Strides forward in global reanalysis

February 17th, 2022

Two new papers led by CSIRO with Bureau of Meteorology co-authors mark significant advances in Bluelink’s global ocean modelling science. The BRAN2020 (the Bluelink ReANalysis) dataset was released last year and has attracted global interest. The papers describe a novel data assimilation approach developed under Bluelink, and quantify the improvements this approach yields in the latest version of BRAN. The new data assimilation approach has also been recently adopted into the Ocean Model, Analysis and Prediction System (OceanMAPS), Australia’s operational short-range ocean forecast system.

Multi-scale data assimilation

Chamberlain et. al. (2021a) describe the innovative data assimilation scheme used in BRAN2020. It employs a novel technique for its ensemble optimal interpolation (EnOI) method: a two-step data assimilation over multiple length scales. The first step is done at coarse resolution to correct the model state at the broad-scale, and is shown to reduce bias in the subsurface. The second step is at a higher-resolution to correct at the eddy-scale, and is shown to reduce errors in mesoscale features. The paper demonstrates clear improvements in the representation of temperature and salinity at all depths around Australia, especially within or below the thermocline.

BRAN2020 skill analysis

In the second paper, Chamberlain et. al. (2021b) describe the BRAN2020 system as a whole and examine the model skill globally. BRAN2020 results show increased skill and reduced bias. For example, misfits between the reanalysis and observations are reduced by more than 30% for subsurface temperature and salinity. The paper also describes BRAN2020’s realistic representation of all major current systems in the near-global domain, including western boundary currents, equatorial circulation, Southern Ocean dynamics, and mesoscale eddies.

“The latest BRAN experiment is our best yet” – Peter Oke, CSIRO

An example figure below from Chamberlain et. al. (2021b) compares sea level variability from BRAN2020 to results from a model without data assimilation, and illustrates how BRAN2020 results are much more closely aligned with the observations.

figure from the paper

Figure from Chamberlain et. al. (2021b). Sea level variability over 1993-2014 (detrended) from (a) BRAN2020, (b) a model without data assimilation, and (c) observations.


For BRAN2020 data, see our data access page. The dataset continues to be updated; at the time of this article, output is available to the end of 2021.

For more Bluelink papers, see our publication list.