CT 2

Core Theme 2:  Assessing sources of uncertainty in ocean analyses and forecasts

Climate forecasts are inevitably subject to uncertainty (or ‘error bars’) of various types. Uncertainty arising from the specification of initial conditions, and from parameter uncertainty in the current generation of climate models, can be explored using ensemble forecasts, and this will be done in CT4. Here, CT2 considers more ‘structural’ sources of uncertainty that are generic to all currently practical forecasting systems. Such uncertainty arises from the limitations of current observations and analysis systems (WP 2.1), and from generic limitations in climate models, specifically the lack of explicit modelling of fresh water input from the Greenland ice sheet (WP 2.2), and poor resolution of key, small-scale components of the THC such as sill overflows, eddies and boundary currents, which may play a key role in redistributing fresh water inputs and so modulating the THC (WP 2.3). Science outputs of this core theme will feed into an overall assessment of priorities for the development of observing systems and models, required for a quasi-operational decadal THC prediction system (Deliverable D6.1.5). This will be one of the major scientific and technological outputs of THOR.

The overall goal of CT2 is to understand to what extent the above limitations restrict our ability to forecast the THC and related climate variables, and to identify priority areas for future development in observations, modelling and data assimilation.  

Lead: Steffen M Olsen (DMI), Co-lead: H. Drange (NERSC)

WP 2.1 Assessing uncertainty in ocean analyses

The workpackage evaluates existing ocean reanalyses and simulations, looking for key processes that drive inter-annual to decadal variability in key THC components and identifying the reasons for discrepancies among currently available analyses. In order to make forecasts (WP4) it is necessary to produce an estimate of the current state of the ocean to provide initial conditions. Two approaches have been used:

• an ‘analysis’ uses (generally sparse) ocean observations to constrain a physically based model, producing a combination of model and observed fields and hence a more or less dynamically consistent estimate of the ocean state. A number of such ocean analyses have been produced in recent years (including under EU programmes ENACT, ENSEMBLES). However, many of these analyses were designed for seasonal forecasting, with a focus on variability associated with El Nino, and it is not clear how well the THC is constrained by these analyses (see www.clivar.org/organization/gsop/synthesis/synthesis.php ).

• An alternative method of estimating the state of the ocean is to run ocean models forced by the history of atmospheric surface fluxes or boundary layer properties, without assimilating ocean observations (‘simulations’). This approach can work well for seasonal El Nino forecasting, because much of the variability in the tropical Pacific is driven by wind stress variations. However for decadal THC variability the approach is unproven. On the other hand, such models are now being run at much higher resolution than is possible for the analyses.

WP 2.1 exploits a number of existing ocean analyses and hincasts, which have been produced recently by our project partners to evaluate the different methods, with a focus on the THC. The evaluation will use independent observations (i.e. not assimilated into the models), including  estimates of the THC around 25°N, and estimates of Atlantic inflow and overflow fluxes. Close contact will be maintained with CT3, which will be extending the available timeseries of many of these observations. A new reconstruction of sea surface salinity, from 1970-2005, will be produced as part of WP2.1 and used as a further test of the analyses/simulations.

The evaluation will go beyond simply comparing the different state estimates, and through a set of focused sensitivity experiments with the DePreSys analysis system will endeavour to establish what are the key factors required to estimate the current THC state, e.g. resolution of eddy and boundary current processes; choice of covariances in assimilating observations; critical levels of data density (e.g. Argo); relative value of different data sources in constraining the THC. These experiments complement studies of data impact on forecast skill in WP 4.2.

Lead: Marie-Noelle Houssais (UPMC-LOCEAN)

WP 2.2 Assessing modelling uncertainty: Greenland ice sheet

WP 2.2 considers the potential impact on THC forecasts of the incomplete representation of the Greenland ice sheet mass balance in current climate models – specifically, could additional fresh water input from Greenland result in a substantial error in THC forecasts, possibly including the crossing of a threshold to a permanent THC shutdown?

The feedbacks between ice sheet and climate will be investigated with a novel coupled atmosphere-ocean-ice sheet model (ECHAM5-MPIOM). The model yields a realistic climate and runs without flux corrections. The model will be forced with different scenarios of anthropogenic greenhouse gas concentrations. By comparing with identical simulations without active ice sheets, the effect the potential melting of the Greenland ice sheet will be assessed. These will be compared with a single existing run using a different climate model (HadCM3) coupled to a different ice sheet model, in order to compare the different roles of modelling uncertainty and forcing scenario uncertainty.

Potential feedbacks of the THC on the Greenland ice sheet mass balance will be investigated by forcing the ocean component towards a collapse of the THC. The anticipated cooling and reduced precipitation are expected to strongly affect the mass balance of the Greenland ice sheet and thus the net melt water input into the North Atlantic.

Plausible rates of melt water input will be added to a range of AOGCMs (BCM, DKCM, ECHAM5-MPIOM, HadCM3, HadGEM1, IPSL-CM4), starting in 2050 and following the IPCC SRES A2 forcing scenario, to estimate whether deficiencies in the representation of the Greenland ice sheet may be resulting in an underestimate of the likelihood of major THC change in present-generation AOGCMs. Initially, idealised patterns of fresh water input will be used, based on existing ice sheet models. Some runs will use scaled-up amounts of fresh water to investigate the impact of extreme melting scenarios. Later in the project, refined fresh water patterns will be used, derived from the ECHAM5-MPIOM model runs above.

Scientific outputs from this WP will be an assessment of the importance of Greenland meltwater input for THC forecasting, and hence of the priority of including interactive ice sheet models in climate forecasting systems.

Lead: Uwe Mikolajewicz (MPG-M)

WP 2.3 Assessing modelling uncertainty: Ocean fresh water distribution

This WP addresses the question “Do climate models properly represent the processes by which fresh water input to the ocean modulates the THC?” In this WP a coordinated set of model experiments will be carried out in which the same fresh water perturbations that were used in WP 2.2 are applied to a range of coupled and ocean only models, at resolutions ranging from 2° to 0.1°. This will be the first time that such coordinated experiments have been performed spanning climate models and high resolution ocean models, DRAKKAR, MOEN and NANSEN-MICOM (ocean only) and BCM, DKCM, ECHAM5-MPIOM, HadCM3 and HadGEM1 (coupled). In order that there is no need to run dedicated controls for the fresh water experiments, they will be run parallel to existing simulations: from 1965-2005 with the ocean only models, and either pre-industrial control runs or ‘All Forcings’ 20th Century simulations for the same period with the coupled models.

The goal of the experiments in this WP is not to simulate a particular event, but rather to elucidate the processes by which the ocean distributes fresh water input to modify the THC, and the sensitivity of those processes to model formulation and resolution. To inform later analysis, existing integrations of the ocean models from 1965-2005 will be analysed to establish the processes controlling fresh water and THC variations, during the early part of the project (while the new experiments are being run), with a focus on the role of meso-scale processes in determining overflows and subpolar deep convection.

As a by-product of these runs it will also be possible, by analysis of common features in ocean only and in coupled model responses, to develop further understanding of the importance of atmospheric feedbacks on THC variability. Close liaison will be maintained here with the analysis of long coupled model runs in WP 1.1.

In order to investigate further the role of the freshwater signals exported from the Arctic Ocean and the Nordic Seas on the THC, the patterns of high latitude atmospheric or oceanic anomalies which most influence the variability of the overflows will be determined from available ocean reanalyses. These anomaly patterns will be applied stochastically to force ensembles of short (typically 10-20 years) model runs and, by comparison with a control run, to estimate their impact on the high latitude ocean and on the THC.

Because the fresh water forcing runs in the WP use a common forcing with the runs in WP 2.2 it will be possible, towards the end of the project, to assess the relative importance of uncertainty in the magnitude of Greenland fresh water input and poorly resolved ocean processes for THC forecasting, leading to further recommendations on priorities for development.

Lead: Arne Biastoch (IFM-GEOMAR)