ensemble data assimilation and uncertainty quantification
DESCRIPTION
Ensemble Data Assimilation and Uncertainty Quantification. Jeffrey Anderson, Alicia Karspeck , Tim Hoar, Nancy Collins, Kevin Raeder, Steve Yeager National Center for Atmospheric Research. What is Data Assimilation?. Observations. Observations combined with a Model forecast…. - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/1.jpg)
Ensemble Data Assimilation and Uncertainty Quantification
Jeffrey Anderson, Alicia Karspeck, Tim Hoar, Nancy Collins, Kevin Raeder, Steve Yeager
National Center for Atmospheric Research
Ocean Sciences Meeting 24 Feb. 2012 1
![Page 2: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/2.jpg)
2
Observations
…to produce an analysis(best possible estimate).
What is Data Assimilation?
+
Observations combined with a Model forecast…
Ocean Sciences Meeting 24 Feb. 2012
![Page 3: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/3.jpg)
3
What is Ensemble Data Assimilation?Use an ensemble (set) of model forecasts.
Use sample statistics to get covariance between state and observations.
Often assume that ensemble members are random draw.
Ensemble methods I use are optimal solution when:1. Model is linear,2. Observation errors are unbiased gaussian,3. Relation between model and obs is linear,4. Ensemble is large enough.
Ocean Sciences Meeting 24 Feb. 2012
![Page 4: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/4.jpg)
Atmospheric Ensemble Reanalysis, 1998-2010Assimilation uses 80
members of 2o FV CAM forced by a single ocean
(Hadley+ NCEP-OI2) and produces a very
competitive reanalysis.
O(1 million) atmospheric obs are assimilated every day.
500 hPa GPHFeb 17 2003
Ocean Sciences Meeting 24 Feb. 2012 4
![Page 5: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/5.jpg)
Ensemble Filter for Large Geophysical Models
Ensemble state estimate after using previous observation (analysis)
Ensemble state at time of next observation (prior)
1. Use model to advance ensemble (3 members here) to time at which next observation becomes available.
Ocean Sciences Meeting 24 Feb. 2012 5
![Page 6: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/6.jpg)
Theory: observations from instruments with uncorrelated errors can be done sequentially.
Ensemble Filter for Large Geophysical Models
Ocean Sciences Meeting 24 Feb. 2012
2. Get prior ensemble sample of observation, y = h(x), by applying forward operator h to each ensemble member.
6
![Page 7: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/7.jpg)
Ensemble Filter for Large Geophysical Models
Ocean Sciences Meeting 24 Feb. 2012
3. Get observed value and observational error distribution from observing system.
7
![Page 8: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/8.jpg)
Note: Difference between various ensemble filters is primarily in observation increment calculation.
Ensemble Filter for Large Geophysical Models
Ocean Sciences Meeting 24 Feb. 2012
4. Find the increments for the prior observation ensemble (this is a scalar problem for uncorrelated observation errors).
8
![Page 9: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/9.jpg)
Theory: impact of observation increments on each state variable can be handled independently!
Ensemble Filter for Large Geophysical Models
Ocean Sciences Meeting 24 Feb. 2012
5. Use ensemble samples of y and each state variable to linearly regress observation increments onto state variable increments.
9
![Page 10: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/10.jpg)
Ensemble Filter for Large Geophysical Models
Ocean Sciences Meeting 24 Feb. 2012
6. When all ensemble members for each state variable are updated, there is a new analysis. Integrate to time of next observation …
10
![Page 11: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/11.jpg)
Ensemble Filter for Lorenz-96 40-Variable Model
Ocean Sciences Meeting 24 Feb. 2012 11
40 state variables: X1, X2,..., X40.dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.Acts ‘something’ like synoptic weather around a latitude band.
![Page 12: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/12.jpg)
Ensemble Filter for Lorenz-96 40-Variable Model
Ocean Sciences Meeting 24 Feb. 2012 12
40 state variables: X1, X2,..., X40.dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.Acts ‘something’ like synoptic weather around a latitude band.
![Page 13: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/13.jpg)
Ensemble Filter for Lorenz-96 40-Variable Model
Ocean Sciences Meeting 24 Feb. 2012 13
40 state variables: X1, X2,..., X40.dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.Acts ‘something’ like synoptic weather around a latitude band.
![Page 14: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/14.jpg)
Ensemble Filter for Lorenz-96 40-Variable Model
Ocean Sciences Meeting 24 Feb. 2012 14
40 state variables: X1, X2,..., X40.dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.Acts ‘something’ like synoptic weather around a latitude band.
![Page 15: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/15.jpg)
Ensemble Filter for Lorenz-96 40-Variable Model
Ocean Sciences Meeting 24 Feb. 2012 15
40 state variables: X1, X2,..., X40.dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.Acts ‘something’ like synoptic weather around a latitude band.
![Page 16: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/16.jpg)
Ensemble Filter for Lorenz-96 40-Variable Model
Ocean Sciences Meeting 24 Feb. 2012 16
40 state variables: X1, X2,..., X40.dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.Acts ‘something’ like synoptic weather around a latitude band.
![Page 17: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/17.jpg)
Ensemble Filter for Lorenz-96 40-Variable Model
Ocean Sciences Meeting 24 Feb. 2012 17
40 state variables: X1, X2,..., X40.dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.Acts ‘something’ like synoptic weather around a latitude band.
![Page 18: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/18.jpg)
Ensemble Filter for Lorenz-96 40-Variable Model
Ocean Sciences Meeting 24 Feb. 2012 18
40 state variables: X1, X2,..., X40.dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.Acts ‘something’ like synoptic weather around a latitude band.
![Page 19: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/19.jpg)
Ensemble Filter for Lorenz-96 40-Variable Model
Ocean Sciences Meeting 24 Feb. 2012 19
40 state variables: X1, X2,..., X40.dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.Acts ‘something’ like synoptic weather around a latitude band.
![Page 20: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/20.jpg)
Ensemble Filter for Lorenz-96 40-Variable Model
Ocean Sciences Meeting 24 Feb. 2012 20
40 state variables: X1, X2,..., X40.dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.Acts ‘something’ like synoptic weather around a latitude band.
![Page 21: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/21.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 21
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 22: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/22.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 22
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 23: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/23.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 23
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 24: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/24.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 24
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 25: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/25.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 25
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 26: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/26.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 26
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 27: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/27.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 27
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 28: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/28.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 28
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 29: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/29.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 29
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 30: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/30.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 30
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 31: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/31.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 31
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 32: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/32.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 32
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 33: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/33.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 33
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 34: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/34.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 34
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 35: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/35.jpg)
Lorenz-96 is sensitive to small perturbations
Ocean Sciences Meeting 24 Feb. 2012 35
Introduce 20 ‘ensemble’ state estimates.Each is slightly perturbed for each of the 40-variables at time 100.Refer to unperturbed control integration as ‘truth’..
![Page 36: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/36.jpg)
Assimilate ‘observations’ from 40 random locations each step.
Ocean Sciences Meeting 24 Feb. 2012 36
Observations generated by interpolating truth to station location.Simulate observational error: Add random draw from N(0, 16) to each.Start from ‘climatological’ 20-member ensemble.
![Page 37: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/37.jpg)
Assimilate ‘observations’ from 40 random locations each step.
Ocean Sciences Meeting 24 Feb. 2012 37
Observations generated by interpolating truth to station location.Simulate observational error: Add random draw from N(0, 16) to each.Start from ‘climatological’ 20-member ensemble.
![Page 38: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/38.jpg)
Assimilate ‘observations’ from 40 random locations each step.
Ocean Sciences Meeting 24 Feb. 2012 38
Observations generated by interpolating truth to station location.Simulate observational error: Add random draw from N(0, 16) to each.Start from ‘climatological’ 20-member ensemble.
![Page 39: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/39.jpg)
Assimilate ‘observations’ from 40 random locations each step.
Ocean Sciences Meeting 24 Feb. 2012 39
Observations generated by interpolating truth to station location.Simulate observational error: Add random draw from N(0, 16) to each.Start from ‘climatological’ 20-member ensemble.
![Page 40: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/40.jpg)
Assimilate ‘observations’ from 40 random locations each step.
Ocean Sciences Meeting 24 Feb. 2012 40
Observations generated by interpolating truth to station location.Simulate observational error: Add random draw from N(0, 16) to each.Start from ‘climatological’ 20-member ensemble.
![Page 41: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/41.jpg)
Assimilate ‘observations’ from 40 random locations each step.
Ocean Sciences Meeting 24 Feb. 2012 41
Observations generated by interpolating truth to station location.Simulate observational error: Add random draw from N(0, 16) to each.Start from ‘climatological’ 20-member ensemble.
![Page 42: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/42.jpg)
Assimilate ‘observations’ from 40 random locations each step.
Ocean Sciences Meeting 24 Feb. 2012 42
Observations generated by interpolating truth to station location.Simulate observational error: Add random draw from N(0, 16) to each.Start from ‘climatological’ 20-member ensemble.
![Page 43: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/43.jpg)
Assimilate ‘observations’ from 40 random locations each step.
Ocean Sciences Meeting 24 Feb. 2012 43
Observations generated by interpolating truth to station location.Simulate observational error: Add random draw from N(0, 16) to each.Start from ‘climatological’ 20-member ensemble.
![Page 44: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/44.jpg)
Assimilate ‘observations’ from 40 random locations each step.
Ocean Sciences Meeting 24 Feb. 2012 44
Observations generated by interpolating truth to station location.Simulate observational error: Add random draw from N(0, 16) to each.Start from ‘climatological’ 20-member ensemble.
![Page 45: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/45.jpg)
Assimilate ‘observations’ from 40 random locations each step.
Ocean Sciences Meeting 24 Feb. 2012 45
This isn’t working very well.Ensemble spread is reduced, but...,Ensemble is inconsistent with truth most places.
Confident and wrong. Confident and right.
![Page 46: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/46.jpg)
Some Error Sources in Ensemble Filters
Ocean Sciences Meeting 24 Feb. 2012
1. Model error
2. Obs. operator error;Representativeness
3. Observation error4. Sampling Error;Gaussian Assumption
5. Sampling Error;Assuming LinearStatistical Relation
46
![Page 47: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/47.jpg)
Observations impact unrelated state variables through sampling error.
Ocean Sciences Meeting 24 Feb. 2012
Plot shows expected absolute value of sample correlation vs. true correlation.
Unrelated obs. reduce spread, increase error.
Attack with localization.
Don’t let obs. Impact unrelated state.
47
![Page 48: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/48.jpg)
Lorenz-96 Assimilation with localization of observation impact.
Ocean Sciences Meeting 24 Feb. 2012 48
![Page 49: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/49.jpg)
Lorenz-96 Assimilation with localization of observation impact.
Ocean Sciences Meeting 24 Feb. 2012 49
![Page 50: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/50.jpg)
Lorenz-96 Assimilation with localization of observation impact.
Ocean Sciences Meeting 24 Feb. 2012 50
![Page 51: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/51.jpg)
Lorenz-96 Assimilation with localization of observation impact.
Ocean Sciences Meeting 24 Feb. 2012 51
![Page 52: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/52.jpg)
Lorenz-96 Assimilation with localization of observation impact.
Ocean Sciences Meeting 24 Feb. 2012 52
![Page 53: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/53.jpg)
Lorenz-96 Assimilation with localization of observation impact.
Ocean Sciences Meeting 24 Feb. 2012 53
![Page 54: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/54.jpg)
Lorenz-96 Assimilation with localization of observation impact.
Ocean Sciences Meeting 24 Feb. 2012 54
![Page 55: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/55.jpg)
Lorenz-96 Assimilation with localization of observation impact.
Ocean Sciences Meeting 24 Feb. 2012 55
![Page 56: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/56.jpg)
Lorenz-96 Assimilation with localization of observation impact.
Ocean Sciences Meeting 24 Feb. 2012 56
![Page 57: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/57.jpg)
Lorenz-96 Assimilation with localization of observation impact.
Ocean Sciences Meeting 24 Feb. 2012 57
This ensemble is more consistent with truth.
![Page 58: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/58.jpg)
Localization computed by empirical offline computation.
Ocean Sciences Meeting 24 Feb. 2012 58
![Page 59: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/59.jpg)
Some Error Sources in Ensemble Filters
Ocean Sciences Meeting 24 Feb. 2012
1. Model error
2. Obs. operator error;Representativeness
3. Observation error4. Sampling Error;Gaussian Assumption
5. Sampling Error;Assuming LinearStatistical Relation
59
![Page 60: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/60.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 60
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
Time evolution for first state variable shown.Assimilating model quickly diverges from ‘true’ model.
![Page 61: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/61.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 61
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
![Page 62: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/62.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 62
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
![Page 63: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/63.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 63
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
![Page 64: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/64.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 64
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
![Page 65: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/65.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 65
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
![Page 66: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/66.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 66
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
![Page 67: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/67.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 67
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
![Page 68: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/68.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 68
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
![Page 69: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/69.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 69
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
![Page 70: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/70.jpg)
Assimilating in the presence of simulated model error.
Ocean Sciences Meeting 24 Feb. 2012 70
dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F.For truth, use F = 8.In assimilating model, use F = 6.
This isn’t working again!It will just keep getting worse.
![Page 71: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/71.jpg)
Reduce confidence in model prior to deal with model error
Ocean Sciences Meeting 24 Feb. 2012 71
Use inflation.Simply increase prior ensemble variance for each state variable.Adaptive algorithms use observations to guide this.
![Page 72: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/72.jpg)
Assimilating with Inflation in presence of model error.
Ocean Sciences Meeting 24 Feb. 2012 72
Inflation is a function of state variable and time.Automatically selected by adaptive inflation algorithm.
![Page 73: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/73.jpg)
Assimilating with Inflation in presence of model error.
Ocean Sciences Meeting 24 Feb. 2012 73
Inflation is a function of state variable and time.Automatically selected by adaptive inflation algorithm.
![Page 74: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/74.jpg)
Assimilating with Inflation in presence of model error.
Ocean Sciences Meeting 24 Feb. 2012 74
Inflation is a function of state variable and time.Automatically selected by adaptive inflation algorithm.
![Page 75: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/75.jpg)
Assimilating with Inflation in presence of model error.
Ocean Sciences Meeting 24 Feb. 2012 75
Inflation is a function of state variable and time.Automatically selected by adaptive inflation algorithm.
![Page 76: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/76.jpg)
Assimilating with Inflation in presence of model error.
Ocean Sciences Meeting 24 Feb. 2012 76
Inflation is a function of state variable and time.Automatically selected by adaptive inflation algorithm.
![Page 77: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/77.jpg)
Assimilating with Inflation in presence of model error.
Ocean Sciences Meeting 24 Feb. 2012 77
Inflation is a function of state variable and time.Automatically selected by adaptive inflation algorithm.
![Page 78: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/78.jpg)
Assimilating with Inflation in presence of model error.
Ocean Sciences Meeting 24 Feb. 2012 78
Inflation is a function of state variable and time.Automatically selected by adaptive inflation algorithm.
![Page 79: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/79.jpg)
Assimilating with Inflation in presence of model error.
Ocean Sciences Meeting 24 Feb. 2012 79
Inflation is a function of state variable and time.Automatically selected by adaptive inflation algorithm.
![Page 80: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/80.jpg)
Assimilating with Inflation in presence of model error.
Ocean Sciences Meeting 24 Feb. 2012 80
Inflation is a function of state variable and time.Automatically selected by adaptive inflation algorithm.
![Page 81: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/81.jpg)
Assimilating with Inflation in presence of model error.
Ocean Sciences Meeting 24 Feb. 2012 81
Inflation is a function of state variable and time.Automatically selected by adaptive inflation algorithm.It can work, even in presence of severe model error.
![Page 82: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/82.jpg)
82Ocean Sciences Meeting 24 Feb. 2012
(Ensemble) KF optimal for linear model, gaussian likelihood, perfect model.
In KF, only mean and covariance have meaning.
Ensemble allows computation of many other statistics. What do they mean? Not entirely clear.
What do they mean when there are all sorts of error? Even less clear.
Must Calibrate and Validate results.
Uncertainty Quantification from an Ensemble Kalman Filter
![Page 83: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/83.jpg)
Ocean Sciences Meeting 24 Feb. 2012 83
Obs DART
POPCoupler
2D forcing3D restart
3D state2D forcingfrom CAM assimilation
DATM
Current Ocean (POP) Assimilation
![Page 84: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/84.jpg)
FLOAT_SALINITY 68200 FLOAT_TEMPERATURE 395032 DRIFTER_TEMPERATURE 33963 MOORING_SALINITY 27476 MOORING_TEMPERATURE 623967 BOTTLE_SALINITY 79855 BOTTLE_TEMPERATURE 81488 CTD_SALINITY 328812 CTD_TEMPERATURE 368715 STD_SALINITY 674 STD_TEMPERATURE 677 XCTD_SALINITY 3328 XCTD_TEMPERATURE 5790 MBT_TEMPERATURE 58206 XBT_TEMPERATURE 1093330 APB_TEMPERATURE 580111
World Ocean Database T,S observation countsThese counts are for 1998 & 1999 and are representative.
• temperature observation error standard deviation == 0.5 K.• salinity observation error standard deviation == 0.5 msu.
Ocean Sciences Meeting 24 Feb. 2012 84
![Page 85: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/85.jpg)
Ocean Sciences Meeting 24 Feb. 2012
Coup
led
Free
Run
POP
forc
ed b
y ob
serv
ed
atm
osph
ere
(hin
dcas
t)
Ensemble Assimilation48 POP oceans
Forced by 48 CAM reanalyses
Physical Space: 1998/1999 SST Anomaly from HadOI-SST
![Page 86: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/86.jpg)
Example: cross section along Kuroshio; model separates too far north.
86
Challenges where ocean model is unable, or unwilling, to simulate reality.
Ocean Sciences Meeting 24 Feb. 2012
Regarded to be accurate.
Free run of POP, the warm water is too far North.
![Page 87: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/87.jpg)
87
time = 10days time = 20days time = 30days
time = 60daystime = 50daystime = 40days
Initially warm water goes too far north.
Many observations are rejected (red), but others (blue) move temperature gradient south.
Adaptive inflation increases ensemble spread as assimilation struggles to push model towards obs.
Latitude Latitude Latitude
Tem
pera
ture
Tem
pera
ture
Tem
pera
ture
Challenges in correcting position of Kuroshio.
60-day assimilation starting from model climatology on 1 January 2004.
Ocean Sciences Meeting 24 Feb. 2012
![Page 88: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/88.jpg)
88
time = 10days time = 20days time = 30days
Observations keep pulling the warm water to the south; Model forecasts continue to quickly move warm water further north. Inflation continues to increase spread.
Model forecasts finally fail due to numerical issues. Black dashes show original model state from 10 January.
Challenges in correcting position of Kuroshio.
60-day assimilation starting from model climatology on 1 January 2004.
time = 60daystime = 50daystime = 40daysLatitude Latitude Latitude
Tem
pera
ture
Tem
pera
ture
Tem
pera
ture
Ocean Sciences Meeting 24 Feb. 2012
Green dashed line is posterior at previous time,Blue dashed line is prior at current time,Ensembles are thin lines.
![Page 89: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/89.jpg)
89
time = 10days time = 20days time = 30days
time = 60daystime = 50daystime = 40days
60-day assimilation starting from model climatology on 1 January 2004.
• Assimilation cannot force model to fit observations.• Use of adaptive inflation leads to eventual model failure.• Reduced adaptive inflation can lead to compromise between observations
and model.
Latitude Latitude Latitude
Tem
pera
ture
Tem
pera
ture
Tem
pera
ture
Challenges in correcting position of Kuroshio.
Ocean Sciences Meeting 24 Feb. 2012
![Page 90: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/90.jpg)
90Ocean Sciences Meeting 24 Feb. 2012
Very certain that model predictions are different from observations. Very certain that small correlations have large errors. Moderately confident that large correlations are ‘realistic’. Very uncertain about state in sparse/unobserved regions.
Must Calibrate and Validate results (adaptive inflation/localization).
There may not be enough observations to do this many places.
First order of business: Improving models. DA can help with this.
Ensemble DA and UQ for Global Ocean Models
![Page 91: Ensemble Data Assimilation and Uncertainty Quantification](https://reader036.vdocuments.us/reader036/viewer/2022062410/56815ff6550346895dcef586/html5/thumbnails/91.jpg)
Ocean Sciences Meeting 24 Feb. 2012
http://www.image.ucar.edu/DAReS/DART/
Code to implement all of the algorithms discussed are freely available from:
91