Advanced data-assimilation methods for satellite observations
Data-Assimilation Research Centre DARCUniversity of Reading
ESA Advanced Data Assimilation project July 2012
Overview
• Task 1: Data-assimilation methods for nonlinear non-Gaussian and multi-scale problems
• Task 2: Quantifying and representing uncertainty in models and observations at multiple scales
• Task 3: Exploration of advanced data-assimilation schemes to retrieve new snow products
The Equivalent-Weights Particle Filter
• Use simple proposal at each time step, e.g. relaxation to observations.
• Use different proposal at final time step to ensure that weights are very similar.
t=0 t=50 t=100
y y
Balance preservation
• Geophysical flows exhibit certain relations between variables called balance relation
• Examples are geostrophic balance and hydrostatic balance• It is crucial that the data-assimilation method retains these
balances to a large extend to avoid strongly unbalanced states, like strong gravity waves
• This is studied here in an ocean model
Quality of the ensemble: ensemble mean
Quality of the ensemble: rank histogram
How a-geostrophic is the flow?
Energy spectra
Unforced stochastic model After Particle filter update
Energy in unbalanced modes
Conclusions
• Equivalent-weights particle filter performs well for the ocean model.
• The scheme does not introduce gravity wave energy beyond what the stochastic forcing does.
• Gravity-wave energy varies substantially over the particles, suggesting that underlying state and random effects are important
• Present work: sensitivity to observation strategy (WP 1.2)