Evaluation and Intercomparison of Ocean Reanalysis

With the availability of satellite altimetry in near real-time since the launch of ERS-1 (1991) and TOPEX/Poseidon (1992), assimilation techniques have been developed in order to provide more realistic descriptions of the ocean dynamics with ocean models.

A first approach is to carry out reanalysis experiments, where models and assimilation are tuned to provide in the past the best description of the ocean circulation. Usually, the set of selected observations for assimilation is processed to remove possible biases and take into account differences in different types of observations. The set of forcing fields is also prepared in order to minimize errors, and long-term trend effects. Forcing estimates might merge observed and modelled parameters. During the experiment, successive intermediate runs might be performed in order to reduce errors identified in the meantime. And because state-of-the art ocean models are used, ocean reanalysis offer the most accurate description of the ocean for a set of "components" (i.e., choice of model and configuration, choice of observations and assimilation methodology).

In fact, historically in the ocean community, the intercomparison of ocean reanalysis have been the first where objective was to be compared to the ocean truth. In the framework of GODAE and CLIVAR, the GSOP project aimed to intercompare different reanalysis computed over one to several decades (Fig. 23.3). One of the goal being to offer synthesis on ocean state estimation for climate research (Lee et al. 2009a, b; Stammer et al. 2009). The idea being that multi-model ensemble approaches can be useful to obtain better estimates of the ocean. In practice, the GSOP objectives are (1) to assess the consistency of the synthesis through inter-comparison; (2) to evaluate the accuracy of the products, possibly by comparison to observations; (3) to estimate uncertainties; (4) to identify areas where improvements are needed; (5) to evaluate the lack of data that directly impacts the synthesis, and propose future observational requirements; (6) to work on new approaches, like coupled data assimilation.

Another use of ocean reanalysis is to provide initial conditions for seasonal and climate forecasts. This is a much more "close-to-real-time-operation" application. The idea is to offer for present time, or for few weeks before, the best possible ocean description together with its error estimates, in order to start coupled ocean/atmosphere forecast for seasonal prediction (Balmaseda et al. 2009).

Fig. 23.3 From (Stammer et al. 2009), Fig. 1, summarizing reanalyses taken into account by GSOP, sorted by forcing fields (green), type of ocean models (orange), assimilation methodology (pink), and résolutions (different blue)

For these two uses of reanalysis in ocean synthesis, errors listed in Table 23.1 are still relevant. One of the conclusions of the GSOP is that the full use of multiensemble assessment requires the detailed error information not only about data and models, but also about the estimated states. Figure 23.1 is illustrating that ocean estimates tend to cluster around methodologies and may not be independent from each other (see discussion in Stammer et al. 2009).

An important aspect of reanalysis accuracy, and the way intercomparison has to focus on, is their dependence on data to be assimilated in the past. Many ocean reanalysis are starting during the 1950s, when atmospheric reanalysis (NCEP and ECMWF ERA40) are available. Until 1978—first satellite with radiometer that provided Sea Surface Temperature (SST) with a global coverage—reanalysis can only rely on in-situ observations that are clearly under-sampling the ocean. As mentioned above, the ocean observability was strengthened with satellite altimetry in the 1990s. And since 2002, the ARGO array changed radically the ocean interior observability (e.g., Roemmich and Argo-Science-Team 2009). Note that atmospheric forcing accuracy has also been improved with satellite observations (radiometer for SST, heat content and exchanges in the atmosphere, and scatterometer for wind estimates). This lack of data in the past makes difficult any rigorous analysis of the ocean interannual and decadal variability.

Another accuracy aspect is linked with multi-data assimilation approaches. Nowadays, most of the assimilation methods use multivariate scheme that corrects their background15 fields using information from temperature, salinity profiles, altimeter sea level measurements, SST from satellite and in-situ observations. Some also take into account satellite sea-ice data, satellite gradiometry, current measurements deduced from current meters or drifters etc In these assimilation schemes, every observation is impacting the model parameters. For instance, temperature observation should correct the salinity field, but also the sea-level, and reversely. Which means accuracy and intercomparison assessments have to considered carefully the relations between the corrected ocean parameters, and the observation errors in the framework of each forecasting system. Moreover, "representativeness" of data has to be taken into account in the assimilation scheme. For instance, coarse resolution models (e.g. 2° horizontal resolution), can clearly not reproduce ocean fronts and water mass distribution as observed by gliders on scale of few kilometres.

The GSOP activity highlights most of these difficulties. A large number of studies have been performed using the reanalysis, among them, sea level variability, water mass pathways, variability of upper and mixed layer heat content, surface flux and run-off estimations, biogeochemistry, geodesy (see Lee et al. 2009a for more details). Note that most of the topics are similar to those studied with free simulations (as mentioned above). In particular the MOC, corresponding to the regulation of the meridional heat transport that affects climate variability was subject to several analysis. Figure 23.4 provides a synthesis for the North Atlantic meridional heat transport. One can notice, compared to Fig. 23.2 that some re-analysis provide more accurate estimates compared to hydrography (Ganachaud and Wunsch 2000) in the subtropical gyre. It means that since the DYNAMO project, models, associated with data assimilation succeeded in improving the representation of the ocean general circulation. However, the spreading of the six estimates in Fig. 23.4 are larger than error bars from (Ganachaud and Wunsch 2000). Moreover, the four reanalysis based on ECCO are similarly below the reference, showing here correlated errors in ECCO systems that will strongly affect an ensemble mean.

Figure 23.5 illustrates the difficulties in providing a robust evaluation of upper ocean heat content over more than 50 years. As mentioned earlier, the spreading before the 1970s seems associated with a lack of in-situ data. The ensemble standard deviation is reduced in the 1990s. However, since 2000, spreading appears again. This clearly raises the question of outliers with respect to the mean. Here, independent estimates should be used in order to evaluate reanalysis error levels. However, one can note a general tendency from all the time series: there is a clear warming of the upper ocean since the 1990s.

The GSOP effort will continue in the future. Multi-model assessment and ensemble mean approach has been identified as the only way to provide reliable

15 In the framework of assimilation the background is the state of the ocean model prior any correction by the assimilation method.

Fig. 23.4 North Atlantic meridional heat transport, from Armin Koehl GSOP présentation at the CLIVAR/GODAE meeting on ocean synthesis evaluation, held at ECMWF, UK, in August 2006 (http://www.clivar.org/data/synthesis/intercomparison.php). (Point and error bars correspond to estimates from Ganachaud and Wunsch 2000)

Fig. 23.4 North Atlantic meridional heat transport, from Armin Koehl GSOP présentation at the CLIVAR/GODAE meeting on ocean synthesis evaluation, held at ECMWF, UK, in August 2006 (http://www.clivar.org/data/synthesis/intercomparison.php). (Point and error bars correspond to estimates from Ganachaud and Wunsch 2000)

ocean estimates. Which means that (a) intercomparison will still be used to evaluate discrepancies, and (b) that effort is needed to characterize uncertainties from each system. Data assimilation techniques should provide more robust control on analysis16 and innovations17. In parallel, the ocean model community is still working on improvements (see Griffies et al. 2009 for a review). Moreover, work is still needed in order to reduce biases and make consistent historical dataset, but also clearly measure the impact of data type and availability on uncertainties (Heimbach et al. 2009).

The scientific assessment of these reanalysis will follow in a similar way. The main goal is still to characterise and understand the ocean medium and large scale patterns prior any further analysis. It means that the same ocean estimates analysed during CME or DYNAMO experiments will be evaluated first.

16 Here in the assimilation framework, the analysis is the production of an accurate image of the true state of the ocean at a given time, represented in a model as a collection of numbers. An analysis can be useful in itself as a comprehensive and self-consistent diagnostic of the ocean. It can also be used as input data to another operation, notably as the initial state for a numerical ocean forecast, or as a data retrieval to be used as a pseudo-observation.

17 The innovation is the discrepancies between observations and ocean model state, that is the vector of departures at the observation points.

12m-rm seasonal anom: NATL Averaged temperature over the top 300m

ukdp ukoi cfcs2

cfas2

ecco50y gfdl soda ecmfa -ecmfc ukgs ingv mri-

eccoSIO

cfasa mct2

- eccoJPLa

- eccoJPLc

- eccoMIT

ukdp ukoi cfcs2

cfas2

ecco50y gfdl soda ecmfa -ecmfc ukgs ingv mri-

eccoSIO

cfasa mct2

- eccoJPLa

- eccoJPLc

- eccoMIT

spread = 0.101

1960

1970

1980 Time

1990

2000

spread = 0.101

1950

1960

1970

1980 Time

1990

2000

Fig. 23.5 Seasonal anomalies of integrated [0-300 m] température in the North Atlantic Ocean. Figure from Balmeseda and Weaver, GSOP présentation at the CLIVAR/GODAE meeting on ocean synthesis evaluation, held at ECMWF, UK, in August 2006 (http://www.clivar.org/data/ synthesis/intercomparison.php). Color code are indicated for each reanalysis. Gray shaded area correspond to ensemble mean standard déviation

Was this article helpful?

0 0
Guide to Alternative Fuels

Guide to Alternative Fuels

Your Alternative Fuel Solution for Saving Money, Reducing Oil Dependency, and Helping the Planet. Ethanol is an alternative to gasoline. The use of ethanol has been demonstrated to reduce greenhouse emissions slightly as compared to gasoline. Through this ebook, you are going to learn what you will need to know why choosing an alternative fuel may benefit you and your future.

Get My Free Ebook


Post a comment