Data Assimilation in Meteorology and Oceanography
(IAPSO, IMAS)
Location: Poynting Physics S02LT
Location of Posters: Bridge Poynting/Watson
Monday 19 July AM
Presiding Chair: Prof Motoyoshi Ikeda (Graduate School of Environmental Earth Science, Hokkaido University and Frontier Research System for Global Change, Sapporo, Japan)
JSP05/E/04-A1 0830
ESTIMATION OF VERTICAL EDDY VISCOSITY COEFFICIENTS BY A ONE-DIMENSIONAL DATA ASSIMILATION MODEL WITH THE ADJOINT EQUATION
Shin'ichi SAKAI (Environmental Science Department, Central Research Institute of Electric Power Industry, 1646 Abiko, Abiko City, Chiba, 270-1194, JAPAN, email: s-sakai@criepi.denken.or.jp)
The vertical eddy viscosity or the surface wind drag coefficients is significant parameter in the coastal ocean model. The field investigation of these parameters is however very difficult, so the actual value of these parameters are still unknown. A variational approach with the adjoint equation is useful to estimate these model parameters. At first, the basic behaviours of the assimilation model were investigated through the identical twin experiments with varying the initial values of control variables, the observation period, and the spatial or temporal intervals of observation data. The main results are as follows: (a) the underestimated conditions of the initial values of the control variables tend to make calculation stable, (b) the appropriate temporal interval of the observation data is 1/5 times as short as the characteristic time scale of the forcing term (wind), (c) the surface current data is especially important in this type of wind forcing model, and (d) the observation period must be longer than the characteristic time scale of the forcing term. Next, the assimilation of in situ data was conducted by using the current data measured by HF ocean radar (-1m) and the moored currentmeter(-2, -5,-10m) in the sea off Fukushima, the open sea in the western Pacific. Two days data were assimilated when the continuous northern wind averaged 5m/s was dominant. The results showed the vertical eddy viscosity coefficients were of the order of 10^(-3) (m^2/s) and the surface wind drag coefficient was about 1.1 * 10^(-3) which were consistent with the past investigation results. One of the advantages of the data assimilation is to estimate the vertical or horizontal profile of the eddy viscosity coefficients.
JSP05/E/02-A1 Invited 0850
ESTIMATES OF AIR-SEA HEAT FLUXES USING THE VARIATIONAL ASSIMILATION SYSTEM FOR THE 1-DIMENSIONAL MIXED LAYER MODEL
YOICHI ISHIKAWA, Toshiyuki Awaji, Masatora Iida, Teiji In (Department of Geo physics, Kyoto University, Kyoto 606-8502, Japan, Email: ishikawa@kugi.kyoto-u.ac.jp), and Bo Qiu(Depar tment of Oceanography, University of Hawaii at Manoa, Hawaii)
The ocean surface mixed layer plays a crucial role in the climate system through the air-sea interactions and hence it is strongly desired to construct the 4-dimensional data assimilation system capable of proving the integrated data set of the mixed-layer temperature and depth and the air-sea heat flux. In this study, a data assimilation system using the 1-dimensional bulk mixed layer model with the variation adjoin method is constructed to estimate the net air-sea heat flux as well as the mixed layer variables from observations of the mixed-layer temperature. When the mixed layer model is designed to reproduce the seasonal cycle of the mixed layer variability, it is required to use quite different equations governing the model variables in the deepening and the shoaling phase. Thus the data assimilation system shows strong non-linearity, causing a difficult situation in obtaining the optimum solution. In this study, the problem associated with the strong non-linearity is solved by the scaling of the gradient of the cost function, which has the potential to reduce the distortion of the Hessian matrix. Furthermore, the incremental method is applied to the minimisation of the cost function. To confirm the effectiveness of our approach, identical twin experiments are carried out by assimilating the simulated mixed-layer temperature into the model before conducting the assimilation of real data. The result shows that the RMS error of the estimated surface heat flux in our approach is 8 W m-2, while the RMS error using the previous approach is much larger, about 40 W m-2. Our assimilation system is applied to estimation of the net air-sea heat flux in the equatorial Pacific using the mixed layer temperature measured by the TOGA-TAO buoy. The estimated surface heat flux is compared with the heat flux obtained by in-situ observations, showing good agreement especially on seasonal time scales. The RMS difference between both values low-pass filtered over 90 days is 28 W m-2. Also, other mixed-layer variables such as the mixed-layer depth show very similar tendency to in-situ observations despite the assimilation of the mixed-layer temperature alone.
JSP05/W/04-A1 0910
RECONSTRUCTION OF SUBSURFACE GEOCHEMICAL FIELDS USING ASSIMILATION OF UPPER OCEAN DATA
M. IKEDA (Graduate School of Environmental Earth Science, Hokkaido University and Frontier Research System for Global Change, Sapporo, Japan, email: mikeda@ees.hokudai.ac.jp) and Y. Sasai (Graduate School of Environmental Earth Science, Hokkaido University)
Oceanographic data are collected much more extensively from the surface than subsurface. A data assimilation method is proposed for determining geochemical fields in the subsurface. The method is tested with a bulk mixed-layer model. The period from fall to winter and the subpolar North Pacific Ocean are chosen, in which the ocean interacts most extensively with the atmosphere via mixed layer development. In addition to physical properties, CO2 and Alkalinity are also considered as model variables. Total carbonate and pCO2 increase in fall through winter, as the mixed layer develops and entrain carbonate-rich subsurface water.
A cost function has squared differences between data and a model solution. Initial values and subsurface properties are chosen as control variables. The seasonal evolutions of the physical properties are well reproduced even without data assimilation. If we attribute the mismatch of the chemical properties to the subsurface chemical data, zonal variabilities become remarkable in the subsurface ocean. The total carbonate is high in the northwest region, suggesting high biological productivity. In the eastern Pacific, alkalinity is low, suggesting high alkalinity pump effects.
JSP05/E/08-A1 0930
INVERSE ESTIMATION OF SEA SURFACE HEAT FLUXES OVER THE EQUATORIAL PACIFIC OCEAN
DONGLIANG YUAN and Michele Rienecker
In this study, we use the Poseidon quasi-isopycnal ocean model and the Reynolds weekly-mean sea surface temperature data combined with TOGA-TAO subsurface temperature observations to estimate the surface heat fluxes over the equatorial Pacific ocean. A variational data assimilation method optimizes the model initial temperature state and the sea surface heat fluxes simultaneously through assimilation of surface and sub-surface temperature data. Because the Reynolds sea surface temperature compilation is weekly, only monthly mean sea surface heat fluxes are obtained. The variational scheme is implemented on a thermodynamic sub-model of the Poseidon model. The velocity field that advects the temperature in the sub-model is provided by a full forward run of the model. After the variational assimilation, the optimized initial temperature state and the sea surface heat fluxes are substituted back into the full run of the forward model to modify the velocity fields used in the next round of variational assimilation. The iteration stops when the differences between two successive estimates of the sea surface heat fluxes satisfy a convergence criterion. The optimized sea surface heat flux will be compared with other estimates.
JSP05/W/01-A1 0950
STRUCTURAL MODEL ERROR, THE LINEAR RANGE, AND VARIATIONAL ASSIMILATION
J.A. HANSEN (1,2) and L.A. Smith (2) (1) Rutherford Appleton Laboratory, Space Science Department, Chilton/Didcot, OXON, OX11 0QX, UK, (2) Mathematical Institute, University of Oxford, Oxford, OX1 3LB, UK. E-mail: J.A.Hansen@rl.ac.uk Telephone No: (01235) 446 220 Fax No: (01235) 445 848)
The perfect model assumption places an operational limit on four dimensional variational assimilation. Models are wrong. Four-dimensional variational assimilation assumes models are right, specifically a model trajectory exists which is consistent with observations and their associated uncertainty. When model error is large, minimising the cost function defined by model/data misfit can result in analyses that differ significantly from both observations and truth. We present the impact of different levels of structural model error in a simple model on 4D-Var assimilation results as a function of assimilation window length, and demonstrate the importance of the type of model error considered (structural vs. parametric). Model error also impacts the linearity assumption intrinsic to the 4D-Var approach. While a difference between the linearisation of a model about a model state and the linearisation of a system about a system state need not limit the 4D-Var minimisation process, it will impact the quality of uncertainty estimates produced by the resulting Hessian, and can limit the effectiveness of an incremental 4D-Var approach.
JSP05/E/10-A1 Invited 1010
4 DIMENSIONAL DATA ASSIMILATION FOR THE LAND SURFACE PROCESSES
Ken-ichi Kuma Numerical Prediction Division Japan Meteorological Agency
For the atmosphere and the ocean, 4DDA is widely used to give the initial condition for the numerical integration. Although the land surface 4DDA has been recently introduced in several modelling centers, it is still far behind to the atmosphere and the ocean. Two main purposes for the land surface 4DDA are 1) Improvement of predictability for the atmosphere in various time scale and 2) Monitoring the land surface processes including hydrosphere, cryosphere, and biosphere for global domain. There are several difficulties for the land surface 4DDA. 1) Numerical model is not well defined by the physical law. 2) Observation and data exchanges are not well established. 3) Some processes highly depend upon locality.
The most fundamental problem will be the issue related with the observation. If we consider the feasibility of the observation system, we must rely upon the remote sensing from the satellite as well as promotion of in-site data exchange. Passive microwave sensors provide the snow amount for dry snow, surface soil moisture for bare land surface. As to the canopy-covered region, the key information derived by the satellite is on photosynthesis process. Our ultimate goal for the land surface 4DDA is to assimilate the data related with photosynthesis process with the soil-vegetation-atmosphere transfer system model. This will enable us to monitor the carbon cycle as well as the hydrological cycle for the global domain.
JSP05/W/17-A1 Invited 1050
ATMOSPHERIC DATA ASSIMILATION AT JAPAN METEOROLOGICAL AGENCY (JMA)
Nobutaka MANNOJI ( Numerical Prediction Division, Japan Meteorological Agency, 1-3-4
Oote-machi, Chiyoda-ku, Tokyo 100-8122, JAPAN, email: nmannoji@npd.kishou.go.jp )
ERS-2 scatterometer data have been assimilated operationally in the T213L30 global model since July 1998. The feature in assimilating ERS-2 data at JMA is that not only the sea surface wind data but also the sea surface pressure data are assimilated. The sea surface pressure data are retrieved from the ERS-2 sea surface wind data referring to the surface pressure observation by buoys and ships. We found that that ERS-2 data have large positive impact in the Southern Hemisphere and small positive impact in the Northern Hemisphere and the Tropics, not only on the surface field but also on the upper air field. One-dimensional variational method (1DVAR) to assimilate TOVS radiance data is now being developed. An experiment in which retrieved temperature is assimilated through optimum interpolation method into the global model showed that 1DVAR has positive impact in the Southern Hemisphere, negative impact in the Tropics and neutral impact in the Northern Hemisphere. The method is being improved and will be in operation in 1999. Kanto Area Meso-scale Experiment (KAMEX), an observational system experiment (OSE) over Kanto Plain for meso-scale forecast, is now being performed. We are investigating impact of obervational data such as the wind profiler, doppler radar, ACARS, sea surface wind by NSCAT and ERS-2, precipitable water by GPS and SSM/I, and moisture bogus data derived from GMS. Three-dimensional variational method for global analysis is now being developed with a low resolution (T63L30L) model. The method will be in operation in 2000 including variational assimilation of TOVS radiance data.
JSP05/E/07-A1 Invited 1110
MRI/JMA OPERATIONAL OCEAN DATA ASSIMILATION SYSTEM -COMPASS-K-
Masafumi KAMACHI (Meteorological Research Institute, 1-1 Nagamine, Tsukuba 305-0052, Japan, email: mkamachi@mri-jma.go.jp) Tsurane Kuragano and Jiang Zhu (both at Oceanographic
Research Department, Meteorological Research Institute, 1-1 Nagamine, Tsukuba 305-0052,
Japan, email: tkuragan@mri-jma.go.jp) Noriya Yoshioka and Satoshi Sugimoto (both in the
Japan Meteorological Agency, 1-3-4 Ohtemachi, Chiyoda-ku, Tokyo 100-8122, Japan,
email: yoshioka@naps.kishou.go.jp) Francesco Uboldi (FISBAT-CNR, Dipartmento di Fisca, Via Irnerio, 46, I-40126 Bologna, Italy, email: uboldi@pinocchio.fisbat.bo.cnr.it)
An ocean data assimilation system, COMPASS-K (the Comprehensive Ocean Modeling, Prediction, Analysis and Synthesis System in the Kuroshio-region), is developed at the Meteorological Research Institute. The system is for understanding ocean variability, an operational system in the Japan Meteorological Agency, and for the GODAE project.
The model is an eddy resolving version of an MRI-OGCM. Space-time decorrelation scales of ocean variability are estimated with TOPEX/POSEIDON (T/P) altimeter data. The scales show inhomogeneous and anisotropic features and give information about an error covariance matrix in an optimum interpolation process. Subsurface temperature and salinity fields are evaluated from T/P altimetry with a correlation scheme and assimilated into the model with a nudging scheme.
Seasonal and interannual variations in the western North Pacific are investigated. Realistic space-time distribution of the physical quantities, the path of Kuroshio and its separation from Honshu are captured well. The assimilation results give useful information about an effect of eddy-mean flow interaction to Kuroshio meander and dynamical insights of the separation.
JSP05/E/03-A1 1130
TEST AND IMPLEMENTATION OF A MESOSCALE DATA ASSIMILATION CHAIN FOR VERY SHORT RANGE FORECAST PURPOSES.
Pier Paolo Alberoni, STEFANO COSTA, Tiziana Paccagnella, Paolo Patruno (both at Servizio Meteorologico Regionale, A.R.P.A. Emilia-Romagna, Bologna, Viale Silvani, 6, 40122 Italy,
email: p.alberoni@arpamet.regione.emilia-romagna.it) Vincenzo Levizzani (Institute FISBAT-CNR, Bologna, via Gobetti 101, 40129, Italy, email: v.levizzani@fisbat.bo.cnr.it and EUMETSAT,
Darmstadt, Germany)
The growth of observational capability over the last decades allows a detailed description of atmospheric structures, from large-scale phenomena to cloud features. The integrated use of whole information available bring together a NWP techniques, to produce very short range forecast (vsrf) product operationally, is till now a challenge for the forecaster community. The coupling of a powerful mesoscale analysis tool with our limited area model (LAMBO) make up the base of SMR vsrf system.
The Local Analysis and Prediction System (LAPS) has been developed by NOAA-FSL as an integrated system to ingest and analyse meteorological data from different observational sources. The system is able to ingest all kinds of data routinely available, i.e. mesonet and regular surface data, radar and profiler data and satellite imagery. LAPS combines and harmonises the dataset to obtain surface and 3D fields of meteorological parameters, that is temperature, pressure, humidity, wind, cloud and other derived variables.
The resulting fields are finally fed into LAMBO, as a very detailed initialisation in order to produce a very short-range forecast. The analysis and assimilation cycles are carried out every hour. The overall system and some examples of forecast products will be presented.
JSP05/W/10-A1 1150
FES98, A REVISED HYDRODYNAMIC AND ASSIMILATION MODEL FOR THE OCEANIC TIDES
F. Lyard (email: Florent.Lyard@cnes.fr), F. Lefevre (email: Fabien.Lefevre@cnes.fr), C. Le Provost (email: Christian.Le-Provost@cnes.fr) and F. Ponchaut (email: ponchaut@cnes.fr) (all at LEGOS/GRGS, UMR5566 CNES-CNRS-UPS, 14 Avenue E. Belin, 31400 Toulouse, France)
A new set of solutions from our FE hydrodynamic tidal model coupled with a revised data assimilation software has recently been made available as the FES98 solutions. The CEFMO model, previously used to compute the FES94.1 solutions, was improved to allow us to compute tidal solutions at a global scale without adding extra continuity conditions at the basin open limits. "Open limit constraints" free solutions, forced only by the astronomical potential and secondary effects (loading and self-attraction), were computed for the main tidal components. The assimilation method, based on the representer approach, has been used to compute the FES95.2 solutions in a multi-basin version also. Thus it was revised in a similar fashion as the hydrodynamic model. As an immediate benefit, the pathologic behaviour of the assimilation model over some places, due to the addition of continuity conditions into the asssimilation problem, has vanished. For instance, the anomalous resonance observed over some specific coastal areas in FES95.2 were eliminated. A careful selection of in situ tide gauge data from different data banks (IAPSO, WOCE and BHI) allowed us to build a collection of about 800 data for the assimilation model. Referred as FES98, the new assimilation solutions are available on a 0.25x 0.25 degrees gridded version of the full finite element solutions. They are totally independent of altimetric measurements, and are as much accurate as the best available altimetric empirical solutions. In addition to the new solutions, the assimilation model gives us some insight look on the model, like the prior and posterior error covariances. These aspects will be discussed as well as the FES98 model improvements compared to FES94 and FES95.
Monday 19 July PM
Presiding Chair: Dr MASA KAMACHI
JSP05/W/16-A1 1400
Experimental seasonal climate forecasts
SEGSCHNEIDER, J (European Center for Medium-Range Weather Forecasting, Shinfield Park, Reading, Berks, Rg2 9ax, UK. E-mail: ned@ecmwf.int Telephone No: 01180 9499 257 Fax No: 01180 986 9450)
Experimental seasonal climate forecasts as they are currently issued on a quasi-operational basis at ECMWF require an initialization. For the oceanic component of the coupled forecast system this is provided by an ocean analysis where subsurface temperature data are assimilated into a global ocean model using an optimum interpolation scheme. Altimeter data provide additional information about the upper ocean heat content. To make use of this information TOPEX/Poseidon and ERS1/2 data are assimilated into the HOPE ocean model by vertical adjustment of the density profiles. Coupled forecasts started from the so-obtained ocean analysis allow a first estimate of the impact on the forecasted SST. The analysis is global, but the focus of this study is the equatorial eastern Pacific, where impact of oceanic conditions on seasonal prediction of climate is largest. Cross-comparisons of TAO/XBT analysis with T/P sea level and T/P analysis with TAO/XBT temperatures show that both assimilation of subsurface temperature and altimeter data improve the ocean analysis as compared to the control (no assimilation). Forecasts of Nino 3 SSTA initialized from altimeter assimilation ocean analysis are closer to observed SSTA than the control forecasts, but because of the probabilistic nature of the problem it is not yet possible to decide whether forecasts are improved compared to those initialized from subsurface temperature assimilation analysis. However, some of the altimeter initialized forecasts are closer to observed SST than the best member of the ensemble of subsurface temperature initialized forecasts. Therefore, and to enlarge the ensemble spread of the forecasts in general, a combination of the two data sources in the ocean analysis will be a next step.
JSP05/W/14-A1 1420
REVERSE INTEGRATIONS OF AN ICE-SHEET MODEL USING A KALMAN FILTER
RICHARD C.A. HINDMARSH, British Antarctic Survey, High Cross, Madingley Road, Cambridge, England CB30ET. E-mail: rcah@bas.ac.uk . Telephone No: +44 1223 221 495 Fax No: +44 1223 362 616
The best constraints of time-dependent ice-sheet behaviour come from the dated marginal retreat sequences following the last glacial maximum. Typically, they comprise geological deposits which define the margin position and which can be dated directly or correlated with dated sites using varve deposits. Excellent sequences are obtainable from Scandinavia and Northern America.
These sequences have been modelled using forward techniques, but the approach used in this paper is to use a Kalman filter integrated backwards in time to force the model to match the margin position. The ability of the Kalman filter to constrain dynamics on unstable manifolds means that reverse integrations oft he essentially diffusive ice-sheet equation can be carried out in this way. The theory for applying the extended Kalman filter is outlined and some examples are presented.
JSP05/W/15-A1 1440
REGIONAL STUDIES AND APPLICATIONS WITH A VARIABLE RESOLUTION STRETCHED GRID DATA ASSIMILATION SYSTEM
Michael FOX-RABINOVITZ (Department of Meteorology/ESSIC, University of Mary-land, College Park, MD 20742, USA, Email: foxrab@atmos.umd.edu) Dick Dee and Lawrence Takacs
(both at General Sciences Corporation,6100 Chevy Chase Drive, Laurel, MD 20707,USA,
Email: ddee@dao.gsfc.nasa.gov, ltakacs@dao.gsfc.nasa.gov)
The variable resolution stretched grid (SG) version of the Goddard Earth Observing System (GEOS) Data Assimilation System (DAS) incorporating the GEOS SG-GCM, is used for regional analysis, forecast, and climate appli- cations. The region of interest with enhanced horizontal resolution, mostly used in experiments, is a rectangle over the U.S. The SG-DAS is capable of reproducing regional mesoscale fields, patterns and diagnostics that are not produced by the medium uniform resolution run with the same amount of grid points as for the SG. The SG-DAS regional analyses and diagnostics are used for: validation of regional climate simulation experiments produced with the SG-GCM for the U.S.1988 summer drought; and are planned to be used for atmospheric chemistry transport experiments. Also, a case study is conducted on a super-typhoon development in December 1997. The SG-DAS appears to be a viable candidate for a variety of regional studies and applications.
JSP05/W/11-A1 1500
CORRECTING SALINITY USING T-S RELATIONS DURING ASSIMILATION OF THERMAL PROFILE DATA.
ALBERTO TROCCOLI and Keith Haines , (Department of Meteorology, University of Edinburgh, Kingís Buildings - Mayfield Road, Edinburgh EH9 3JZ (UK). E-mail: alberto@met.ed.ac.uk Telephone No: 0131 650 5041 Fax No: 0131 650 5780)
Ocean data assimilation has been continuously investigated over the last decade. Most research has focussed on either satellite altimeter data or near surface temperatures, as measured by either eXpendable BathyThermographs (XBT) or moored buoys. Recently it has been realised that salinity also plays an important role determining density and circulation not only at high latitudes, (e.g. Reverdin et al., 1997), but also in the tropics (e.g. Donguy, 1994). However, little attention has been paid to updating salinity (S) in ocean circulation models used for data assimilation (DA). This is mainly due to the paucity of salinity data. As a result, the easiest solution is usually adopted, namely salinity is left unmodified by DA (i.e. preserving the a priori S(z) profile). But this may lead to increased errors in the velocity field (Cooper, 1988).
We will present a simple approach using in situ T-S relations to including salinity field adjustments when only temperature measurements are available. This DA method is then applied to an ocean general circulation model in a twin experiment framework. The true and the parallel runs differ by the external forcings. Simulated XBT data are inserted along selected sections. Results show that S(z) is generally largely improved along those sections. We also address how these changes reflect upon the downstream water properties and whether we can learn something useful from recording the modifications introduced by the assimilation.
JSP05/W/18-A1 1520
DIAGNOSING VERTICAL VELOCITIES USING THE QG OMEGA EQUATION: AN EXAMINATION OF THE ERRORS DUE TO SAMPLING STRATEGY
J. T. ALLEN, D. A. Smeed, A. J. G. Nurser and J. W. Zhang (Southampton Oceanography Centre, Southampton , UK, SO14 3ZH, email: jta@soc.soton.ac.uk) M. Rixen (G.H.E.R., Universite de
Liege, Belgium)
Vertical motion at the mesoscale plays a key role in ocean circulation, ocean-atmosphere interaction, and hence the control of climate variability. It is not yet possible to measure vertical velocities less than 1000 m day-1 directly. However, by assuming quasi-geostrophic (QG) balance, they can be diagnosed from the geostrophic velocity field and suitable boundary conditions. Significant errors in the accuracy of this diagnosis arise from the necessary compromise between spatial resolution and synopticity of a hydrographic survey. Observations of a numerical ocean model have been made on similar temporal and spatial patterns to observations made of the real ocean during a typical mesoscale physics research cruise. The balance between the number of observations and the synopticity of observations effects the apparent flow and in particular the diagnosed vertical motion. A combination of effects can typically lead to errors of 85% in the estimation of net vertical heat flux. An analytical two layer model is used to understand the sources of the components of this error and suggest key parameters for the design of mesoscale sampling.
JSP05/W/13-A1 1600
THE IMPACT OF ANISOTROPIC ERROR CORRELATION
R. SWINBANK (Universities Space Research Association, NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA; email: swinbank@dao.gsfc.nasa.gov) L.P. Riishojgaard, R. Menard
(both at JCET, NASA GSFC, Greenbelt, MD 20771, USA; email: riishojgaard@dao.gsfc.nasa.gov, menard@dao.gsfc.nasa.go)
Most data assimilation systems assume isotropic forecast error correlations, but results from two dimensional Kalman Filter experiments indicate that the correlations can be far from isotropic. In this paper we use a simple two-dimensional data assimilation system, which analyses trace chemical species such as ozone, to assess different approaches to modelling the error correlations. We compare assimilation results using isotropic correlations with results obtained using different approaches to modelling anisotropic correlations: first, using correlations based on the concentrations of the trace chemicals, and secondly using an advective correlation model. We show that these relatively cheap ways of modelling anisotropic correlations give objectively better results than using isotropic correlations. We discuss the possible extension of these approaches to a full 3-D meteorological data assimilation system.
JSP05/W/12-A1 1620
THE USE OF DIRECT NEURAL NETWORK INVERSE MODELS IN DATA RETRIEVAL AND ASSIMILATION
Dan CORNFORD and Ian T Nabney (Neural Computing Research Group, Aston University, Aston Triangle, Birmingham B4 7ET, UK, Email:d.cornford@aston.ac.uk)
The increasing volume of remotely sensed data available for use in Numerical Weather Prediction (NWP) models provides a challenge to data assimilation systems. It is not always feasible to assimilate the remotely sensed variables directly since their relation to the meteorological variables is typically complex and non-linear. Often the 'forward sensor model' relating the meteorological variables to the sensor measurements is available; however, inversion of this model is computationally expensive and often done in 'ad-hoc' ways.
Using the example of satellite scatterometer data we show how direct neural network inverse models allow inference of the local conditional probability of a surface wind vector over the ocean given the local scatterometer observation. The inverse models can then be used in several ways: as preconditioners to any retrieval/assimilation algorithm; and, using the scaled likelihood method, for inference in a Bayesian retrieval procedure. The procedure we have developed permits the retrieval of the wind fields without background information, and can thus be performed independently of the NWP model. The Bayesian retrieval procedure uses a Gaussian Process based prior wind field model which resolves scales relevant to NWP models and has been extended to include fronts. Thus these winds could be used directly in the assimilation procedure, saving time by considerably reducing the burden of forward model inversion in a variational assimilation system.
JSP05/E/09-A1 1620
STATISTICAL ESTIMATION OF THE GRIN FUNCTION
Leonid I.Piterbarg (Center of Applied Mathematical Sciences, USC, Los Angeles, USA,
email: Piter@cams.usc.edu) Sergei V.SEMOVSKI (Limnological Institute SB RAS, P.O.Box 4199 Irkutsk 664033 Russia, email: Semovsky@lin.irk.ru)
The general statement for heat transport is the equation dT/dt=LT+w. If linear operator L is independent of time, random field w is stationary by time and sequence of field observations T(x,y,t) is available, then statistical procedure is presented for estimation of Green function. Leith (1975) proposed method for delta-correlated noise, based on multivariate correlation matrix calculation. Piterbarg & Semovski (1987) considered finite correlation case, the method is based on calculation of multivariate spectra.
Grin function estimation can be used for simulation of transport for arbitrary initial conditions. Estimation of advection field, diffusion, feedback factors can be derived. However, the procedure can also used for "teleconnection" description. Some functionals, derived from Grin function estimation, can be used for description of different characteristics of process under study, including spreading of elementary heat impulse, "sensitivity" and "influence" functions.
Examples of Grin function computation are presented for few model cases. Statistical estimation of Grin function has been used for description of anomalies formation of sea surface temperature for the North Pacific, global surface temperature analysis and "influence function" calculation for the Central Siberia.
JSP05/W/05-A1 1700
THE DETERMINATION OF SURFACE IRRADIANCE DIURNAL CYCLES BY ASSIMILATION OF JOINT GROUND-BASED AND REMOTE SENSING DATA
Oleg Pokrovsky , Nadya Korolevskaya ( Main Geophysical Observatory, Karbyshev str.7, St.Petersburg, 194021, Russia, e-mail: pokrov@main.mgo.rssi.ru)
Subsystem of surface radiative budget (SRB) is one of the most important modules in global observational system (GOS). Ground-based actinometrical network (AN) does not meet such important requirements, as uniformly coverage and spatial density. Satellite data could be potentially to compensate the lack of observational information. Now AN and satellite data are used in separate mode. AN data are implemented for calibration and validation aimes. But the most promising approach could be based on simultaneous assimilation of both kind of data, as only in this case known mutualy supplement features (relatively high accuracy of AN data and global uniformly coverage, accompanied by increasing of spatial resolution and density of satellite data ) could give some synergy effect. Assimilation method should overcome principal difference of these two kind of data to supply the radiative fields, having sufficient diurnal resolution. It has been developed the retrieval method of diurnal cycles of surface radiation budget variables based on satellite observation dataset. The method has been examined by means of the simulation of satellite measurement dataset ERBE and ScaRaB. It has been elaborated a statistical model of surface downward irradiance (SDI) diurnal cycles including a classification into cases of the clear sky, overcast cloudness and some cases of partly cloudness. The complete set of classes changes in wide range from 3 in the winter up to 6 in the summer. To achieve an effective classification by using the method of the discriminant analysis it is necessary to have at least five satellite measurements per day in July and three instantaneous observations - in March. When the classification procedure has been performed it is necessary to consider the retrieval problem of missing diurnal cycle values. Having the 3-5 satellite observations during a day the retrieval of not only daily sums but also the diurnal distributions of irradiance could be effective. So, for example, in July the relative retrieval error decreases to 20-30 %, and in March - to 10-20 %. The study conducted has shown that some reserve for the increasing of the retrieval accuracy is connected to an optimum choice of satellite observation instants during a day.
Tuesday 20 July AM
Presiding Chair: Dr Keith Haines (Department of Meteorology, University of Edinburgh, King's Buildings, Edinburgh, EH9 3JZ, UK)
JSP05/W/08-A2 0930
THE PERFORMANCE OF ALTIMETER DATA ASSIMILATION IN THE OCCAM GLOBAL OCEAN MODEL
Alan FOX and Keith Haines (Department of Meteorology, University of Edinburgh, King's Buildings, Edinburgh, EH9 3JZ, UK. Email: A.D.Fox@ed.ac.uk). Beverly de Cuevas and Dave Webb (Southampton Oceanography Centre, Empress Dock, Southampton SO14 3ZH, UK)
Sequential assimilation of TOPEX/POSEIDON and ERS-1 altimeter data every 10 days into the OCCAM 1/4 degree global ocean model has been performed from the beginning of 1993. The assimilation method uses mapped sea level anomalies provided by CLS and a mean sea level which has been corrected with hydrographic data in the western boundary current separation regions. An error analysis on the surface height fields is performed with the model errors calculated at low resolution on each assimilation using an adaptive method. Surface height updates thus calculated are projected vertically into the model interior using the Cooper and Haines water property preserving method. The principle aims of our assimilation system are two-fold: firstly, to produce improved analyses of surface and subsurface ocean fields; and secondly, using the analyses produce useful forecasts of the ocean. Changes to fields and transports introduced by the altimeter assimilation are highlighted via comparisons with an OCCAM control simulation. Bearing our aims in mind, the performance of the assimilating model is assessed by comparisons with independent XBT data and WOCE hydrographic and ADCP sections, and the ability of the system to predict sea level changes on the 10-40 day timescale is demonstrated.
JSP05/W/07-A2 0950
ALTIMETER ASSIMILATION IN A GLOBAL OCEAN MODEL: MODEL ERROR ESTIMATION.
Keith Haines and Alan FOX (Department of Meteorology, University of Edinburgh, King's Buildings, Edinburgh, EH9 3JZ, UK. Email: A.D.Fox@ed.ac.uk)
Sequential assimilation schemes in oceanography rely on some estimation of model error at assimilation time. The Kalman Filter represents the optimal estimate, but remains computationally unfeasable in any realistic application. An alternative approach is described in which model errors are estimated by application of a simple consistency condition between the known model-data misfits and observational errors, and the unknown model errors at each assimilation time. This calculation is performed at low resolution and results in an adaptive estimate of model error.
The technique is described in relation to assimilation of mapped altimetry data into the OCCAM, 1/4 degree, 36 level, global ocean model. Surface height updates found using the estimated model errors are projected vertically into the interior using the Cooper and Haines water property preserving method. Results of experiments assimilating TOPEX and combined TOPEX+ERS1 altimetry data over the period 1993-1995 are shown. At various times short 'forecast' runs are also performed with durations of up to 2 months. In a useful dual application, the model error estimation procedure described can also be used to gauge forecast skill.
JSP05/W/03-A2 1010
WHAT CAN WE LEARN BY ASSIMILATING ALTIMETRY?
JENS SCHROETER, Bernadette Sloyan and Martin Losch (Alfred Wegener Institute for Polar and Marine Research, Postfach 12 01 61, 27515 Bremerhaven,Germany. Email: jschroeter@awi-bremerhaven.de Telephone No: +49-(0)471-4831 762 , Fax No: +49-(0) 471-4831 797)
The assimilation of satellite altimetry referenced to a geoid is discussed. The first problem that stands out in this context is the geoid error which can dominate the sea-surface-height data to be assimilated and make this information almost useless. The second major problem is the fact that good resolution and high quality geoid models like EGM96 include information about an oceanic climatology. By assimilating altimetry referenced to such a geoid we implicitely assimilate ocean climatology. We apply a box-inverse model of the southern ocean and show how only the long wavelength information of the altimeter data can be used for assimilation. The change in the models inverse solution and in the associated error covariance is presented. The possible impact by future satellite gravity missions on our results will be dicussed.
JSP05/W/09-A2 Invited 1050
DATA ASSIMILATION FOR INITIALIZING COUPLED CLIMATE PREDICTIONS
Antonio J. BUSALACCHI (Laboratory for Hydrospheric Processes, NASA Goddard Space Flight Center, Greenbelt, MD, 20771, USA, Email: tonyb@neptune.gsfc.nasa.gov)
At the beginning of CLIVAR, data assimilation in support of coupled climate prediction is proving to be the link that binds the modeling and observational components of interannual climate research. Initialization studies of coupled ocean-atmosphere prediction models have become an active area of research in the past few years now that the Tropical Pacific Ocean Observing System has a reached a mature stage of deployment. Data assimilation into coupled forecast models of seasonal to interannual climate variability can take the form of assimilation (OI or adjoint) of in situ and remotely sensed data into the ocean model in a stand-alone mode to improve initial conditions (e.g., reduce systematic biases), for parameter estimation, or as part of a coupled initialization procedure. This presentation will discuss the prospects for improved prediction skill as a result of data assimilation approaches to initializing coupled climate models. Predictions of the 1997-1998 El Nino will be used to illustrate how forecast skill was enhanced as a result of various approaches to assimilating surface wind, sea level, subsurface thermal structure and sea surface temperature observations.
JSP05/E/05-A2 1130
ASSIMILATION OF ALTIMETER DATA IN A TROPICAL PACIFIC OGCM
PETER MCLEAN and Mike Davey, UK Meteorological Office, London Road, Bracknell, Berkshire, RG12 2SZ, UK
A Tropical Pacific OGCM has been used to simulate temperature and currents inthis region. The OGCM used in the UK Met. Office has some systematic errors in common with other similar models, with sea surface temperatures having a warm bias in the central and eastern equatorial Pacific. The thermocline in the analysis is also too diffuse which in turn leads to a too weak equatorial undercurrent. Interannual variability of sea surface temperature in different regions of the equatorial Pacific follows the trends of observed anomalies, but underestimates some warm and cold events. Topex Poseidon altimeter data which is produced in gridded format by CLS was assimilated into the tropical Pacific OGCM to try to improve the analysis. Results suggest that there is a tightening of the thermocline so that subsurface temperatures more closely resemble those observed, and that sea surface temperature anomalies are closer to those observed for example during the 1997 El Nino event. However altimeter assimilation was found to be of less use away from the equator with little improvement in the analysis in the subtropical regions.
JSP05/W/13-A2 1150
VARIATIONAL ADJUSTEMENT OF TSUNAMI PARAMETERS BY ASSIMILATION OF TIDAL STATIONS DATA
Carlos A.L. PIRES (Centro de Geofisica da Universidade de Lisboa, Rua da escola Politecnica 58,
1250 Lisboa Codex, Portugal, email: cpires@fc.ul.pt); Pedro A M Miranda (Centro de Geofisica da Universidade de Lisboa, Rua da escola Politecnica 58, 1250 Lisboa Codex, Portugal,
email: pmiranda@fc.ul.pt)
Time series at tidal stations constitute a potential source of information that may be assimilated into oceanic models. That information is too restrictive too infer the full dynamical state of the sea with any assimilation technique. However, it can be relevant or even sufficient in cases where a simplified hydrodynamical model can be applied and initial conditions can be described by a small set of parameters. Both conditions are nearly satisfied in the case of large tsunami events. We develop a four-dimensional (4DVAR) data assimilation scheme for tsunami signs (after a tide removal) at tidal stations along the Iberian Peninsular and North Africa for the Gorringe Bank large earthquake of February 1969. Observations are assimilated into a shallow-water model for a period covering the first two or three significant waves in each station. Minimisation is performed in a functional space of 9 parameters using the well known Okada formulas described a simple tsunami source. This method is able to recover parameter values that are comparable to those obtained by pure geophysical inversion techniques and can eventually be used to complement them.