Report on the Joint SPARC Workshop on Data Assimilation and Stratospheric Winds

Banff, Canada, September 12-15, 2005

S. Polavarapu, Environment Canada, Canada (saroja.polavarapu@ec.gc.ca)

T. G. Shepherd, University of Toronto, Canada (tgs@atmosp.physics.utoronto.ca)

Background

Data assimilation is the process whereby observations are combined with model forecasts to produce an optimal estimation of the state of the atmosphere, known as an analysis. While the primary motivation for data assimilation is to provide an initial condition for weather forecasts, analyses also provide a record of the global state of the atmosphere that includes all relevant variables.  Thus analyses can be used for process studies and, in some cases, to examine long-term changes.  At the same time, data assimilation can help to identify observation biases or sudden changes in observation quality.

For these reasons, data assimilation contributes significantly to SPARC science.  Yet there are many issues with the quality of assimilation products in the stratosphere (Rood 2005).  SPARC Report No. 3 (2002) documented many of these.  Especially in the tropics, differences between analyses can sometimes exceed the seasonal or interannual variability, making them of little value for studies of atmospheric variability (Figure 1).  There are also significant polar temperature biases, which give uncertainty to studies of polar stratospheric clouds — this will be the subject of a future SPARC Report.  Many studies (e.g. Schoeberl et al. 2003) have highlighted the severe errors that can arise when using assimilated winds in off-line transport models.

Figure 1: Climatological seasonal cycle of zonal mean zonal wind at the equator from various analyses at 30 hPa (bottom left), 10 hPa (bottom right), 5 hPa (top left), and1 hPa (top right).  (From SPARC (2002).) For better resolution, please contact the SPARC IPO

These problems arise because of the special challenges of data assimilation in the stratosphere.  In the troposphere, the models are mature and the availability of plentiful observations from numerous independent observation types helps to separate model and observation bias.  In the stratosphere, in contrast, models are known to exhibit severe biases (Pawson et al. 2000) such as the “cold pole problem” and the lack of a quasi-biennial oscillation (QBO), while at the same time there are relatively few observations, especially of winds, and little redundancy between those that do exist.  In addition, the interest in stratospheric chemical transport has exercised the analyses in ways they were not intended for.  In particular, stratospheric applications are often limited by errors that involve long time scales, which are not adequately reflected in the error covariances that underlie operational data assimilation.

Because the process of data assimilation requires as inputs not only measurements and model forecasts, but also estimates of their accuracy (the mean and covariance of their errors), knowledge of the underlying physics of each measurement type, and of a model’s numerical discretization and physical parameterizations, is required.  In addition, data assimilation itself requires expertise in statistics and estimation theory.  Finally, the outputs of assimilation must be assessed not only statistically, but also in terms of physical realism.  Thus, data assimilation is a multi-disciplinary activity that requires the involvement of many different research communities to be effective: the measurement community, modellers, assimilators and theoreticians who understand the physics and chemistry of the real atmosphere.

TOP

The SPARC Data Assimilation Working Group

The goal of the SPARC Data Assimilation Working Group (DAWG) is to advance data assimilation science in areas relevant to SPARC science.  While other coordinated assimilation activities certainly exist (e.g. WGNE, THORPEX), the middle atmosphere context provides a rather different perspective on the process of data assimilation.  For example, the interest in long time scales and the ubiquitous presence of gravity waves in the mesosphere challenges assimilation schemes designed for the troposphere (e.g. Polavarapu et al. 2006).  The SPARC community already contains the wide variety of expertise needed to advance the science of middle atmosphere assimilation, and the SPARC organization provides the means to achieve the goals of the Working Group.

The goals of the SPARC DAWG are many.  Firstly, the group will collect and document information on data assimilation systems.  This is important information for the SPARC community members who use assimilation products for diagnostic studies.  At the same time, the Working Group will encourage users of assimilation products to include multiple analyses in studies of processes or long-term changes, to emphasize the differences between analyses from different systems.  The data assimilation community requires information on measurement data sets (meteorological and chemical) such as quality, availability, and the software needed to access it.  The Working Group intends to gather such information and make it accessible on the SPARC Data Center website.  An important feedback to data assimilators is obtained from process-oriented assessments.  One obvious example of such feedback is the identification of errors in assimilated winds on long time scales obtained from examining age of air and trajectory calculations.  Since assimilated products have particular difficulties in polar regions and in the tropics, physically based diagnostics of these regions will be the starting point.  Because the SPARC community contains many users of reanalyses, the Working Group can provide guidance for reanalysis efforts.   Finally, the Working Group will liaise with space and other agencies (e.g. IGACO, GCOS) on SPARC data needs through the SPARC SSG.

How will all these goals be achieved?  At the very minimum, an annual workshop will be held.  This will serve to gather assimilators interested in the middle atmosphere, and from which information on the various systems can be documented.  By inviting experts in the dynamics and chemistry of the true system (the middle atmosphere), a more physically based assessment or criticism of assimilation products can take place.  A specific theme can be identified to help select a few experts for a given workshop.  The workshops can also serve to connect assimilators with the users of assimilation products.  Thus attendance at these workshops should include assimilators, dynamicists and chemists with an interest in atmospheric processes, and users of assimilation products.  Besides specialized workshops, periodic special sessions at large conferences will be organized to facilitate interaction with other research communities.

Reports such as this one will be written to disseminate information on the Working Group’s activities, or to provide an overview of current or new ideas in assimilation to the wider SPARC community.  In addition, the value of an article on the issues in middle atmosphere data assimilation for a general audience (e.g. Bulletin of the American Meteorological Society) is being considered.

The workshops will include more than scientific presentations.  The workshops will also serve to identify the need for collaborative work, such as intercomparison projects.  Such projects will advance the science of assimilation through the assessment of many different schemes with standard, physically-oriented diagnostics.  The themes of intercomparison projects, or of the workshops themselves, will identify outstanding issues in the field.  Proposed themes for upcoming workshops include: transport, water vapour, the tropical tropopause layer, and gravity wave drag.  The latter topic refers to the goal of using the assimilation process to help identify the missing gravity wave drag force that parameterizations try to account for.  An outcome of such research could be the estimation of parameters needed by gravity wave drag parameterizations.  Finally, an ozone analysis intercomparison project has been proposed to continue the coordination started by the ASSET project.

The 2005 Joint SPARC Workshop

As noted above, observations of stratospheric winds are very limited.  There are some historical observations from rocketsondes and from the HRDI instrument on UARS. Rocketsondes are extremely sparse and geographically biased, while HRDI’s stratospheric measurements had poor signal-to-noise and thus were heavily smoothed. Above 10 km, operational measurements come only from radiosondes, which have an inherent geographical bias and leave enormous data gaps; even then, only 20-30% reach 10 hPa (~35 km).  While some information on winds can be obtained from satellite-derived temperatures, this relies on a balance between the mass and wind fields, which is a powerful constraint in the extratropics but a far weaker constraint in the tropics.  The unbalanced component of the flow also is believed to increase strongly with altitude (Koshyk et al. 1999; Shepherd et al. 2000).  Furthermore, operational temperature observations come from nadir sounders, which have poor vertical resolution and significant bias problems in the upper stratosphere.

The poor quality of stratospheric wind analyses in the tropical lower stratosphere is illustrated by comparisons with direct wind measurements, where those exist.  In Figure 2, NCEP reanalyses are compared with winds measured from the ER-2 aircraft, flying at around 20 km altitude.  In the extratropics, the agreement is excellent over a wide dynamic range.  In the tropics, in contrast, the agreement is seen to be very poor.  In Figure 3 (colour plate IV), ECMWF analyses are compared with winds measured from long-duration balloon flights at around 60 hPa.  Again, the agreement is found to be much better in the extratropics than in the tropics.  In Figure 4 (colour plate IV), the directly measured winds are filtered to exclude periods shorter than 12 hours, thus filtering the inertia-gravity waves which are not represented in the analysis (Hertzog et al. 2002).  Now the agreement in the extratropics is remarkable, but the filtering has little impact on the discrepancies in the tropics.

	Figure 2:  Scatter plot comparing total horizontal winds in the lower stratosphere around 20 km altitude from in situ aircraft measurements (horizontal axis) with NCEP/NCAR reanalyses (vertical axis) in the wintertime (a) extratropics and (b) tropics.  Note the different scale in the two plots. (Figure courtesy of Paul Newman, NASA GSFC.) For better resolution, please contact the SPARC IPO

Figure 3:  Comparison of direct wind measurements from long-duration balloon flights at 60 hPa (Vial et al., 2001) with ECMWF analyses at the same point in the flight path.  The left column shows equatorial measurements (taken in 1998), right column shows high-latitude measurements (taken in 2002).  The top row shows zonal wind velocity and the middle row, meridional wind velocity; in both cases black denotes the balloon measurements and blue the ECMWF analyses. The bottom row shows the differences: black for zonal wind, and orange for meridional wind.  (Figure courtesy of Albert Hertzog, LMD.) For better resolution, please contact the SPARC IPO

Figure 4:  The same as for Figure 3, but with the directly measured winds filtered to exclude periods shorter than 12 hours. The difference with Figure 3 is attributed to inertia-gravity waves. For better resolution, please contact the SPARC IPO

TOP

Yet analysed winds are highly important for SPARC science.  Many important problems of stratospheric variability, including the QBO and solar effects on climate, involve tropical winds which are poorly characterized at present.  This makes it difficult to validate stratospheric climate models, whose representation of the QBO and Semi-Annual Oscillation (SAO) tends to be very model-dependent.  Also, present analyses appear to be of inadequate quality for long-term transport studies.  This compromises our attribution of ozone changes, and also introduces biases in chemical data assimilation.  Although better assimilation methods may improve the situation, there remain fundamental limitations that are ultimately tied to the quality of the wind observations.

While there is some prospect of inferring winds from observations of chemical species using four-dimensional assimilation methods, such observations are only sensitive to the component of the wind parallel to the tracer gradient.  Unfortunately, tracer gradients tend to align perpendicular to the wind (the “stirring” effect).  Moreover most tracer gradients tend to be quite slack in the tropics, where wind observations are most needed. In any case, any such derived wind products would require validation.

Thus, it seemed timely to assess the current knowledge of stratospheric winds, the science questions that require such knowledge, and the prospects for improved knowledge in the future.  This led to the 2005 Joint SPARC Workshop on Data Assimilation and Stratospheric Winds.  There were a total of 37 participants: 14 from Europe, 13 from Canada, 8 from the USA, 1 from India and 1 from Japan.  Of this number, only 16 were from the data assimilation community; the others represented climate modelling, diagnostics, measurements, theory, and process studies. This sort of cross-fertilization between the data assimilation and the more ‘physically oriented’ communities is an exciting development, and as noted above is key to the success of the SPARC DAWG. The results of the Joint Workshop are now summarized, organized by theme.

Transport errors

One of the most important issues concerning middle atmosphere data assimilation is the inability of analysed winds to adequately represent the Brewer-Dobson circulation.  Such a failing limits their use in chemistry-transport models (CTMs) when the scientific interest lies in processes spanning more than a few weeks.  Recent work at the ECMWF suggests that there is some hope for improving analysed winds in this respect.  S. Polavarapu, in an overview presentation on issues in middle atmosphere data assimilation, showed a slide of recent age-of-air calculations by B. Monge-Sanz and M. Chipperfield (University of Leeds) (Figure 5).  The ages of air computed from operational analyses (using 4D-Var) were far older (and more realistic) than those computed from ERA-40 analyses (using 3D-Var). Since 4D-Var analyses tend to be in better balance than 3D-Var analyses (e.g. Gauthier and Thépaut 2001), the former are less noisy and therefore produce less spurious mixing of tracers.  Furthermore, using an even more recent version of the ECMWF system(also 4D-Var) resulted in ages of air which are approaching observed values in the northern extratropics.  This latter result is likely due to the improved bias correction of ATOVS data, and the introduction of an improved balance in the initial forecast error covariance matrix (based on the nonlinear balance and omega equations)  (A. Simmons, personal communication).

Figure 5:  Mean age of air at 20 km altitude from simulations of the TOMCAT CTM (solid coloured lines) using different ECMWF meteorological analyses to drive the model, compared with the mean age of air derived from ER-2 aircraft observations of CO2 and SF6 (dashed lines).  2σ error bars have been included for the observations. (Figure courtesy of Beatriz Monge Sanz, University of Leeds.) For better resolution, please contact the SPARC IPO

Because of the transport errors associated with assimilated winds, K. Miyazaki and colleagues developed a system which first nudges ECMWF winds into a GCM, which then drives a CTM.  The idea is that the GCM will be kept close to analyses, while model dynamics will reduce imbalance.  Miyazaki found that nudging of wind fields created imbalances between the nudged fields and the GCM fields, while nudging of both temperature and wind fields was slightly better in this regard, although biases in the GCM remain despite the nudging.  This system will be used for an extensive (1957-2002) ozone reanalysis using ERA-40 fields to clarify the roles of mean and eddy transports on the interannual and decadal variability of constituent distributions.

Dynamical variable assimilation

The assimilation of satellite data remains both a major motivation and a major challenge for middle atmosphere data assimilation.  One challenge arises from the fact that satellite measurements are generally related to atmospheric variables integrated over some path, whether vertical (nadir) or slant (limb).  Y. Rochon noted that background error correlations can introduce small-scale structure into the analyses that is not necessarily physical.  The specification of variances can also cause problems; e.g. the assimilation of total column ozone can result in the displacement of the climatological ozone maximum from the lower stratosphere to the upper troposphere, if background error variances reflect the variability of the upper tropospheric jets.  While ad hoc procedures such as the removal or modification of covariances can reduce undesirable effects, Rochon developed a more general mathematical framework for choosing how to spread analysis increments along a path of integration.

Since analyses may not be in balance, as far as the forecast model is concerned, gravity waves may be generated in a geostrophic adjustment-like process when a model integration is started from an analysis.  In the past, with 3D assimilation schemes, a separate filtering (or initialization) scheme was implemented.  D. Sankey investigated the impact of several schemes on the middle atmosphere and found that a digital filtering of the full fields reduced vertical dispersion of parcels in trajectory analyses, but also eliminated the diurnal tide.  An incremental analysis updating scheme was found to preserve the model’s tide but to enhance vertical dispersion of parcels.

Since air irreversibly enters the stratosphere through the tropical tropopause, it is important to correctly capture upwelling in the tropical troposphere in order to simulate stratospheric moisture.  However, moisture is an exceedingly difficult variable to assimilate.  H. Thornton showed that unrealistic analysis increments are produced in the upper stratosphere as a result of improper background error covariance specification.  Modification of covariances can reduce the problem.  The choice of moisture variable for assimilation can also be important.  Using relative humidity results in more Gaussian forecast errors than, say, the logarithm of specific humidity, but relative humidity will be adjusted even in the absence of moisture data if temperature data is assimilated.  The Dee and da Silva (2003) approach was found to improve the assimilation up to about 5 hPa but did little to help above this level.  The non-Gaussianity of errors in the moisture variable was addressed using the Holm (2002) variable, but preliminary results showed that the impact on stratospheric analyses was minimal.

Background error covariance specification is important not only for moisture, but for all variables.  Thus proper covariance specification remains an area of active research.  D. Jackson showed results from assimilation experiments with the Met Office’s research stratospheric model (60 levels up to 84 km) comparing background error statistics derived using (i) the NMC-method and (ii) Yves Rochon’s method which is based on model climatology (with the diurnal tide removed).  Rochon’s method provided better verification against analyses, but work is still ongoing.  S. Pawson showed that by using inhomogeneous and anisotropic covariances (longer correlation lengths in the zonal direction in the tropics), the GEOS-4 system was better able to capture the easterly to westerly transition of the QBO by allowing the equatorial radiosonde wind observations to have more weight.  The rationale is that these sparse radiosonde observations are believed to be representative of the zonal mean wind.  (Moreover there are reasons to expect zonally elongated error correlations in the tropics.)  Without this modification, analyses would lag several months behind the Singapore wind observations in switching from easterly to westerly winds. With the modification, the analyses appear to be much better, based on comparison with Strateole balloons and with the QBO signal in ozone seen at Nairobi.

Y. Jaya Rao noted that cirrus clouds and aerosols are both observable with lidar measurements.  Vertical velocity measurements with a VHF radar over a tropical station show vertical wind reversal within the clouds suggesting enhanced mixing.  Such measurements may be very useful for determining the radiative impact of cirrus clouds on climate.  M. Salby discussed the impact of convection on the temperature structure of the tropical tropopause.  Representation of this process is a challenge for data assimilation, because of the small length scales involved both in convection and in the mean state response.

TOP

Chemical data assimilation

Chemical data assimilation remains an important activity at major weather centres around the world.  As D. Jackson noted, the main motivations for assimilating ozone include the improved assimilation of radiance measurements, improved radiative heating rates, the ability to exploit new satellite measurements, and the capability to assess proposed new satellite instruments.  At the Met Office, a 3D-Var system based on an N48L50 version of the model is used for ozone assimilation with a Cariolle-based chemistry parameterization.  The University of Cambridge is also developing a coupled chemistry model approach within a version of the Met Office Unified Model, and some preliminary results of UARS MLS ozone assimilation were shown by M. Parrington. 

NCEP has been assimilating SBUV ozone data operationally since 1997 with a 3D-Var scheme.  Now ozone data from EOS-Aura, consisting of OMI total column and profile measurements, are being assimilated in test mode and will later be assimilated operationally.  S. Zhou showed some results that indicated that OMI data is dominant due to its high density.  Assimilation of high resolution profile data from MLS is in progress and assimilation of HIRDLS data is also planned. 

R. Ménard described the operational assimilation of surface ozone data at the Canadian Meteorological Centre, as well as plans for on-line chemistry assimilation and transport with the operational forecast model and complex chemistry models for both the troposphere and stratosphere.  B. Bregman presented examples of current work at KNMI involving the assimilation of SCIAMACHY total column measurements of ozone.  This has been running operationally since January 2004.  KNMI uses an off-line CTM with a sequential (Kalman filter-type) assimilation scheme.  H. Elbern described an ambitious new assimilation scheme SACADA (Synoptic Analyses of Chemical constituents by Advanced Data Assimilation) in development at a consortium led at the University of Cologne and designed to run operationally at the German Space Agency (DLR-DFD). The system employs the German Weather Centre's  (DWD) global forecast model GME online, based on an icosahedral grid and including a complex chemistry module. A 4D-Var approach was taken to ensure a-temporal consistency of chemical constituents. Sample results from the assimilation of ten species from MIPAS were shown.

Middle atmosphere measurements from the recent ENVISAT, EOS-Aura and Odin satellites have motivated work outside of operational centres as well.  A. O’Neill showed that MIPAS ozone improved analyses between 100 and 10 hPa when assimilated using 3D-Var with the ECMWF model at T159L60.   J.Rösevall used an off-line isentropic transport model with a Kalman filter-type algorithm to assimilate Odin SMR data and estimate descent rates and ozone loss rates within the polar vortex.

Because data assimilation is a very expensive operation, the additional cost of chemistry models remains an important issue.  M. Bourqui presented a new chemistry solver which uses pre-computed nonlinear transfer functions that represent average diurnal chemistry.  This new approach could be very useful in a data assimilation context.  The issue of the level of complexity of chemistry needed in the assimilation context was a topic of considerable discussion.  R. Ménard presented results of a surface ozone assimilation which showed better analyses (in terms of standard deviation and bias against observations) without chemistry than with it.  However, the forecast biases were greatly reduced with chemistry. In addition, W. Lahoz showed results from an intercomparison study (under the auspices of the EU-funded ASSET project) of five ozone assimilation models and found that having complex chemistry did not necessarily improve the ability to capture ozone depletion in the polar vortex.  The same study also found that variants of the Cariolle parameterization gave rather different instantaneous rates of change of ozone and led to different ozone distributions.  The discussion led to a proposal of a theme for a future SPARC-DAWG workshop: the minimum level of complexity of chemistry needed for data assimilation. 

Quality of current analyses

G. Manney assessed the impact of differences between analyses for the study of transport in the 2002 Antarctic Stratospheric Sudden Warming (Manney et al. 2005).  While the different analyses exhibited overall qualitative agreement in a coarse sense, there were very significant differences in detail, as revealed in sensitive diagnostics such as effective diffusivity.  These differences are also reflected in the sensitivity to the choice of analysed data sets of other diagnostic studies such as Match-estimated ozone loss, age of air, and stratosphere-troposphere mass fluxes — all of major importance for SPARC science.  In terms of dynamical fields, while the zonal-mean zonal wind at 60N and 10 hPa agrees between all analyses, the NCEP/NCAR reanalysis is clearly unreliable for stratospheric temperature, while ERA-40 temperatures are unreliable for PSC studies because of spurious vertical oscillations.

M. Giorgetta addressed the question of whether a model needs to be able to simulate a self-generated QBO in order to properly represent a QBO through assimilation.  This is not so clear; once the zonal mean winds are reasonably close to the observations, the wave fluxes in the model, both resolved and parameterized, will presumably respond to the shear layers.  On the other hand, the model will certainly have a bias that will depend on the state of the real atmosphere, and if these wave fluxes are poorly represented then the model response to QBO winds will be incorrect.

E. Manzini focused on variability in the Arctic wintertime vortex associated with ENSO, contrasting its representation in ERA-40 and in the NCEP/CPC analysis.  Above 10 hPa, there are substantial differences in amplitude (though not in structure).  In general, ERA-40 has a stronger planetary wave 1 signature in the upper stratosphere — though much less so before 1980. 

In a discussion session, the following consensus was reached for our knowledge of stratospheric winds.  In the extratropics, up to 10 hPa there is a reasonable overall agreement between analyses, and between analyses and radiosondes.  Whether the extent of agreement is suitable for long-term transport or mixing is not clear.  Vortex mixing appears to be reasonably well represented, however.  But overall the quantification of the quality of analyses is process dependent.  As a general rule, we would like the differences between different analyses to be less than the interannual variability.  For vertical wind, the only validation data is from tracers, which can be difficult to interpret because of mixing.  It is recommended that analyses should use and provide, diabatic heating rates. Above 10 hPa, analyses differ more substantially.  In this region, there are no direct wind measurements, and no other data constraints on the TOVS radiances.  Furthermore, the model impact on the analysis, either from parameterized gravity-wave drag or from the location of the model lid, is stronger.  Thus, in this region there are likely to be significant biases in both models and observations, leading to unreliable analyses.

In the tropics, the situation is much worse.  With respect to the zonal wind, the SAO is not well characterized (models may force their own SAO quite strongly), and the observations are of temperature, not winds.  The QBO is reasonably well characterized below 10 hPa qualitatively, but not quantitatively.  Above 10 hPa, better characterization is needed. Since models don’t always simulate a self-generated QBO, periodic biases are present.  There is also no validation of the longitudinal structure of the QBO.  With respect to the meridional or vertical wind, we do not know much about the quality of the analyses, but they are likely to be quite poor.  For these fields, tracer observations may provide the only validation opportunity.  On the other hand, a better knowledge of the zonal wind should help constrain the meridional wind, as the horizontal flow can perhaps be expected to be non-divergent to a first approximation (even in the tropics).

TOP

Unbalanced dynamics and mesospheric assimilation

E. Källén described work addressing the assimilation of equatorial waves — Kelvin waves, mixed Rossby-gravity waves, equatorial Rossby waves and inertia-gravity waves — exploiting knowledge of the spatio-temporal structure of the waves.  For ECMWF, 60-70% of the forecast error variance in the tropics is associated with equatorial waves, so it is important to attempt to represent them in analyses.  The study was motivated by the ADM-Aeolus instrument (see below), which will only measure line-of-sight wind and so requires assimilation in order to obtain the vector wind. Rather surprisingly, 4D-Var does not do much better than 3D-Var, and observing the height field is not so helpful in getting the wind components.

While the middle atmosphere includes the mesosphere, most assimilation work is concentrated on the stratosphere.  However, interest in estimation of the mesosphere and lower thermosphere has been increasing as operational centres raise their lids.  (The U.S. Navy has model versions with lids at 85 and 100 km.  ECMWF will raise their lid to 0.01 hPa and the Met Office will have a version of their model with a lid at 110 km.)  With mesospheric data assimilation it becomes possible to couple the neutral atmosphere to “space weather”.  Since the mesosphere is dominated by wave disturbances generated from below (i.e. tides, planetary, gravity and Kelvin waves), an assimilating model should ideally include the wave source region.

R. Lieberman and D. Ortland both addressed the subject of data assimilation in the mesosphere and lower thermosphere (MLT), where unbalanced motions are strong.  In this respect there may be parallels with the ocean, where inertial oscillations are the dominant signal of current-meter measurements.  The subject is timely because of the launch of the TIMED spacecraft in December 2001.  A dominant feature of MLT variability is thermal tides, which cause strong aliasing problems for most satellite observations (which are slowly precessing in local time) and can lead to a misrepresentation of tides and transient planetary waves.  While ground-based observations can resolve the diurnal variability, they need to be optimally combined with satellite observations to produce a complete picture (Figure 6).  Thus, for TIMED, ground-based measurements were an integral part of the science plan and represented a “fifth instrument”.  For this strategy to work as part of assimilation, however, the underlying model must have a realistic representation of thermal tides.

T. Matsuo found that one challenge of using an ensemble Kalman filter for assimilation in the mesosphere-lower thermosphere region was the collapse of the ensemble spread due to insufficient model variability compared to observed variability. This may be related to the use of a mechanistic model, which forces planetary waves at the tropopause.  A covariance inflation factor, estimated from observations, was a proposed solution. 

Figure 6:  (a)  Retrieval of the migrating diurnal tide at 100 km modelled in NCAR-WACCM (thick solid curve) by a sequential estimator based upon TIDI sampling alone (thin curve).  The retrieval is hampered due to TIDI undersampling in local time. (b) As in (a), including sampling by 7 ground-based radar wind profilers.  The retrieval is subtantially improved due to the higher sampling rate of the ground-based stations, located at a sufficient number of sites (> 4) so as to resolve the migrating semidiurnal tide. (c) Longitude versus universal time sampling of satellite (negatively sloped curves) and ground-based observations (vertical curves).  (Figure courtesy of Ruth Lieberman, Colorado Research Associates.) For better resolution, please contact the SPARC IPO

Problems requiring improved wind measurements

T. Dunkerton highlighted the upper stratosphere/lower mesosphere as a region of particular interest.  This is where variances tend to maximize and much interesting dynamics occurs:  planetary waves break down violently, barotropic and inertial instabilities occur, and mean flows evolve rapidly, e.g. during the spring and autumn transitions.  Thus, it is an exciting frontier for middle atmosphere data assimilation.  There is some evidence that in polar regions, stratospheric disturbances are associated with mesospheric precursors (e.g. stratospheric warmings with mesospheric coolings).  In order to understand the dynamics of such connections, and whether there is any causal element, it is necessary to separate upward and downward components of planetary wave fluxes.  This will require accurate measurements of winds in this region.  In the tropics, there is a medley of interesting dynamical phenomena involving winds, which will be a challenge to properly characterize from observations.  Inertial instability is clearly occurring, but is difficult to observe because of its rapid timescale and small (as yet unknown) vertical scale.  (In models, it tends to occur on the smallest resolved vertical scale.)  Equatorial waves, both of intermediate frequency (1-3 day period) and intermediate scale (zonal wavenumber 10), play a crucial role in the QBO.  Observing the properties of inertia-gravity waves requires high vertical resolution in the lower stratosphere, and high temporal resolution in the upper mesosphere.

In related comments, T. Shepherd emphasized the increasing dominance of inertia-gravity waves (IGWs) with increasing altitude, leading to drastically shortened autocorrelation times for winds, for example.  This represents a challenge for data assimilation in the upper stratosphere and mesosphere.  IGWs are present even in relatively coarse resolution global models, although only represent the tip of the iceberg with respect to the real atmosphere.  By the same token, any measurement technique will only observe part of the IGW spectrum, a constraint that needs to be borne in mind when interpreting the measurements (Alexander, 1998).  Considerable thought will need to go into understanding forecast and observation “errors” under these conditions.  In the tropical stratosphere, while conditions for a “semi-geostrophic” balance (which would include Kelvin waves) appear to exist, it seems likely that the flow is nevertheless highly imbalanced because of direct forcing of unbalanced motions.

New wind instruments

There are two new wind instruments that promise to increase our knowledge of stratospheric winds in the near future.  P. Ingmann discussed ADM-Aeolus (Stoffelen et al. 2005), which will be launched by the European Space Agency in 2008 to address the most important measurement need identified by WMO (2001): namely a global coverage of direct wind measurements.  ADM-Aeolus will measure line-of-sight winds using an active optical system (lidar), and is focused on the troposphere.  However, it now appears that it will be possible to obtain measurements up to 30 km, making these measurements of great interest for SPARC.  As noted earlier, the use of ADM-Aeolus measurements will require the use of data assimilation to get information on the two horizontal wind components, a concept that was built into the instrument design from the outset.

The other stratospheric wind instrument is SWIFT, which has been approved for launch by the Canadian Space Agency in the 2010 time frame.  SWIFT is a passive imager that will measure vector winds in the 20-50 km altitude range, and will thus not be dependent on data assimilation — although there is certainly great interest in the prospect of using SWIFT winds for assimilation, or at least validation.  A. Scott presented the SWIFT instrument concept and its implementation on the proposed Chinook satellite, while Y. Rochon discussed its expected error characteristics.  Assessing the potential value of measurements from a new satellite instrument is always a challenge.  C. McLandress presented the results of simple calculations, using the Canadian Middle Atmosphere Model to provide a surrogate atmosphere, in order to determine the required measurement errors for SWIFT in order to meet its science goals (e.g. meridional residual circulation, zonal mean wind, amplitude and phase of equatorial planetary waves).  An example is shown in Figure 7.  The results show the expected dominant winter hemisphere branch of the poleward Brewer-Dobson circulation in the upper stratosphere, and the two-cell structure in the lower stratosphere.  If the goal is to capture the dominant structure of the meridional transport, then at 20 km s = 3 m/s is optimal, s = 5 m/s is acceptable, and s = 10 m/s is unacceptable, while at 40 km s = 5 m/s is optimal and s = 10 m/s is acceptable.  W. Lahoz described the results of a more sophisticated Observation System Simulation Experiment (OSSE) performed for ESA to assess the impact of SWIFT winds relative to the current operational suite of observations.  Not surprisingly, the greatest impact was found to be in the tropics.  Finally, S. Pawson briefly described some lessons learned from an OSSE performed by Ricky Rood’s group at the NASA DAO for the SWIRLS instrument, which was proposed for EOS-B (eventually EOS-Aura) in the early 1990’s, but then never flown (in part because of this OSSE).  However, the problems in interpreting OSSE’s were noted, because OSSE’s are so dependent on the assimilation system used for the study and therefore may not be relevant to the real data when the instrument is actually flown.

Figure 7:  Estimation of SWIFT measurement requirements for the meridional residual circulation v* using pseudo-observations computed from the Canadian Middle Atmosphere Model for July: upper stratosphere (top panels) and lower stratosphere (bottom panels).  The blue curve denotes the ‘truth’; the black lines show results of ten different estimations, for one month of simulated SWIFT measurements, with Gaussian random noise added to both the meridional and zonal wind components before computing v*.  The noise standard deviation is given in the lower right-hand corner of each panel. (Figure courtesy of Charles McLandress, University of Toronto.) For better resolution, please contact the SPARC IPO

TOP

There are also some prospects of observing winds in the mesosphere.  This is of interest for data assimilation, now that several assimilation models include the mesosphere within their model domain. W. Ward discussed plans for a potential Canadian instrument called WaMI, and the possibility of an Atmospheric Dynamics Mission.

In a discussion session, the measurement requirements developed for SWIFT while it was being considered by ESA as a Stratospheric Dynamics mission were discussed (Table 1).  Those requirements were developed on the basis of certain science goals.  Other potential science goals for SWIFT were raised in the discussion.  One was the possibility of assessing whether current analyses are adequate to reproduce the tracer distributions that are arising with new measurements during periods of rapid evolution in the polar vortex (e.g. sudden warmings).  Another was characterization of the subtropical mesospheric jet, the bottom part of which would be seen by SWIFT.  A third was the potential for providing a constraint on extratropical upper stratospheric temperature, which would help with the known AMSU bias problems.  Finally, the question was raised (but not answered) as to whether there would be any value in complementary ground-based measurements.

International Polar Year

The international polar year (IPY) refers to an extensive, multi-national, interdisciplinary period of observations covering the 2007 and 2008 calendar years.  SPARC has prepared a proposal for participation in the IPY which was described in the Newsletter No. 25 (July 2005).  SPARC’s contribution is entitled “The structure and evolution of the stratospheric polar vortices during IPY and their links to the troposphere.”  The main idea is to document the dynamics, chemistry, transport and microphysical processes of polar vortices, highlighting the themes of ozone depletion, and the links between the stratosphere and troposphere and between the stratosphere and mesosphere.  This is a step toward the ultimate goal of understanding the connection between the polar climate and the stratosphere.  The proposal involves the coordination of satellite and ground-based campaigns, as well as specific initiatives to increase understanding of major features and processes.   The SPARC Data Center will play an essential role in this effort by archiving key data.

Because the SPARC IPY proposal aims to document the current state of the stratosphere in the Arctic and Antarctic, data assimilation can play an important role in this effort.  The IPY effort was discussed during the workshop and a number of recommendations were made.  Firstly, the SPARC-DAWG will contribute to the overall SPARC proposal by archiving assimilation products from many centres and research groups for the 2007-8 period.  The SPARC Data Center would be used as a repository for the collected products.  The precise products and the participating centres and research groups will be identified in the coming months.  In order to attract many participants, the products requested will have to be as flexible as possible.  Thus they will likely be entirely at the discretion of the data provider.  Other details such as data formats will be determined in collaboration with the SPARC Data Center.  Such a repository of analyses and forecasts over the IPY year can serve as a resource for process-oriented studies of the polar regions.  A second way that SPARC-DAWG proposes to assist the overall SPARC IPY activities is to consider how to combine multi-model assimilations into a probability density function.  Just as an ensemble of model forecasts can be used to indicate forecast uncertainty, an ensemble of analyses may be able to indicate analysis uncertainty.  At the very least, a range of possible states can be provided.  The question is then whether value can be added by combining products in an objective way.

Besides the two activities outlined above, the discussion led to some recommendations for SPARC.  Firstly, the group recommends gathering and archiving special purpose data sets for validation of assimilation products.  This would be very helpful for participants who will provide assimilation products to be archived by SPARC.  The DAWG also requests an attempt to define observation requirements for surface measurements relevant to data assimilation.  Similarly, it would also be useful for assimilation if gaps in measurements (e.g. during the polar night) were identified, and then filled.  What is the potential for in situ observations in this respect?  Finally, the group would like to use IPY activities to argue for the extension of existing satellite measurements such as Odin, TIMED and SCISAT-1.

It will be important to link to other IPY activities related to this one.  SPARC has endorsed the POLARCAT IPY proposal which deals with the impact of long range aerosol transport on climate through an examination of pollutant transport into and out of the Arctic.  Another related activity is ORACLE-O3 which focuses on stratospheric ozone measurement and the processes leading to ozone loss.  Because ORACLE-O3 does not include an assimilation activity thus far, the repository of analysis products created by SPARC IPY work could be of interest.  Finally, WGNE has a polar vortex forecasting activity which may already have coordinated groups willing to contribute to the SPARC data repository. 

Other SPARC-DAWG activities

In the short term, the IPY will be the main focus for the DAWG.  Nevertheless, a few other activities are planned for the coming year.

What has become very clear from analysis intercomparison exercises in the past (e.g. SPARC 2002) is that analyses differ from each other, especially in the middle atmosphere.  Therefore, considerable caution must be taken when comparing modelling results or measurements to a single analysis.  What is preferable is to plot several different analyses, to indicate the level of uncertainty in the analyses.  Because many intercomparison studies have already been done, it would be useful for the climate research community to be able to easily access the results.  Therefore, the DAWG plans to create a “clearing house” for intercomparison studies located on the SPARC Data Center website.

While the “clearing house” just described can serve the climate research community, what is needed for the middle atmosphere data assimilation community is the identification of a standard set of physically oriented diagnostics that can be used to validate assimilation products.  G. Manney has agreed to help coordinate this activity, by providing appropriate diagnostics for polar processes.  The Tropical Tropopause Layer (TTL) is another region where assimilation products can be improved.  Relevant diagnostics for this region should assess water vapour, temperature, clouds, and diabatic heating rates.  SPARC’s CCMVal activity (see report this issue) has already identified many diagnostics for their model intercomparison exercise.  Some of these diagnostics may also be useful for assimilated fields, and therefore the CCMVal diagnostics can serve as a starting point.  At the same time, the DAWG can contribute to CCMVal by helping to assess the analyses used for some of the CCMVal diagnostics.

The presentations of Matsuo, Lieberman and Ortland reflect a growing interest in the estimation of the mesosphere and lower thermosphere.  The SPARC Data Center will be used to add links to mesospheric measurements (pre-existing websites for the TIMED satellite mission and ground-based measurements, to start with).  Similarly, the SPARC Data Center will provide a link to the website for the ASSET ozone intercomparison project described by Lahoz.

Next meeting

The next SPARC-DA working group meeting will be held during the week of 18-22 September 2006 in Noordwijk, the Netherlands.  As with this meeting, the goal is to focus on process-oriented evaluation of middle atmospheric (and upper tropospheric) analyses.  For this, experts from outside the assimilation field are necessary, and will be invited to participate.  The planned themes for the next meeting include transport errors on long time scales, and the tropical tropopause region.  Both themes will highlight problems with assimilated fields in the tropics.

Acknowledgement

We would like to acknowledge financial support for the Joint Workshops from the WCRP, the Canadian Space Agency, and Environment Canada.

References

Alexander, M. J., 1998. Interpretations of observed climatological patterns in stratospheric gravity wave variance. J. Geophys. Res., 103, 8627-8640.

Dee, D. P. and da Silva, A. M.  2003. The choice of variable for atmospheric moisture analysis.  Mon. Wea. Rev., 131, 155-171.

Gauthier, P. and Thépaut, J.- N. 2001.  Impact of the digital filter as a weak constraint in the preoperational 4D-Var assimilation system of Meteo-France. Mon. Wea. Rev., 129, 2089-2102.

Hertzog, A. et al., 2002. Quasi-Lagrangian measurements in the lower stratosphere reveal an energy peak associated with near-inertial waves. Geophys. Res. Lett., 29, Art. No. 1229.

Holm, E. 2002. Revision of the ECMWF humidity analysis: construction of a Gaussian control variable. Pp. 1-6 in Proceedings of ECMWF/GEWEX Workshop on Humidity analysis, 8-11 July 2002, Reading, U.K.