-FGAT is sufficient for future studies that may wish to identify the impact of the stratosphere on tropospheric medium-range forecasts. However, 4DVAR clearly has an advantage in that bias and standard deviations of background error against GPSRO data are lower (Fig. 17). Furthermore, this advantage of 4DVAR increases with height, perhaps because it better resolves the divergent or faster motions that increasingly dominate energy spectra with height.

APPENDIX

The New Hybrid Vertical Coordinate

The GEM model hybrid vertical coordinate obeys the following relations:
eq1
where p is the pressure at the hybrid level η, is a reference pressure chosen to be 800 hPa, is the surface pressure, is the hybrid vertical level at the model top, and is a relaxation coefficient chosen to be 1.6.
The pressure is therefore related to the hybrid level following
eq4
At the model top, , and the following relation must hold:
eq5

REFERENCES

  • Andrews, D. G., J. R. Holton, and C. B. Leovy, 1987: Middle Atmosphere Dynamics. Academic Press, 489 pp.

  • Auligné, T., A. P. McNally, and D. P. Dee, 2007: Adaptive bias correction for satellite data in a numerical weather prediction system. Quart. J. Roy. Meteor. Soc., 133, 631642.

    • Search Google Scholar
    • Export Citation
  • Baldwin, M. P., and T. J. Dunkerton, 2001: Stratospheric harbingers of anomalous weather regimes. Science, 294, 581584, doi:10.1126/science.1063315.

    • Search Google Scholar
    • Export Citation
  • Baldwin, M. P., D. B. Stephenson, D. W. J. Thompson, T. J. Dunkerton, A. J. Charlton, and A. O’Neill, 2003: Stratospheric memory and skill of extended-range weather forecasts. Science, 301, 636640, doi:10.1126/science.1087143.

    • Search Google Scholar
    • Export Citation
  • Bélair, S., M. Roch, A.-M. Leduc, P. A. Vaillancourt, S. Laroche, and J. Mailhot, 2009: Medium-range quantitative precipitation forecasts from Canada’s new 33-km deterministic global operational system. Wea. Forecasting, 24, 690708.

    • Search Google Scholar
    • Export Citation
  • Brasseur, G. P., and S. Solomon, 2005: Aeronomy of the Middle Atmosphere. 3rd ed. Springer, 658 pp.

  • Buehner, M., P. L. Houtekamer, C. Charette, H. L. Mitchell, and B. He, 2010: Intercomparison of variational data assimilation and the ensemble Kalman filter for global deterministic NWP. Part II: One-month experiments with real observations. Mon. Wea. Rev., 138, 15671586.

    • Search Google Scholar
    • Export Citation
  • Cardinali, C., S. Pezzulli, and E. Andersson, 2004: Influence-matrix diagnostic of a data assimilation system. Quart. J. Roy. Meteor. Soc., 130, 27672786, doi:10.1256/qj.03.205.

    • Search Google Scholar
    • Export Citation
  • Charlton, A. J., A. O’Neill, W. A. Lahoz, and A. C. Massacand, 2004: Sensitivity of tropospheric forecasts to stratospheric initial conditions. Quart. J. Roy. Meteor. Soc., 130, 17711792.

    • Search Google Scholar
    • Export Citation
  • Charlton, A. J., A. O’Neill, P. Berrisford, and W. A. Lahoz, 2005: Can the dynamical impact of the stratosphere on the troposphere be described by large-scale adjustment to the stratospheric PV distribution? Quart. J. Roy. Meteor. Soc., 131, 525543.

    • Search Google Scholar
    • Export Citation
  • Christiansen, B., 2001: Downward propagation of zonal mean zonal wind anomalies from the stratosphere to the troposphere: Model and reanalysis. J. Geophys. Res., 106 (D21), 27 30737 322.

    • Search Google Scholar
    • Export Citation
  • Christiansen, B., 2005: Downward propagation and statistical forecast of the near-surface weather. J. Geophys. Res., 110, D14104, doi:10.1029/2004JD005431.

    • Search Google Scholar
    • Export Citation
  • CMC, 2008: Addition of new data in the deterministic data assimilation system and the ensemble Kalman filter. Environment Canada’s Tech. Notes on Operational Implementations, 46 pp. [Available online at http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/lib/op_systems/doc_opchanges/technote_20080523_ajoutdonnees_e.pdf.]

  • CMC, 2009: A new stratosphere-troposphere assimilation and medium range forecast system for operational numerical weather prediction. Environment Canada’s Tech. Notes on Operational Implementations, 64 pp. [Available online at http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/lib/op_systems/doc_opchanges/technote_glbstrato_20090618_en.pdf.]

  • Côté, J., S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998a: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 13731395.

    • Search Google Scholar
    • Export Citation
  • Côté, J., J.-G. Desmarais, S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998b: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part II: Results. Mon. Wea. Rev., 126, 13971418.

    • Search Google Scholar
    • Export Citation
  • Dee, D. P., and S. Uppala, 2009: Variational bias correction of satellite radiance data in the ERA-Interim reanalysis. Quart. J. Roy. Meteor. Soc., 135, 18301841.

    • Search Google Scholar
    • Export Citation
  • Douville, H., 2009: Stratospheric polar vortex influence on Northern Hemisphere winter climate variability. Geophys. Res. Lett., 36, L18703, doi:10.1029/2009GL039334.

    • Search Google Scholar
    • Export Citation
  • Gauthier, P., M. Tanguay, S. Laroche, S. Pellerin, and J. Morneau, 2007: Extension of 3DVAR to 4DVAR: Implementation of 4DVAR at the Meteorological Service of Canada. Mon. Wea. Rev., 135, 23392354.

    • Search Google Scholar
    • Export Citation
  • Hines, C. O., 1997a: Doppler-spread parameterization of gravity-wave momentum deposition in the middle atmosphere. Part 1: Basic formulation. J. Atmos. Sol.-Terr. Phys., 59, 371386.

    • Search Google Scholar
    • Export Citation
  • Hines, C. O., 1997b: Doppler-spread parameterization of gravity-wave momentum deposition in the middle atmosphere. Part 2: Broad and quasi monochromatic spectra, and implementation. J. Atmos. Sol.-Terr. Phys., 59, 387400.

    • Search Google Scholar
    • Export Citation
  • Kita, K., and A. Sumi, 1986: Reference ozone models for middle atmosphere. University of Tokyo Geophysical Institute, 26 pp. [Available from Division of Meteorology, Geophysical Institute, University of Tokyo, Bunkyo, Tokyo 113, Japan.]

  • Koshyk, J., B. A. Boville, K. Hamilton, E. Manzini, and K. Shibata, 1999: Kinetic energy spectrum of horizontal motions in middle-atmosphere models. J. Geophys. Res., 104, 27 17727 190.

    • Search Google Scholar
    • Export Citation
  • Kursinski, E. R., G. A. Hajj, J. T. Schofield, R. P. Linfield, and K. R. Hardy, 1997: Observing Earth’s atmosphere with radio occultation measurements using the Global Positioning System. J. Geophys. Res., 102 (D19), 23 42923 465.

    • Search Google Scholar
    • Export Citation
  • Langland, R. H., and N. L. Baker, 2004: Estimation of observation impact using the NRL atmospheric variational data assimilation adjoint system. Tellus, 56, 189201.

    • Search Google Scholar
    • Export Citation
  • Laroche, S., P. Gauthier, M. Tanguay, S. Pellerin, and J. Morneau, 2007: Impact of the different components of 4DVAR on the global forecast system of the Meteorological Service of Canada. Mon. Wea. Rev., 135, 23552364.

    • Search Google Scholar
    • Export Citation
  • Li, J., and H. W. Barker, 2005: A radiation algorithm with correlated-k distribution. Part I: Local thermal equilibrium. J. Atmos. Sci., 62, 286309.

    • Search Google Scholar
    • Export Citation
  • Limpasuvan, V., D. W. Thompson, and D. L. Hartmann, 2004: The life cycle of the Northern Hemisphere sudden stratospheric warmings. J. Climate, 17, 25842596.

    • Search Google Scholar
    • Export Citation
  • Marshall, A. G., and A. A. Scaife, 2010: Improved predictability of stratospheric sudden warming events in an atmospheric general circulation model with enhanced stratospheric resolution. J. Geophys. Res., 115, D16114, doi:10.1029/2009JD012643.

    • Search Google Scholar
    • Export Citation
  • Nezlin, Y., Y. J. Rochon, and S. Polavarapu, 2009: Impact of tropospheric and stratospheric data assimilation on mesospheric prediction. Tellus, 61A, 154159.

    • 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
  • Paul, J., P. Fortuin, and H. Kelder, 1998: An ozone climatology based on ozonesonde and satellite measurements. J. Geophys. Res., 103, 31 70931 734.

    • Search Google Scholar
    • Export Citation
  • Polavarapu, S. M., T. G. Shepherd, S. Ren, and Y. J. Rochon, 2005: Some challenges of middle atmosphere data assimilation. Quart. J. Roy. Meteor. Soc., 131, 35133527.

    • Search Google Scholar
    • Export Citation
  • Ren, S., S. M. Polavarapu, and T. G. Shepherd, 2008: Vertical propagation of information in a middle atmosphere data assimilation system by gravity-wave drag feedbacks. Geophys. Res. Lett., 35, L06804, doi:10.1029/2007GL032699.

    • Search Google Scholar
    • Export Citation
  • Ren, S., S. M. Polavarapu, S. R. Beagley, Y. Nezlin, and Y. J. Rochon, 2011: The impact of gravity wave drag on mesospheric analyses of the 2006 stratospheric major warming. J. Geophys. Res., 116, D19116, doi:10.1029/2011JD015943.

    • Search Google Scholar
    • Export Citation
  • Sankey, D., S. Ren, S. Polavarapu, Y. J. Rochon, Y. Nezlin, and S. Beagley, 2007: Impact of data assimilation filtering methods on the mesosphere. J. Geophys. Res., 112, D24104, doi:10.1029/2007JD008885.

    • Search Google Scholar
    • Export Citation
  • Scinocca, J. F., N. A. McFarlane, M. Lazare, J. Li, and D. Plummer, 2008: Technical note: The CCCma third generation AGCM and its extension into the middle atmosphere. Atmos. Chem. Phys., 8, 70557074.

    • Search Google Scholar
    • Export Citation
  • Shaw, T. A., and T. G. Shepherd, 2007: Angular momentum conservation and gravity wave drag parameterization: Implications for climate models. J. Atmos. Sci., 64, 190203.

    • Search Google Scholar
    • Export Citation
  • Shepherd, T. G., K. Semeniuk, and J. N. Koshyk, 1996: Sponge layer feedbacks in middle-atmosphere models. J. Geophys. Res., 101, 23 44723 464.

    • Search Google Scholar
    • Export Citation
  • Simmons, A., 2000: Assimilation of satellite data for numerical weather prediction: Basic importance, concept and issues. Proc. Seminar on Exploitation of the New Generation of Satellite Instruments for Numerical Weather Prediction, Reading, United Kingdom, ECMWF, 21–46. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/seminar/2000/sem2000_simmons.pdf.]

  • Vallis, G. K., 2006: Atmospheric and Oceanic Fluid Dynamics. Cambridge University Press, 745 pp.

  • Von Storch, H., and F. W. Zwiers, 2003: Statistical Analysis in Climate Research. Cambridge University Press, 470 pp.

  • Waters, J. W., and Coauthors, 2006: The Earth Observing System Microwave Limb Sounder (EOS MLS) on the Aura satellite. IEEE Trans. Geosci. Remote Sens., 44, 10751092.

    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences: An Introduction. Academic Press, 467 pp.

  • WMO, 2010: Manual on the global data-processing and forecasting system. World Meteorological Organization Publ. WMO-485, 256 pp. [Available online at http://www.wmo.int/pages/prog/www/DPFS/Manual_GDPFS.html.]

  • Zadra, A., M. Roch, S. Laroche, and M. Charron, 2003: The subgrid-scale orographic parameterization of the GEM model. Atmos.–Ocean, 41, 155170.

    • Search Google Scholar
    • Export Citation
  • Zhou, S., A. J. Miller, J. Wang, and J. K. Angell, 2002: Downward-propagating temperature anomalies in the preconditioned polar stratosphere. J. Climate, 15, 781792.

    • Search Google Scholar
    • Export Citation
1

Shading may appear around a zero difference in absolute bias because of interpolation of confidence intervals when plotting. A zero difference in bias occurring between vertical levels will be contoured. However, if the small difference is significant at both vertical levels, shading will appear at both levels and be interpolated over the zero contour.

Save