Data assimilation with sea ice models

Transcription

1 SIPN workshop NCAR, Boulder 1-2 April 2014 Data assimilation with sea ice models François Massonnet

2

3 brine content mixed layer heat content heat content concentration thickness

4 brine content mixed layer heat content heat content concentration thickness Observation

5 brine content mixed layer heat content heat content concentration thickness Observation Whole sea ice state

6 Data assimilation consists in optimally updating the whole sea ice state, given partial observations brine content heat content thickness Updated whole state Observation mixed layer heat content concentration Whole sea ice state

7 1. Simple sea ice model 2. Comprehensive sea ice model

8 1. Simple sea ice model 2. Comprehensive sea ice model

9 2-variable sea ice model [Semtner, 1976; Notz, 2005] εσt(t) 4 Q OF (t) 1 α(t) Q SW (t) T t z Q i = k T i z T T i (t, z) h(t) T b Q o

10 The easy part: updating the observed variable Sea ice thickness [m] Observations Ensemble Analysis True thickness (not known) Free run Model years

11 The easy part: updating the observed variable Sea ice thickness [m] Observations Ensemble Analysis True thickness (not known) Free run Model years

12 The tricky part: updating the other variable Ensemble 20th April 00

13 The tricky part: updating the other variable Ensemble 20th April 00

14 Primitive nudging may bring the system into a non-physical state Ensemble Updated ensemble 20th April 00

15 Ensemble 20th April 00

16 Multivariate data assimilation accounts for model state covariance Ensemble Updated ensemble 20th April 00

17 1. Simple sea ice model The update of the whole state should be consistent with the model dynamics 2. Comprehensive sea ice model

18 1. Simple sea ice model The update of the whole state should be consistent with the model dynamics 2. Comprehensive sea ice model

19 Given a comprehensive ocean-sea ice model and observations of total ice concentration, How to update concentration in individual categories? How to update fields that hold sea ice memory? How to keep model dynamics as balanced as possible? Model: NEMO-LIM3 [Madec, 2008; Vancoppenolle et al., 2009] Observations: SMMR/SSM-I ice concentration [Eastwood et al., 2011]

20 All variables Given a comprehensive ocean-sea ice model and observations of total ice concentration, How to update concentration in individual categories? How to update fields that hold sea ice memory? How to keep model dynamics as balanced as possible? Observed variable Model: NEMO-LIM3 [Madec, 2008; Vancoppenolle et al., 2009] Observations: SMMR/SSM-I ice concentration [Eastwood et al., 2011]

21 Covariances are spaceand time-dependent! Correlation (concentration, thickness) N=25, air surface temperature + winds perturbed 26th March th September 2000 See also Lisaeter et al., 2003

22 Initialization from sea ice concentration improves seasonal Arctic predictions Error forecast September concentration Not initialized RMSE [%] Initialized in March from observations of ice concentration

23 1. Simple sea ice model The update of the whole state should be consistent with the model dynamics 2. Comprehensive sea ice model Multivariate data assimilation is promising for seasonal Artic sea ice predition

24 Take home message In data assimilation for sea ice prediction, the update of the whole sea ice state is - necessary - not obvious - worthwile

25 Thank you!

26