Imperative 2: Data synthesis, analysis, reanalysis and uncertainty

Provide credibility to climate projections by understanding the past and present state of the ocean

Prepared by: D. Stammer, T. Palmer and K. Trenberth

Rationale

The changing state of the ocean is a critical component of the whole climate system. As such, the oceans have stored more than 90% of the energy content change of the earth system since 1955 and sea level is rising at a rate that appears now higher in the last 50 years compared with the first 50 years of the 20th Century (Bindoff, Willebrand et al., 2007). Perhaps more importantly, all of these changes are projected to continue and accelerate through to 2100 (Meehl et al., 2007). However, to provide credibility to projections of the future of the climate system, we have to understand the past, and what is happening now in the ocean.

Scientific Background and Major Challenges

Observations: Determining past changes of the climate system and inferring dynamics and feedbacks from the limited observations available during the last 50 years, let alone the last century, is challenging, especially due to the profound under-sampling of the ocean. As an example, providing answers to questions about past heat content and global sea level changes in the ocean is presently not feasible with great confidence prior to the most recent decade. The past decade has been characterised by increased sampling of the ocean, especially through the advent of altimetry, the Argo float system, moored buoy arrays, and other in situ and satellite components.

Reanalyses: The ability to analyse climate variability from limited climate data sets depends fundamentally on our ability to bring all available data into consistency with the dynamics of the ocean and the full climate system, as embedded in ocean or climate models. This enables use of the results for studies of variability or changes of either the ocean alone or of the fully coupled climate models, and for the initialisation of climate models. In the past, reanalyses in the atmosphere and syntheses/analyses in the ocean were performed individually dealing with each fluid in isolation. This work has led to important results and it is obvious that respective activities need to continue. At the same time, the challenge WCRP and climate research in general is facing in the future, is to deal with a truly coupled system, which includes the performance of syntheses/reanalyses and analyses in a truly coupled context and by doing so to improve our ability to initialise the coupled system. Respective work is underway under the auspices of CLIVAR’s Global Synthesis and Observations Panel (GSOP) and the WCRP Observation and Assimilation Panel (WOAP). 

Sustained observations: It will require a long-term continuation of satellite altimetry, and top-to-bottom Argo-like profile measurements of temperature and salinity globally, including regions under sea ice cover, to improve our understanding of regional and global sea level, heat content, as well as changes in the water cycles of the ocean. In addition regional changes in sea level are much larger than global numbers, implying that one needs a local tide gauge network on top of global estimates to answer urgent questions about sea level change and coastal security.

External forcing: To properly attribute ocean variability to natural variability or anthropogenic causes, we also need to gather information on changes of the external forcing by the atmosphere, cryosphere and other components of the Earth System (e.g. run-off and river discharge into the oceans). Attribution of ocean variations to changes in these forcing fields is fundamentally about the science of climate predictability, including for climate variability (e.g. El Niño-Southern Oscillation (ENSO), seasonal variability), as well as long-term climate change.

Strategic Plan

Ocean Syntheses: Assessing all changes occurring presently in the ocean requires a global and long-term observing system, capturing the full state of the ocean. As an example, assessing sea level changes, globally and regionally requires a detailed description of the changes in heat and freshwater content over the entire water column and over many decades, as well as changes in the mass of the ocean. All those measurements became feasible only recently, essentially since the advent of altimetry and through Argo and GRACE (before then large parts of the ocean where hardly observed even once over the last 100 years). However, the need is to estimate climate changes in the ocean over the last decades to centuries.

To obtain reliable estimates of long-term variations of climate indices from a limited database, all existing data should be used as best and as carefully as possible. Given the imminent problems associated with anthropogenic climate change, it is essential to learn as much as possible from the past. But using the existing data in the ocean for investigations of decadal and longer term climate variations requires the reprocessing of the entire climate data base to assure that uncertainties from the observations are reduced as best as possible (e.g. remove uncertainties in the XBT fall rate, biases in temperature or salinity profiles).

Global Ocean Heat Content (GOHC) of the upper 700m determined from different datasets, analyses, etc. The Domingues et al. (2008) record was determined using XBT data corrected for depth biases and illustrates the large difference from GOHC records estimated using uncorrected XBT data for the period 1970-1985

CLIVAR and WCRP must continue to show a significant leadership in this direction. An important step in obtaining the best possible estimates of the changing ocean as part of the climate system is then to analyse climate-quality ocean observations in a holistic approach by using all information available and analysing it in ways consistent with our dynamical understanding as embedded in ocean and climate models. Such an approach has to take in account uncertainties of observations as well as of models in which data are being assimilated.

Retrospective work is ongoing in the form of regional and global ocean synthesis. GSOP provides a comprehensive list of available ocean syntheses; several reanalyses of historical ocean observations have been constructed and are being evaluated through the CLIVAR GSOP intercomparison project. Available information from ocean syntheses is used to compute important climate indices, such as heat content of the global ocean and to start attributing changes to various sources, including erupting volcanoes through their effects on atmospheric aerosols.

Future ocean syntheses for climate research must be sustained in support of climate research and climate services. Not every ocean synthesis is useful for this purpose. Results depend on the assimilation approach; some are tailored for mesoscale predictions or for the initialisation of SI models. It is essential that ocean syntheses be accompanied by uncertainty measures (see below). Ultimately the community should compile ocean syntheses from multi-model, multi-approach ensemble estimates that are supposed to be of better quality than any estimate alone. However, the science has to go a long way to reach that goal, which as a pre-requisite requires prior as well as a posteriori error information.

Coupled Data Assimilation: Projecting the ocean state over the next decade or even centuries requires running a coupled climate forecast system, which requires information about the initial/present climate state as well as boundary conditions to the ocean. Indeed one of the goals for an ocean observing and synthesis system is to provide proper initial conditions for ocean and climate forecasts on seasonal to decadal time scales. For predictions of a season to a year or so, the ocean state is most critical although information about sea ice extent, soil moisture, snow cover, and state of surface vegetation over land are all important.

Initialisation of a coupled system has three main components: i) the observing system leading to climate records; ii) the assimilation method leading to estimates of the initial conditions; and iii) the coupled climate model leading to predictions. Uncertainties in each element will impact the skill of predictions. In practice, initial conditions for the ocean component of the coupled model are usually produced in an ocean-only context and are being fed into the coupled model subsequently. This will inevitably lead to mismatches in the physics of the initial condition and the coupled model and major uncertainties in predictions (SI and DEC/CEN) originate from uncertainties in the models initial condition and from errors in coupled models (including the initial coupling shock and subsequent adjustments).

Ocean data assimilation is used operationally in several SI forecast centers around the world to initialise seasonal forecasts with coupled models, and it is from that experience that decadal prediction efforts will build. Temperature and salinity from ocean syntheses have already been used to initialise models for decadal forecasts (Smith et al., 2007, Pohlmann et al., 2009). In this way, modelling groups without data assimilation schemes can perform initialised climate predictions. However, mismatches in model physics leads to initialisation shocks; while errors in coupled models lead to fast degradation. To overcome those, the climate community needs to work both on improving initialisation procedures, but also on improving coupled climate models to better represent and simulate present day climate variability. In particular, the assimilation needs to be performed in the coupled model framework to obtain proper initial conditions. At the same time a coupled assimilation system will lead to reduced model errors through parameter optimisation. In fact it is highly likely that confronting models with observations in this way will lead to improved understanding of error sources and to model improvements.

Fully-coupled data assimilation schemes that take advantage of covariance’s between ocean and atmospheric variables to generate an optimal estimate of the climate system are expected to offer the greatest forecast skill for SI and decadal predictions. Such schemes are under development with some encouraging results (e.g. Zhang et al., 2007). The related activities are to improve and initialise climate models and make “seamless” ensemble climate predictions for the oceans for the time horizons of 30 years. Developing and evaluating coupled assimilation and forecast systems for decadal predictions should be a core activity of WCRP over the next decade and beyond.

Uncertainties: A recent intercomparison of ocean syntheses reveals a wide spread of results and highlights the need for specifying proper uncertainties for model and data errors. These uncertainties in the ocean state become especially relevant for understanding climate change in the ocean, including the joint relationship between atmospheres and oceans, and for understanding future projections of the ocean state in the context of the regional change and with short time horizons. A significant effort is therefore required to understand uncertainties in estimates of climate indices, how they depend on the underlying method and how they affect aspects of prediction.

In analyses of the ocean and the atmosphere, uncertainties arise from several elements:

1. Observing system;
2. Data processing;
3. Model errors;
4. Approaches to estimate the ocean or the climate state; and
5. Initialisation procedures of coupled models.

Historically the subsurface ocean has been very sparsely observed, and some of the data appear to be significantly biased (Domingues et al., 2008), making the development and testing of ocean initialisation schemes difficult. For instance, the non-stationary nature of the ocean observing system, particularly due to the paucity of salinity data as well as XBT data only going to 500 m, can give rise to spurious decadal variability making the assessment of forecasts based on such analyses difficult. More emphasis has to put on the last 10-15 years when sufficient amounts of data are available over the global ocean to infer changes.

An ultimate use of ocean syntheses is to use information to initialise the ocean in coupled climate forecast models. The problem of model error is critical for decadal prediction, since the sub-surface ocean state associated with the initial condition may be significantly different than the climate of the free running coupled model.

A substantial uncertainty can also emerge from the specific choice of estimating initial conditions. Various choices can be made ranging from computationally simple to mathematically rigorous but computationally demanding. All approaches usually can be summarised as “filters” and “smoothers”. A significant effort is required to understand the impact of the approach on the result as well as the validity of the approach for climate timescales.

Finally understanding uncertainties not only in models or data (and realising that data errors can be significant) but understanding the a priori uncertainties of estimated climate indices need to be of high priority for WCRP, since without that information the use of ocean or atmospheric analyses is only of limited value, regardless of application.

Building links to IGBP carbon, biogeochemistry, and ecosystems: Many processes in the ocean depend fundamentally on the physical state of the ocean. An important link between CLIVAR and other programmes, such as IGBP, is therefore to provide best possible estimates of the changing physical ocean to the carbon and ecosystems communities so as to better constrain their problems. An important question to investigate in this context is that of the uptake of carbon dioxide by the ocean. The ocean circulation does play a fundamental role in this context and obtaining best possible descriptions of the ocean climate state during the past, present and future is of great importance for understanding the climate response under future emissions scenarios. Ultimately we want to perform joint syntheses between the physical and the carbon communities so as to better understand the ocean circulation but also to better understand the observed changes of carbon dioxide in the ocean and at the same time utilise the observational constraints in both. This calls for a close link between ocean and coupled climate syntheses and carbon syntheses.

Climate change on time scales of decades and centuries has profound impacts on the ecosystems of the ocean. Ongoing assessments of the physical, biogeochemical and ecosystems of the ocean are important and should be implemented by WCRP jointly with IGBP.

Back to About Us