Wiki chapter | OceanPrediction DCC

In the early days of numerical modelling of the various components of the Earth system, each component was treated individually. Figure 10.1 shows a representation of two systems, ocean and atmosphere, that run independently: the output of one system is used to “force” the other. The interface between the ocean and the atmosphere was considered a phenomenon that had to be modelled independently of the two media.

Figure 10.1. Traditional modelling platform characterised by Systems (S), like ocean model and atmosphere model, and inputs to each System (V).

This representation of the Earth system interactions is in some sense arbitrary. As the complexity of models grew, attempts were made to integrate the components more tightly, particularly in the field of climate modelling. Weather forecasting has a time scale of days to a couple of weeks (Lorenz, 1967) and, as new forecasts would be initialised regularly (typically every day), excessive diffusivity was never considered a problem. Making the early numerical weather prediction models conservative was therefore not a priority. The problem of conserving quantities such as heat, moisture, or momentum to avoid model drift, began to manifest itself only with the advent of long integrations of climate models. It became clear that long climate integrations of the atmosphere needed to also consider the impact of a (slowly) changing ocean, not least because the various climate components interact in nonlinear ways. This produces feedback loops that can fundamentally alter the state of each climate component. Numerical weather prediction models also needed to close the energy budget at the top of the atmosphere (or in the case of climate change, get that imbalance right). This led to the first attempts at coupling ocean and atmosphere models. The ice floating on the ocean and the soil in the ground were also separate from the ocean and the atmosphere. The latter was the first to be incorporated into more complex models, leading to the first coupled models.

Figure 10.2 shows a conceptual representation of systems that can interact through a “mechanism” called coupler. Figure 10.3 shows a more detailed and realistic representation of this coupling process.

Figure 10.2. Coupling modelling platform where Systems (S) communicate with each other through an interface code called “coupler”.

Figure 10.3. A schematic of the components (ocean, waves, etc.), the models (NEMO, WWIII, etc.), and the coupling exchanges between them, based on the system described in Lewis et al. (2019). Note the use of the coupler OASIS, the use of input forcing between Jules and the river flow model, direct coupling between Jules and the UM and direct forcing between the NEMO and ERSEM systems. A relatively simple coupled system (no ice) that includes 6 different models and 4 different approaches to coupling between them.

Theoretical challenges to producing skilful weather forecasts were noted early in the history of NWP. For example, Lorenz (1963) pointed to the phenomenon of sensitive dependence on initial conditions. This means that small changes in our current best guess of the atmosphere or ocean could lead to very large changes in the forecasts. As a consequence, skillful weather prediction is limited to a finite time horizon of around 1-2 weeks. However, this perspective tends to focus on synoptic scale atmospheric dynamics. When a numerical model of the atmosphere is coupled to numerical models of the ocean and other Earth system components, new timescales are introduced into the system. In such multiscale systems, fast growing errors tend to be associated with processes that evolve quickly but saturate at smaller scales (Harlim et al., 2005), while slower growing or decaying errors tend to be associated with larger scale oscillations (Penland and Sardeshmukh, 1995; Penland and Matrosova, 1998; Vannitsem and Duan, 2020).

DA is the process of integrating information from numerical models with observations derived from real world measurements. At operational centres, DA systems have typically been built for each Earth system component independently. Early efforts to produce coupled forecasts maintained this separation of components when applying DA to provide initial conditions (Saha et al., 2006, 2010, and 2014; Zhang et al., 2007), an approach that is now called WCDA. More recently, there have been efforts to treat the entire coupled Earth system as one state and update accordingly. This more integrated approach allows observations to have immediate influence across domain boundaries (e.g. the air-sea interface), and as such is called SCDA. There are also approaches that fall on the spectrum between these extremes, such as the CERA system at the ECMWF that applies different DA systems to the atmosphere and ocean but still allows influence across the air-sea interface via an iterative cycling over a moving 6-12 hour time window (Laloyaux et al., 2018).

Beyond these theoretical considerations, there are many technical complications involved in transitioning to coupled prediction. Many centres have developed monitoring and prediction tools independently for individual Earth components (e.g. atmosphere, ocean, land, waves, etc.). This is natural based on the historical context of their development and limitations on computing capabilities, but it has created an infrastructure within and across institutions that adds complexity to the task of unifying prediction systems. The major prediction centres are making progress towards an integrated approach by unifying software infrastructure for models and data assimilation capabilities, as well as providing opportunities to increase interactions among the development teams of each system component. Data formats for model output and observational data sets have not been fully standardised across the various Earth system domains, and so this adds further steps before seamless integration.

A very important practical limitation that has most certainly curtailed research and development in coupled prediction is the extreme demands it places on computational resources. The best performing applications for atmospheric prediction and ocean prediction have already been pushed to their limits of resource consumption. Acknowledging the fact that coupled systems can perform very differently at low resolutions versus high resolutions, there remain very few organisations with the resources needed to explore unanswered questions in coupled prediction at relevant resolutions for operational prediction. For this reason, there are efforts underway to identify methods to reduce the computational demands at bottlenecks within the cycled data assimilation and forecast systems.

Follow us

Coupled Prediction: Integrating Atmosphere-Wave-Ocean forecasting

10.1 Introduction to coupled prediction

References

Coupled Prediction: Integrating Atmosphere-Wave-Ocean forecasting

10.1 Introduction to coupled prediction

10.2 Coupling processes

10.2.1 Waves and their role in air-sea exchange

10.2.2 Land/sea exchanges

10.2.3 Air-sea exchanges across sea ice

10.2.4 The importance of air-sea exchanges during storms and other extreme events

10.3 Benefits expected from coupling

10.4 Ocean Information Services based on Coupled Frameworks

10.4.1 Establishing service needs

10.4.2 Identifying the required information