Sea ice assimilation within ECMWF’s next generation ocean and sea ice reanalysis, and beyond�

Phil Browne, Hao Zuo, Sarah Keeley

ORAS5 sea ice concentration Prototype ORAS6 sea ice concentration

Relevant ingredients of ORAS6

LIM2 – operational single category model

SI3 – new multi category model

Prognostic variables of LIM2 and SI3

So we go from 1.5 variables to around 50!

How do we make this a tractable problem?

• Our observations are only of sea ice concentration (grid box averages)
• We want to keep quantities that are orthogonal to concentration constant

​

​

​

​

​

​

​

​

​

​

• We change volumetric quantities in proportion to concentration changes

​

concentration

thickness

thickness

new concentration

Single category to multicategory and treatment of prognostic variables

• NEMOVAR gives increment of total ice concentration in a grid box
• We have a degree of freedom to choose how to distribute this increment across thickness categories
• Without better information, we trust the model
• Keeps state close to balance/model attractor
• Should minimize changes to thickness introduced by concentration observations
• Need to make some extra assumptions when the model has no ice but observations and therefore DA says otherwise
• e.g. temperature, salinity, thickness distribution of new ice from DA increment

Can the model maintain increments? – Ice Induced Temperature Increments

Ice advection/assimilation increment interaction

• Ice advection scheme (Prather, 1986) conserves first and second order gradients of tracers
• Applying an assimilation increment with IAU is, by definition, modifying 0^th order moment of a tracer
• Conserving the higher order moment while modifying the underlying field leads to nasty grid-scale noise/checkerboarding patterns (top image)
• We modify the advection scheme to recompute gradients instead of conserving initial values (bottom image)

Future developments for multicategory control vector

The challenge for variational DA: specifying the cross-category terms in the background error covariance matrix

​

Can be implemented with either a diffusion operator or balance transform

​

The values and structures we hope to learn from EnKF experience

This gives us sensitivity to new observation types!

Nonlinear SIT obs operator

Jacobian of SIT

If we have only a single category (i.e. L = 1) then H(x) ≡ 0, i.e. SIT obs will have no impact!

More thin ice

Less thick ice

Overall reduction in ice thickness to fit obs by:

Future developments for NWP

Sea ice concentration retrievals computed with combination of 4DVar and ML

In the coupled framework we will rely less on external products and more on self-consistent analyses.

​

All-sky and all-surface approaches are therefore vital to use observations appropriately

Abstract for 11th International Workshop on Sea Ice Modelling, Assimilation, Observations, Predictions and Verification

Sea ice assimilation within ECMWF’s next generation ocean and sea ice reanalysis, and beyond

​

The next operational sea ice model to be used at ECMWF for reanalysis and NWP required major data assimilation (DA) developments. The fundamental difference between the old sea ice model (LIM2) and the next model (SI3) is the change from a single category model to one of multi-categories.

In this talk we will present the work we have done to allow variational DA with the multi-category model and describe the many scientific and pragmatic choices we have made along the way.

Along with the challenges of the new multi-category model come opportunities. We will touch on future planned developments to fully support multi-category sea ice within the DA system, which amongst other things will allow to assimilate thickness (equiv. freeboard/altimeter profile) observations without needing to further extend the variational control vector.