This file contains basic background information to help you use the Met Office Assimilated data held at the BADC.
This section contains only a brief outline of the assimilation scheme and the data processing steps which involved in the production of these data. Users requiring a more detailed description should consult Swinbank and O'Neill (1994)
1.1 Data Assimilation
Assimilation is essentially an analysis technique that fits a model to observed data. The model (in this case a numerical forecast model of the stratosphere and troposphere) provides a strong constraint to the analysis problem, i.e. the model ensures that the analysed fields are physically self-consistent. The assimilation technique allows data of many different types (for example satellite soundings, radiosondes and aircraft data) to be included in the analysis. By contrast, conventional analysis schemes can only use data of a single type - for example the Met Office TOVS analyses described by Bailey et al. (1991) which are also available form the BADC.
While the model is being integrated forward in time, the model state is nudged towards observed values; the observation weights take into account the expected accuracy of each observation type.
The UARS assimilation system has been developed from the new Met Office data assimilation scheme for operational weather forecasting - the so called analysis-correction scheme described in Lorenc et al. (1991). The forecast model at the heart of the assimilation system is the Met. Office Unified Model which is capable of being run at many different resolutions. For the UARS project it is configured so that its vertical coverage extends from the earth's surface to above the stratopause.
The model is a global primitive equation model, with a split-explicit time integration scheme. It incorporates a comprehensive range of physical parametrization schemes. It uses a hybrid vertical coordinate system, with terrain-following model levels at low levels, gradually changing to pressure levels in the stratosphere.
1.2 Overview of Data Processing
The assimilation system can be divided into the following separate stages, which are shown schematically in Fig. 1:
Extraction of Observations
The first stage in the production is the extraction the necessary data from the Met. Office Data banks, or in case of Level 3A UARS data, from the UARS Central Data Handling Facility (CDHF) at NASA Goddard.
Observation pre-processing and quality control:
The observational data are transcribed into a standard format and are quality controlled by checking against forecast data, using a Bayesian approach devised by Lorenc and Hammond (1988). Data are extracted from forecast fields at observation locations, and a statistical test is performed to establish whether an observation is grossly in error. In addition a "buddy check" is carried out so that if observations of a different type are consistent with each other, then they are more likely to be accepted, even if they differ substantially from the forecast.
Assimilation and forecast:
The assimilation is run by integrating the numerical model in assimilation mode. The model is integrated forward in time and model fields are adjusted towards the observed data so that the model state is a realistic representation of the atmosphere within the assumed errors. The model is also run in forecast mode - in which the model state is governed purely by the model equations.
At the end of the assimilation, the analyzed fields are interpolated to standard pressure levels and written out.
Observation post-processing:
In the final stage of the observation processing, an observation processing dataset (OPD) is compiled. For each observation used in the analysis, this contains the observed value, the difference between the observation and the analysis value and the difference from the background value. Thus the OPD is a valuable resource for validating both the assimilation and the observations.
Error statistics generated from the OPD give users of the assimilated data information on the quality of the analysis.
1.3 "Non-UARS" Assimilations
The standard product consists of analyses produced using operational data - essentially the same data as is used for operational weather forecasting i.e. the World Weather Watch network of surface and upper air observations and satellite data. They are also referred to as "correlative analyses" since they are produced to allow UARS investigators to check their retrievals against an independent analysis. The analyses have been produced operationally since October 1995.
1.4 "UARS" Assimilations
The assimilation scheme has also been run in experiments to incorporate UARS data in addition to the standard meteorological observations. These do not form a continuous sequence, but rather a set of experiments to assimilate particular combinations of data from different instruments.
Note that in the case of the 1991-92 NH winter, the assimilations incorporating temperatures from the UARS ISAMS instrument are thought to be superior to the "non-UARS" data, and should be used in preference.
2.1 Spatial Coverage
Vertical Coverage
All fields are on the 22 UARS pressure levels from 1000 HPa to 0.316 HPa inclusive - approximately 0-55km. (These fields are vertically interpolated from the model data; currently the model has 42 levels, with a vertical resolution in the stratosphere of about 1.6 km)
Horizontal Coverage
The data fields are on the grid used by the numerical model which covers the whole globe. The grid is described in the section on the Resolution below.
2.2 Temporal Coverage
Analyses are available at the CDHF for 12 GMT daily from Oct 17th 1991 with a time lag of a 2-3 days, and shortly thereafter from the BADC. There is a separate file for each day. For periods of special interest, files may be provided four times per day at 0:00, 6.00, 12.00 and 18.00 GMT. The near real-time assimilation data have version numbers V0001 or V0002; higher version numbers denote periods re-run at a later date. Most of the changes made to the assimilation system have not made large differences to the analyses. However, users should note the following significant changes:-
07-Dec-91 The original assimilation run has poor upper levels (p < 3 hPa), which was improved mainly by quality control changes (from 7 Dec to 16 Dec. 07-Jan-92 Polar filtering improved. 27-May-92 Polar filtering improved. This error gave very poor flow near the pole on a few occasions where there were strong winds at high latitudes. 26-Aug-92 Vertical velocities introduced. 22-Feb-93 Divergence damping increased. Gave less noisy temperatures at upper levels (p ~1hPa), with a better fit to observations. 16-Oct-95 Switch over to operational stratospheric run. Still uses the same version of the model, but is run as four six-hourly cycles per day instead of one 24hour assimilation job. It is run closer to real-time, so the observation coverage might be slightly poorer if data is delayed.The temporal coverage of the various assimilation types is summarised in Fig. 2.. Users should note that there is a sequence of data from the 1991-92 northern winter which assimilates temperatures from the ISAMS instrument. These files should be used in preference to the standard product.
2.3 Resolution
The assimilated data are stored on standard UARS pressure levels from 100hPa to 0.316hPa (approximately 55km). The UARS pressure levels are given by the following relation:
p(i) = 1000 x 10^(-i/6), i=0 to 21The horizontal resolution is the same as the numerical model used for data assimilation. The model uses a staggered grid (known as an "Arakawa B grid", so that wind fields are stored on different points from the other fields.
The wind fields (u,v,w) are on a 2.5 degree (latitude) * 3.75 degree (longitude) global grid, with the first point at 88.75N, 1.875E ( 72 rows of 96 points making 6912 points in all).
(88.75,1.875) (88.75,5.625) (88.75,9.375) ... (88.75,358.125) (86.25,1.875) ... ... ... ... ... ... ... ... ... (-88.75,1.875) ... ... ... (-88.75,358.125)The other fields (geopotential height and temperature) are on a 2.5 degree (latitude)* 3.75 degree (longitude) global grid, with the first point at 90.0N, 0E (73 rows of 96 points making 7008 points in all).
(90,0) (90,3.75) (90,7.5) .... (90,356.25) (87.5,0) .... .... .... .... .... .... .... .... .... (-90,0) .... .... .... (-90, 356.25)2.4 Data Quality
The tables presented below are provided to give users a rough estimate of the errors and bias in the analysed fields, using information provided by the Met Office.
Temperature
Pressure Est. RMS error Est. bias} _______________________________________________________ 1000 hPa 1.0K magnitude <0.3K tropopause 1.5K 100 hPa 1.0K magnitude <0.3K 10 hPa 1.0K magnitude <0.3K increasing to... 1hPa 2.0K magnitude <0.5K _______________________________________________________These are representative global-average errors, derived from the `Observation Processing Dataset' (or OPD), a dataset of observed and analysed values compiled as part of the assimilation process. Note that the figures do NOT take into account any bias in the original operational (NOAA/NESDIS) temperature retrievals. In general, the errors will be more than average at high latitudes and in winter. In particular, errors will be larger (perhaps 10-20K locally) during dynamically active periods such as stratospheric warmings. The errors will also be larger near the tropopause (200-300 hPa in mid-latitudes), as indicated in the table.
Westerly and Southerly wind components
Pressure Est. RMS error Est. bias _____________________________________________________ 1000 hPa 6.0 m/s magnitude <1 m/s tropopause 9.0 m/s 100 hPa 6.0 m/s magnitude <1 m/s 10 hPa (8 m/s) 1hPa (12 m/s) _____________________________________________________The wind errors in the troposphere are derived from OPD statistics for radiosondes, where available. The OPD gives very similar statistics for both westerly and southerly components. Since these errors are applicable to the two wind components independently, the RMS vector wind error will be a factor of approximately sqrt(2) larger. The 10 hPa and 1 hPa figures (in brackets) are indicative estimates; they are consistent with differences found between the Met Office TOVS analyses and the UARS correlative analyses. As with the temperature statistics these are global figures; errors will be larger in winter and dynamically active periods.
Geopotential Height
Pressure Est. RMS error Est. bias ______________________________________________ 1000 hPa (10 m) - 100 hPa (20 m) - 10 hPa (70 m) - 1hPa (100 m) - ______________________________________________These are rough estimates based on comparisons with the Met Office TOVS analyses and the NMC analyses. As with the temperature data, they do not take into account any systematic errors in the original operational (NOAA/NESDIS) satellite data, which are used in all three sets of stratospheric analyses
Vertical Velocity
The vertical velocity is a diagnostic quantity produced from the numerical model used in the data assimilation system. These data are subject to considerable uncertainty, and should be used with caution. Experience suggests that the diagnostic appears to capture the broad-scale vertical circulation, but there are unrealistic features at smaller scales.
2.5 Units
The units for the parameters present in the Met Office data files are tabulated below :
Parameter Units ________________________________________ Temperature T Kelvins Zonal wind component u m/s Meridional wind component v m/s Geopotential height z m Vertical velocity dp/dt w Pa/s ________________________________________
In November 2000, a new stratospheric data assimilation system was implemented, based on the 3D variational (3DVAR) data assimilation system used for the standard global forecast suite. The 3DVAR stratospheric system includes the assimilation of radiances (rather than temperature retrievals) from the operational polar orbiter satellites.
In October 2003, the assimilation model was changed to use a new semi-Lagrangian dynamical core, usually referred to as "new dynamics". The model is a 50-level configuration of the Unified Model; the tropospheric levels are the same as those used in the 38-level global forecast model, but additional stratospheric levels have been introduced. This model configuration has been designed to help to demonstrate the benefit of additional stratospheric levels on the processing of satellite temperature soundings. More details about the New Dynamics package is available on the Met Office New Dynamics factsheet.