The research and objectives of Bjerknes Climate Predition Unit
Weather forecasting is a well-established scientific practice serving society. Numerical climate prediction in contrast is in its infancy, but is potentially of equal societal importance: allowing society to better cope with climate fluctuations, reducing risks, optimising resource use, and guiding adaptation.
Climate predictability arises when more slowly varying components of the climate system, like the ocean, significantly influence the atmosphere, especially when two-way interaction reinforces variability, or when external factors impact climate. The former is the case for seasonal prediction of the El Niño Southern Oscillation (ENSO).
While there is promise that climate can be predicted a decade in advance, especially for the North Atlantic region, predictions remain highly experimental. Thus, climate prediction is one of the main frontiers in climate research – ranked a Grand Challenge by the World Climate Research Programme.
The vision of the Bjerknes Climate Prediction Unit (Bjerknes CPU) is to skilfully forecast climate in the Atlantic-Arctic sector from a month to decades into the future for the benefit of society.
To fully realise this we need to address three key challenges – the Research Activities (RA) of the unit: The mechanisms (RA1) underlying climate predictability need to be understood and properly represented in climate models. Data assimilation (RA2) methods need to be developed to optimally use observations to constrain predictable dynamics. The theoretical and practical limits of climate prediction (RA3) need to be understood.
The Bjerknes CPU assembles world-class expertise emerging from the Bjerknes Centre for Climate Research (BCCR; established as a Norwegian Centre of Excellence in 2003) in climate dynamics, modelling and predictability, data assimilation, observations, and in computing required.
The best possible climate prediction system for the Atlantic-Arctic sector can now be realized through a dedicated unit – the Bjerknes CPU – that provides the long-term perspective and resources required to meet these three key challenges.
Key aspects of Bjerknes CPU
First, identifying the mechanisms underlying predictability requires a deep understanding of the key processes. Observations reveal a predictable statistical relation between sea surface temperature (SST) in the northern seas and several parameters important for society: continental climate over Norway, commercially valuable Barents Sea cod stock, and Arctic winter sea ice extent.
However, open questions exist on the role of ocean circulation and how the ocean influences continental climate. We will deepen the understanding of the mechanisms underlying predictability in this region and use this knowledge to improve the Norwegian Climate Prediction Model (NorCPM). NorCPM combines Ensemble Kalman Filter data assimilation (EnKF) and the Norwegian Earth System Model (NorESM).
Second, we will advance the application of data assimilation to properly account for climate dynamics. A first version of NorCPM shows that ocean variability in the North Atlantic can be constrained assimilating SST, leading to skilful forecast in the Eastern Nordic Seas a few years ahead. The system has been extended to assimilate hydrographic profiles and sea ice concentration (in strongly coupled sea ice – ocean assimilation). We will address key theoretical issues in order to constrain all components of the climate system simultaneously.
Third, to demonstrate the societal benefit of improved climate prediction the unit will initiate the first quasi-operational climate prediction system in Norway. Confidence in operational climate services will be built through novel methods to assess skill and prediction limits. Appropriate data flow and collaborations are established to make the predictions available to the broader researcher- and climate service communities.
Research Activity 1: Mechanisms giving rise to climate predictability
Mechanistic understanding is a prerequisite for identifying predictable relations in observations and models. RA1 aims to identify and model the interactions within and among climate system components that give rise to predictability. We will also consider external factors (greenhouse gases, volcanic eruptions, aerosols, solar forcing) that can add predictive skill. RA1’s basic approach is to identify the relevant physical mechanisms in observations and reanalysis (RA2), test them in model experiments, and improve their representation in NorCPM.
T1.1 Predictability here is hypothesized to be rooted in ocean inertia, and more specifically in poleward ocean heat transport. We will investigate the mechanisms and time scales involved in the propagation of ocean heat anomalies along the Gulf Stream’s extension toward the Arctic, how the anomalies interact with the atmosphere, their impact on sea ice predictability, and eventual influence on continental climate variability.
T1.2 It is unclear to what degree air-ice-ocean coupling can explain the observed climate variability. We will study large-scale air-ice-ocean coupling; interaction over sharp SST and sea ice fronts, and on synoptic scales; and identify coupled modes of climate variability and the impact of external forcing.
T1.3 A large component of predictability in the Atlantic-Arctic sector might arise from teleconnections from the tropics, the North Pacific, sea ice, snow cover and land surface conditions. Advanced atmospheric diagnostics will be used to study the importance and mechanisms of the various teleconnections.
T1.4 Models appear to underestimate predictable dynamics and suffer from large model biases35. Motivated by findings of T1.1-T1.3 (and other RA), improvements will be implemented into NorCPM to alleviate such problems. A key first step will be to develop a stratosphere-resolving version of NorCPM25. We will also consider model resolution, and implement new parameterisations.
Research Activity 2 – Data assimilation for improved climate prediction
Advancing the formalism of data assimilation (DA) in the field of climate prediction improves forecast skill. DA is the entire process that provides the best-possible state estimate based on observations, a dynamical model and statistical information. DA has been key in advancing the understanding and exploiting the predictability features of complex nonlinear systems such as the weather. RA2 will extend the application of a state-of-the-art DA class of methods, the ensemble DA, to climate prediction. Ensemble-DA methods have the greatest potential because of their flexibility and straightforward implementation11.
T2.1 The climate system includes complex, coupled phenomena over wide, separated, spatial and temporal scales (atmosphere, ocean, land surface, cryosphere). DA procedures, on the other hand, are mostly designed to deal with a single dominant scale of motion or under the assumption of weak coupling37. BCPU will study strongly coupled DA, which allows information from the observations to be propagated across all model components simultaneously and consistently. In this task we will make extensive use of theory and utilize a hierarchy of low-order coupled models38. The leading average approach39, and more generally the error propagation in coupled dynamics, will be studied in the face of the spatial and temporal scales separation among atmosphere and ocean.
T2.2 Sampling errors limit the efficiencty of ensemble methods in high-dimensional systems. A notorious, somehow ad-hoc, manner to handle this issue is know as localisation, in which only observations within a specified isotropic local domain surrounding the analysis point are considered. We will formulate a new non-isotropic local domain surrounding the analysis points are considered. We will formulate a new non-isotropic, multivariate and spatially varying localisation method in order to handle updates across model components of separated spatial scale.
T2.3 We will combine in-depth analysis of the scale interactions (RA1, T2.1) and improved localization method (T2.2) to develop strongly coupled DA ocean-sea ice-atmosphere assimilation within NorCPM. The strongly coupled DA will be refined through twin experiments and then with observational data.
T2.4 Reanalysis and DA statistics from the ensemble will help improve understanding of the coupling mechanisms and reduce model error. We shall perform two long-term reanalysis; one using a stratosphere- resolving strongly coupled DA ocean – sea ice (NorCPM1) and a second with an improved model (T1.4) and strongly coupled DA ocean – sea ice – atmosphere (T2.3) DA (NorCPM2).
Research Activity 3 – Climate prediction limits
Development of innovative approaches to assess and attribute predictive skill and quantification of the impact of systematic model errors. Idealized and retrospective forecasts experiments have provided estimates of predictability of climate1. However, large systematic model errors, poor initialization, and limited understanding of mechanisms make these estimates uncertain.
We will deliver enhanced NorCPM predictions and estimates of prediction limits by using mechanistic understanding, numerical experimentation with improved models and data assimilation, and multi-model predictions.
T3.1 We will develop a quasi-operational capability by performing once a year real-time forecasts with NorCPM. For skill assessment, retrospective predictions will be repeated with NorCPM1 and NorCPM2. The experimental forecasts will be made freely available to stakeholder partners.
T3.2 New ways of attributing sources of prediction skill is needed to have a better understanding of model behavior and its errors. We will develop new diagnostic tools describing relevant dynamics (RA1), and use these and standard metrics to better understand and quantify the skill of NorCPM. NorCPM skill will be contrasted against skill of observational-based statistical predictions, developed together with RA1.
T3.3 We will assess the impact of systematic model errors and shock on prediction skill firstly by diagnostic analysis of NorCPM and freely available multi-model predictions; and secondly by repeating prediction experiments with (3D ocean35,44) bias corrected versions of NorCPM.
T3.4 We will develop a hierarchy of models1, beginning with slab-ocean atmosphere coupled models, stochastically forced ocean models, introducing simplified representations of ocean circulation and coupled ocean-atmosphere variability (T2.1). Thus, we will provide estimates of climate prediction limits for different temporal and spatial patterns of climate variability.
The Trond Mohn Foundation (formerly The Bergen Research Foundation) is funding the project with 30M NOK over 5 years, in addition to 30M NOK from the partners.