Oceanography modeling and data assimilation

Transverse actions: operational oceanography, modeling and data assimilation
Since a few years, a growing number of 3D hydrodynamics ecosystem coupled models operate in the Mediterranean to answer a large spectrum of questions going from operational oceanography to coastal management or more fundamental problems as the consequences of climate change on marine ecosystems. The constraints defined by the modeler are noticeably different if he is an actor of operational oceanography or at the opposite, if observations are needed through time series in order to build a model to simulate the response of an ecosystem to climate change. In the following, we discuss of the use of observations for research models and for operational oceanography.
Needs for research models
These models have to take into account regions with very diversified functioning as in the northwestern Mediterranean, the deep convection dominated by the winter mixing of nutrients, the Rhone dilution area under the control of continental inputs or the oligotrophic Northern Current.
At best, these models are validated using satellite images after a calibration stage based on 1D comparison with an open sea observation site (DYFAMED for the Med). Considering first the complexity of biogeochemical models (multi functional groups, multi elements) and second the spatial complexity of the represented areas, this approach is far to be adapted.
It is then important that an observation service could provide to modelers: 1) time series of parameters in different sites representative of the diversity of regions of a basin or a sub-basin and 2) time series allowing to characterize fluxes between these different regions. Both information is necessary to check models which are then able to provide quantifications of natural elements or contaminants budgets. However this approach requires to be efficient, preliminary studies to evaluate the representativeness of the observations sites and a frequency of observations compatible with the needs of model validation.
Considering the first point, the modeler needs to know if one observation can be considered representative of a large region or at the opposite if it is “local” compared to the horizontal resolution of his model. For example, a site submitted to intermittent input of freshwater of a small river will be disregarded by a modeler using a few kilometers resolution model while it will be of great interest for a littoral scale modeler. As a conclusion, it looks that each observation site should be the object of specific studies aiming at building an accurate understanding of its meaning.
The sampling frequency is obviously an important point. The time variability of measured parameters at the observation sites should be as much as possible characterized as well as the forcing responsible of this variability. An acceptable sampling frequency for each parameter could then be defined for each site. It is important to consider that the usefulness of an observation depends of the sampling frequency: for example, a monthly observation is well adapted to a climatological approach which requires a long time series before being interpreted while a daily observation allows to capture the meteorological scale or the oceanic mesoscale and is then very rapidly useful for modelers.
Finally a recommendation should be to harmonize as much as possible “observed parameters” and “model variables” (two names for the same concept). This is probably a challenge for the observations as the complexity of models is fast increasing thanks to the computers constant evolution. However, it is really important to maintain a dialog between both communities if we want to increase the use of observations and the quality of models.
Needs for operational oceanography
Progress in a wide range of ocean research and applications depends upon the prompt and dependable availability of ocean information products. The field of physical oceanography has matured to a point where it is now conceivable to combine numerical models and observations via data assimilation in order to provide ocean prediction products on various spatial and time scales, as well as long oceanic re-analyses time series. The latter contribute to the understanding of the long-term water cycle over the Mediterranean basin in terms of variability and trends. Compared to numerical modeling of the ocean circulation, the field of oceanic biogeochemical modeling is much less mature. Moreover, the strong nonlinearity of the biogeochemical models equations makes the data assimilation methods difficult to implement in these models. This explains the use of 1D models tuned to the “ecological area” of interest, as a first stage before the use of 3D models. The data assimilation problems and applications are different in physics and biology. Initial state estimation is frequently used in the field of physical oceanography to initialize new forecast simulations. As the biogeochemical models have various poorly known parameters and functional relationships, this aspect is generally less relevant and data assimilation can be an efficient way to estimate these parameters.
Because of the irregular and incomplete nature of the datasets relative to the spatial and temporal scales of interest, a good estimate of the ocean state does not depend solely on the performance of the data assimilation algorithm but also on the choice and the tuning of the model. Most of the target applications require good representation of, at least, temperature, velocity components and sea level. High resolution operational oceanography requires accurate depiction of mesoscale ocean features such as eddies and the meandering of currents and fronts and of upper ocean structure. Coastal applications require accurate sea level and cross-shelf transport estimates. Seasonal-to-interannual climate forecasts require a good representation of the upper ocean temperature and salinity fields. Decadal climate monitoring and research requires attention to the thermohaline circulation, among other things. Biogeochemical applications require to pay attention to the upper ocean physical parameterizations and the vertical transports (upwelling).
Operational oceanography is one of the key components of the GMES/MERSEA European integrated project aiming at building a capacity for ocean monitoring and forecasting at European level. This ocean monitoring and forecasting is now mature enough to consider the challenge of very high resolution raised by strong user demands and the transition to a strong operational activity oriented towards social needs (monitoring and expertise on extreme ocean climate events, new indicators for ocean pollution risk, …). But this activity is also fully dependant on real-time, continuous and accurate ocean observation networks, and associated actions for in situ and satellite sustained network are critically required. Data needed for model/assimilation systems can be separated into four main classes: atmospheric forcing (wind stress, specific humidity, precipitation, …), data for assimilation (altimetry, Argo, SST, ocean color, …), validation and verification data (hydrography, …) and ancillary data (climatology, reanalysis, bathymetry, …). Note, however, that the separation into data types is neither definitive nor unique (e.g., forcing data can be used as one of the controls on the assimilation process).
Operational oceanography offers among other applications an efficient way to control and adapt the acquisition design of the assimilated data. For instance, “adaptive observations” could reduce the analysis uncertainty by using the forecast to plan their short term deployment in the future. In this context, the gliders, which can be seen as steered profilers, can be redeployed from a given mission to sample, for instance, a particular event of interest. Thus they can supplement ARGO data effectively, as the latter can only marginally monitor the mesoscales and the circulation over continental shelves and margins. Data assimilation methods also allow Observing System Evaluations (OSEs) and Observing System Simulation Experiments (OSSEs), which are powerful and relatively inexpensive ways of assessing the impact of potential new observations, for determining the impact of losing current observing systems, and for refining and redirecting current observing practices. Further, with the advent of new forecast techniques, including high-resolution models, OSSEs would provide an excellent way of fine-tuning the observing system to the forecast needs. This last point could be very beneficial to both operational oceanography and process-oriented studies at regional, coastal, meso and submeso scales, including “green ocean” objectives.