Sponsoring Society: SSC;
Co-sponsored by ASA Section Stat and Environment
Session Slot: 10:30-12:20 Monday
Estimated Audience Size: ???
AudioVisual Request: ???
Session Title: Data Assimilation and Reconstruction in Canadian
Environmental Research
Theme Session: No
Applied Session: Yes
Session Organizer: Zwiers, Francis W. Canadian Ctr for Climate Modelling & Analysis
Address: Canadian Ctr for Climate Modelling & Analysis P.O. Box 1700, Victoria, BC V8W 2Y2, CANADA
Phone: 250-363-8229
Fax: 250-363-8247
Email: fzwiers@ec.gc.ca
Session Timing: 110 minutes total (Sorry about format):
Opening Remarks by Chair - 5 First Speaker - 30 minutes Second Speaker - 30 minutes Third Speaker - 30 minutes Discussant - 10 minutes Floor Discusion - 5 minutes
Session Chair: Zwiers, Francis W. Canadian Ctr for Climate Modelling & Analysis
Address: Canadian Ctr for Climate Modelling & Analysis P.O. Box 1700, Victoria, BC V8W 2Y2, CANADA
Phone: 250-363-8229
Fax: 250-363-8247
Email: fzwiers@ec.gc.ca
1. Assimilation of Data into Dynamical Models of the Coastal Ocean
Thompson, Keith R., Dalhousie University
Address: Department of Oceanography Dalhousie University Halifax, Nova Scotia Canada B3H 4J1
Phone: 902 494 3491
Fax: 902 494 2885
Email: keith@phys.ocean.dal.ca
Abstract: Driven by the need to forecast the state of the ocean and to handle the vast amount of data that streams in continuously from satellites in earth orbit, oceanographers are developing ways of assimilating data into dynamical models of the ocean. I will review some of the assimilation techniques currently in use and relate them to concepts more familiar to statisticians such as multiple regression. I will illustrate one of the techniques, the so-called adjoint method of data assimilation, by applying it to the problem of predicting tides in the Yellow Sea from coastal sea-level records. In particular I will show how this technique can be used to (i) estimate optimal open boundary conditions for a limited area model and (ii) detect errors in the water depths upon which the model is based. I will conclude by describing an extension of the readily-implemented nudging technique which, in its simplest form, involves the addition of a relaxation term to the model equations that is proportional to the difference between observed and modelled quantities.
2. Background Error Statistics Modelling in a 3D Variational Data Assimilation Scheme: Estimation and Impact On the Resulting Weather Forecasts
Gauthier, Pierre, Environment Canada
Address: Meteorological Research Branch Environment Canada 2121 Trans-Canada Highway (room 502) Dorval, P.Q. CANADA H9P 1J3
Phone: 514-421-4695
Fax: 514-421-2106
Email: pierre.gauthier@ec.gc.ca
Abstract: Since June 1997, the Canadian Meteorological Centre has replaced its previous optimal interpolation (OI) scheme by a 3D variational data assimilation (3D-var) for its global analysis. The regional analysis is also done with the 3D-var since September 1997. Although both OI and 3D-var produce a minimum variance estimate, 3D-var is a better framework to assimilate observations that are indirectly linked to the model variables such as radiance measurements. Moreover, the 3D-var implicitly uses all observations to produce the analyzed value at a grid point while OI could only used a limited number of observations. No data selection algorithm is therefore used in the variational analysis. The 3D-var also offers the possibility to use a wider variety of background error covariance models. Typically, background error correlations are considered to be homogeneous, isotropic and separable and these assumptions yield a compact representation in spectral space. Recently, experiments have been carried out to study the impact of non-separable correlations on the analysis which results in sharper wind and temperature increments. Some avenues are also explored to formulate correlation models based on empirical orthogonal functions that would relax the homogeneity and isotropy assumptions. For all correlation models, the issue of accuracy of the estimation has to be raised especially when the estimates are obtained through a time series of lagged forecasts. Perspectives are also given on how the 3D-var is a necessary step towards a 4D data assimilation scheme.
3. Optimal Estimation of Climate Parameters and Climate Data Reconstruction
Shen, Samuel S. , University of Alberta
Address: Dept. of Mathematical Sci. University of Alberta Edmonton, Alberta Canada T6G 2G1
Phone: 403-492-0216
Fax: 403-492-6826
Email: sam.shen@ualberta.ca
Abstract: This talk discusses two optimal assessments of a climate state: optimal averaging and optimal gridding of the historical climate data.Spectral approach to optimal averaging: An optimal scheme is developed that minimizes the mean square error when using finitely many surface stations to measure the various orders of spherical harmonic components of a climate field. An important formula is derived to demonstrate that the sampling error is relatively insensitive to the exact shapes of empirical orthogonal functions. Two examples are given: the global average of the annual surface air temperature using 63 stations and the regional average of the monthly tropical Pacific SST.
Adaptive data gridding: The validation of climate models requires to reconstruct climate fields for the past, say 1985-1930, from the scarce observation data. A field can be reconstructed on a one-by-one degree regular grid. A systematic theory for the interpolation is described that uses the recent and more accurate observational data.
Discussant: Switzer, Paul Stanford University
Address: Dept. of Statistics Stanford University Stanford, CA 94305-4065 U.S.A.
Phone: 415-723-2879
Fax: 415-725-8977
Email: ps@playfair.stanford.edu
List of speakers who are nonmembers: Thompson, Shen, Gauthier