The Bluelink ReANalysis (BRAN) is a multi-year integration of OFAM that assimilates observations using an Ensemble Optimal Interpolation (EnOI) data assimilation system called BODAS (Bluelink Ocean Data Assimilation System). The goal of BRAN is to provide a realistic quantitative description of the three-dimensional time-varying ocean circulation of all physical variables (temperature, salinity, sea-level and three components of velocity) for the last few decades.

BRAN is always a work in progress. Many runs of BRAN have been completed, each incorporating improvements to the model, data assimilation schemes and observational data sets.

This video is of BRAN2020. It shows how true to the observations the modelling system are, which enables users of the forecast system (provided by our partners at the Bureau of Meteorology) to see its strengths as well as its weaknesses. For example, when you watch the video, you will often see quite a strong correspondence between the motion of the surface drifters (the magenta arrows heads looping around) and the modelled movement of the water (as revealed by advection of warm and cold waters). We also do this quantitatively, of course, but tables of statistics do a poor job of portraying the richness and complexity of the circulation of the ocean. A critical eye will notice that the flows and temperatures in the model do not always evolve continuously, but sometimes suddenly change. This is the assimilation cycle at work. In recent years, we have been working to reduce these shocks, as have all our peers around the world. When the model is able to remain perfectly true to the real world and anticipate perfectly, several days in advance, what the satellites and other systems eventually observe, our job will be done.

Animation of daily Sea Surface Temperature from BRAN2020 in the greater Australasian region from July 2017 to June 2021. The trains of black chevrons indicate the position of surface drifters that are drogued so that their motion is dominated by surface current rather than wind. Note that the temperature scale changes depending on the region shown.

Why do a reanalysis?

For the near-real-time model runs that are used to produce forecasting, much observational data may not be readily available due to transmission delay or lack of thorough quality control.

In addition to providing a high fidelity 3D record of the global seas, the BRAN model outputs support the creation of climatologies of mean ocean conditions, across all variables and locations.  These provide an atlas of phenomena, together with indications of variability and uncertainty that can be used for planning operational activity and understanding the impact of long term change.

Results and uptake

  • Results from BRAN2020 are described in Chamberlain et al. 2021.
  • A recent independent study of global reanalyses, Russo et al. 2022, showed that BRAN2020 outperforms FNMOC, HYCOM and GLORYS in the Southern African region, especially for mixed layer depth.
  • An article in Nature by Li et al. 2022 used BRAN2020 to study drivers of ocean warming in western boundary currents.
  • Other recent applications of BRAN2020 data include research into ocean dynamics (e.g., Chen et al. 2022) and fisheries (e.g., Schilling et al. 2022).

Each iteration of BRAN has been subject to peer-review, ensuring that methods and techniques used under Bluelink remain at the cutting-edge of the world’s best practice. Bluelink is also a founding partner of the Global Ocean Data Assimilation Experiment (GODAE), which now falls under GODAE OceanView (; meaning that Bluelink also benefits from a close network of international collaborators, and a healthy competition between our ocean forecasting peers around the world.


BRAN data is publicly available. Details can be found on our Data Access page.

More information


BRAN Version history

Bluelink partners have performed a series of Bluelink ReANalyses (BRAN) experiments over many years. Each BRAN experiment has involved incremental improvements, and many have included step-change improvements. A summary of the development of BRAN follows:

  • BRAN1: the first Australian ocean reanalysis involved the development of a new model configuration and a new capability in ocean data assimilation. BRAN1, described by Oke et al. (2005), was far from perfect – but it opened a new chapter in Australian oceanography.
  • BRAN1p5: the most obvious problem with BRAN1 was the ocean initialisation. The BRAN1 fields were too noisy, and the experiment didn’t assimilate satellite SST observations. BRAN1p5, documented by Oke et al. (2008), was a shorter reanalysis, that assimilated SST and largely eliminated the very noisy fields evident in BRAN1.
  • BRAN2p1: using a setup that was very close to BRAN1p5, BRAN2p1 addressed a few small errors (in topography and forcing fields) and was longer than BRAN1p5, spanning over 14 years. Schiller et al. (2008) used BRAN2p1 to quantify the ocean variability around Australia.
  • BRAN3p0: still based on BRAN1p5 (and 2p1), BRAN3p0, described by Oke et al. (2013), further improved the ocean initialisation and included many incremental improvements to the data assimilation system.
  • BRAN2015: the first Australian, near-global ocean reanalysis, adopted a new model configuration (with high-resolution across all longitudes) and a new data assimilation code ( replaced the system described by Oke et al. (2008). BRAN2015 was only a short experiment (about 3 years initially), but it was kept up to a few months behind real-time until about 2018. BRAN2015 wasn’t documented well, but was used for several applications (e.g., Oke et al. 2018; Griffin et al. 2017; Branson & Sun 2017)
  • BRAN2016: using the same configuration as BRAN2016, BRAN2016 was a long reanalysis that was again not well documented, but used for several applications (e.g., Oke et al. 2019; Benthuysen et al. 2018; Huang et al. 2019). BRAN2016 developed a deep bias in temperature and salinity that hadn’t been detected before.
  • BRAN2020: employed a multi-scale data assimilation step to eliminate the deep ocean biases. The misfit to observations in BRAN2020 are about 30% smaller that for BRAN2015 and BRAN2016. The data assimilation approach for BRAN2020 is documented by Chamberlain et al. (2021), and a description of the BRAN2020 experiment is presented in a data paper, by Chamberlain et al. (2021). BRAN2020 will be updated regularly.