Ensemble Data Assimilation and Uncertainty Quantification Jeffrey Anderson, Alicia Karspeck, Tim Hoar, Nancy Collins, Kevin Raeder, Steve Yeager National.

Slides:



Advertisements
Similar presentations
22 March 2011: GSICS GRWG & GDWG Meeting Daejeon, Korea Tim Hewison SEVIRI-IASI Inter-calibration Uncertainty Evaluation.
Advertisements

1 -Classification: Internal Uncertainty in petroleum reservoirs.
Variational data assimilation and forecast error statistics
OSE meeting GODAE, Toulouse 4-5 June 2009 Interest of assimilating future Sea Surface Salinity measurements.
Polly Smith, Alison Fowler, Amos Lawless School of Mathematical and Physical Sciences, University of Reading Exploring coupled data assimilation using.
Kriging.
Motivation The Carolinas has had a tremendous residential and commercial investment in coastal areas during the past 10 years. However rapid development.
Initialization Issues of Coupled Ocean-atmosphere Prediction System Climate and Environment System Research Center Seoul National University, Korea In-Sik.
Effects of model error on ensemble forecast using the EnKF Hiroshi Koyama 1 and Masahiro Watanabe 2 1 : Center for Climate System Research, University.
Jon Robson (Uni. Reading) Rowan Sutton (Uni. Reading) and Doug Smith (UK Met Office) Analysis of a decadal prediction system:
Empirical Localization of Observation Impact in Ensemble Filters Jeff Anderson IMAGe/DAReS Thanks to Lili Lei, Tim Hoar, Kevin Raeder, Nancy Collins, Glen.
Ibrahim Hoteit KAUST, CSIM, May 2010 Should we be using Data Assimilation to Combine Seismic Imaging and Reservoir Modeling? Earth Sciences and Engineering.
Toward a Real Time Mesoscale Ensemble Kalman Filter Gregory J. Hakim Dept. of Atmospheric Sciences, University of Washington Collaborators: Ryan Torn (UW)
ASSIMILATION of RADAR DATA at CONVECTIVE SCALES with the EnKF: PERFECT-MODEL EXPERIMENTS USING WRF / DART Altuğ Aksoy National Center for Atmospheric Research.
Carbon Flux Bias Estimation at Regional Scale using Coupled MLEF-PCTM model Ravindra Lokupitiya Department of Atmospheric Science Colorado State University.
Advanced data assimilation methods- EKF and EnKF Hong Li and Eugenia Kalnay University of Maryland July 2006.
UNBIASED ESTIAMTION OF ANALYSIS AND FORECAST ERROR VARIANCES
A Regression Model for Ensemble Forecasts David Unger Climate Prediction Center.
Kalman filtering techniques for parameter estimation Jared Barber Department of Mathematics, University of Pittsburgh Work with Ivan Yotov and Mark Tronzo.
ESA DA Projects Progress Meeting 2University of Reading Advanced Data Assimilation Methods WP2.1 Perform (ensemble) experiments to quantify model errors.
EnKF Overview and Theory
Observing Strategy and Observation Targeting for Tropical Cyclones Using Ensemble-Based Sensitivity Analysis and Data Assimilation Chen, Deng-Shun 3 Dec,
Open Source Ensemble Kalman Filtering: the Data Assimilation Research Testbed - DART Tim Hoar, Jeffrey Anderson, Nancy Collins, Kevin Raeder, Hui Liu,
The Importance of Atmospheric Variability for Data Requirements, Data Assimilation, Forecast Errors, OSSEs and Verification Rod Frehlich and Robert Sharman.
Nonlinear Data Assimilation and Particle Filters
Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss High-resolution data assimilation in COSMO: Status and.
Tim Hoar : National Center for Atmospheric Research with a whole lot of help from: Jeff Anderson, Nancy Collins, Kevin Raeder, Bill Sacks : NCAR Yongfei.
Dusanka Zupanski And Scott Denning Colorado State University Fort Collins, CO CMDL Workshop on Modeling and Data Analysis of Atmospheric CO.
A unifying framework for hybrid data-assimilation schemes Peter Jan van Leeuwen Data Assimilation Research Center (DARC) National Centre for Earth Observation.
Sophie RICCI CALTECH/JPL Post-doc Advisor : Ichiro Fukumori The diabatic errors in the formulation of the data assimilation Kalman Filter/Smoother system.
Applications of optimal control and EnKF to Flow Simulation and Modeling Florida State University, February, 2005, Tallahassee, Florida The Maximum.
Assimilating chemical compound with a regional chemical model Chu-Chun Chang 1, Shu-Chih Yang 1, Mao-Chang Liang 2, ShuWei Hsu 1, Yu-Heng Tseng 3 and Ji-Sung.
Sensitivity Analysis of Mesoscale Forecasts from Large Ensembles of Randomly and Non-Randomly Perturbed Model Runs William Martin November 10, 2005.
Development of an EnKF to estimate CO 2 fluxes from realistic distributions of X CO2 Liang Feng, Paul Palmer
Pg 1 A Loosely Coupled Ocean-Atmosphere Ensemble Assimilation System. Tim Hoar, Nancy Collins, Kevin Raeder, Jeffrey Anderson, NCAR Institute for Math.
Data assimilation and forecasting the weather (!) Eugenia Kalnay and many friends University of Maryland.
The European Heat Wave of 2003: A Modeling Study Using the NSIPP-1 AGCM. Global Modeling and Assimilation Office, NASA/GSFC Philip Pegion (1), Siegfried.
© Crown copyright Met Office The EN4 dataset of quality controlled ocean temperature and salinity profiles and monthly objective analyses Simon Good.
Local Predictability of the Performance of an Ensemble Forecast System Liz Satterfield and Istvan Szunyogh Texas A&M University, College Station, TX Third.
Tim Hoar: National Center for Atmospheric Research with a whole lot of help from: Bill Sacks, Mariana Vertenstein, Tony Craig, Jim Edwards: NCAR Andrew.
One-year re-forecast ensembles with CCSM3.0 using initial states for 1 January and 1 July in Model: CCSM3 is a coupled climate model with state-of-the-art.
A Random Subgrouping Scheme for Ensemble Kalman Filters Yun Liu Dept. of Atmospheric and Oceanic Science, University of Maryland Atmospheric and oceanic.
Pg 1 SAMSI UQ Workshop; Sep Ensemble Data Assimilation and Uncertainty Quantification Jeff Anderson National Center for Atmospheric Research.
Future Directions in Ensemble DA for Hurricane Prediction Applications Jeff Anderson: NCAR Ryan Torn: SUNY Albany Thanks to Chris Snyder, Pavel Sakov The.
The application of ensemble Kalman filter in adaptive observation and information content estimation studies Junjie Liu and Eugenia Kalnay July 13th, 2007.
ENSEMBLE KALMAN FILTER IN THE PRESENCE OF MODEL ERRORS Hong Li Eugenia Kalnay.
École Doctorale des Sciences de l'Environnement d’ Î le-de-France Année Modélisation Numérique de l’Écoulement Atmosphérique et Assimilation.
École Doctorale des Sciences de l'Environnement d’Île-de-France Année Universitaire Modélisation Numérique de l’Écoulement Atmosphérique et Assimilation.
École Doctorale des Sciences de l'Environnement d’ Î le-de-France Année Modélisation Numérique de l’Écoulement Atmosphérique et Assimilation.
Demonstration and Comparison of Sequential Approaches for Altimeter Data Assimilation in HYCOM A. Srinivasan, E. P. Chassignet, O. M. Smedstad, C. Thacker,
X 10 km Model Bathymetry (A)Temperature (B) Alongshore Velocity Cross-shore Velocity Figure 1: Panel (A) model grid and bathymetry. The model has closed.
The Ensemble Kalman filter
June 20, 2005Workshop on Chemical data assimilation and data needs Data Assimilation Methods Experience from operational meteorological assimilation John.
National Oceanic and Atmospheric Administration’s National Weather Service Colorado Basin River Forecast Center Salt Lake City, Utah 11 The Hydrologic.
Pg 1 ESPC DA Workshop; Sept The Data Assimilation Research Testbed: A Community Ensemble DA Facility Jeffrey Anderson, Nancy Collins, Tim Hoar, Hui.
ECMWF/EUMETSAT NWP-SAF Satellite data assimilation Training Course Mar 2016.
Jeffrey Anderson, NCAR Data Assimilation Research Section
Data Assimilation Research Testbed Tutorial
Jeffrey Anderson, NCAR Data Assimilation Research Section
Data Assimilation Theory CTCD Data Assimilation Workshop Nov 2005
Data Assimilation Training
Tom Hopson, Jason Knievel, Yubao Liu, Gregory Roux, Wanli Wu
Jeff Anderson, NCAR Data Assimilation Research Section
Parallel Implementations of Ensemble Kalman Filters for Huge Geophysical Models Jeffrey Anderson, Helen Kershaw, Jonathan Hendricks, Nancy Collins, Ye.
Jeff Anderson, NCAR Data Assimilation Research Section
Error statistics in data assimilation
Data Assimilation Initiative, NCAR
Hartmut Bösch and Sarah Dance
Sarah Dance DARC/University of Reading
Presentation transcript:

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

Observations …to produce an analysis (best possible estimate). What is Data Assimilation? + Observations combined with a Model forecast… 2Ocean Sciences Meeting 24 Feb. 2012

What is Ensemble Data Assimilation? 3 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

Atmospheric Ensemble Reanalysis, Assimilation uses 80 members of 2 o 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 GPH Feb Ocean Sciences Meeting 24 Feb

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

Theory: observations from instruments with uncorrelated errors can be done sequentially. Ensemble Filter for Large Geophysical Models Ocean Sciences Meeting 24 Feb Get prior ensemble sample of observation, y = h(x), by applying forward operator h to each ensemble member. 6

Ensemble Filter for Large Geophysical Models Ocean Sciences Meeting 24 Feb Get observed value and observational error distribution from observing system. 7

Note: Difference between various ensemble filters is primarily in observation increment calculation. Ensemble Filter for Large Geophysical Models Ocean Sciences Meeting 24 Feb Find the increments for the prior observation ensemble (this is a scalar problem for uncorrelated observation errors). 8

Theory: impact of observation increments on each state variable can be handled independently! Ensemble Filter for Large Geophysical Models Ocean Sciences Meeting 24 Feb Use ensemble samples of y and each state variable to linearly regress observation increments onto state variable increments. 9

Ensemble Filter for Large Geophysical Models Ocean Sciences Meeting 24 Feb When all ensemble members for each state variable are updated, there is a new analysis. Integrate to time of next observation … 10

Ensemble Filter for Lorenz Variable Model Ocean Sciences Meeting 24 Feb state variables: X1, X2,..., X40. dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. Acts ‘something’ like synoptic weather around a latitude band.

Ensemble Filter for Lorenz Variable Model Ocean Sciences Meeting 24 Feb state variables: X1, X2,..., X40. dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. Acts ‘something’ like synoptic weather around a latitude band.

Ensemble Filter for Lorenz Variable Model Ocean Sciences Meeting 24 Feb state variables: X1, X2,..., X40. dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. Acts ‘something’ like synoptic weather around a latitude band.

Ensemble Filter for Lorenz Variable Model Ocean Sciences Meeting 24 Feb state variables: X1, X2,..., X40. dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. Acts ‘something’ like synoptic weather around a latitude band.

Ensemble Filter for Lorenz Variable Model Ocean Sciences Meeting 24 Feb state variables: X1, X2,..., X40. dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. Acts ‘something’ like synoptic weather around a latitude band.

Ensemble Filter for Lorenz Variable Model Ocean Sciences Meeting 24 Feb state variables: X1, X2,..., X40. dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. Acts ‘something’ like synoptic weather around a latitude band.

Ensemble Filter for Lorenz Variable Model Ocean Sciences Meeting 24 Feb state variables: X1, X2,..., X40. dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. Acts ‘something’ like synoptic weather around a latitude band.

Ensemble Filter for Lorenz Variable Model Ocean Sciences Meeting 24 Feb state variables: X1, X2,..., X40. dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. Acts ‘something’ like synoptic weather around a latitude band.

Ensemble Filter for Lorenz Variable Model Ocean Sciences Meeting 24 Feb state variables: X1, X2,..., X40. dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. Acts ‘something’ like synoptic weather around a latitude band.

Ensemble Filter for Lorenz Variable Model Ocean Sciences Meeting 24 Feb state variables: X1, X2,..., X40. dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. Acts ‘something’ like synoptic weather around a latitude band.

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Lorenz-96 is sensitive to small perturbations Ocean Sciences Meeting 24 Feb 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’..

Assimilate ‘observations’ from 40 random locations each step. Ocean Sciences Meeting 24 Feb 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.

Assimilate ‘observations’ from 40 random locations each step. Ocean Sciences Meeting 24 Feb 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.

Assimilate ‘observations’ from 40 random locations each step. Ocean Sciences Meeting 24 Feb 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.

Assimilate ‘observations’ from 40 random locations each step. Ocean Sciences Meeting 24 Feb 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.

Assimilate ‘observations’ from 40 random locations each step. Ocean Sciences Meeting 24 Feb 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.

Assimilate ‘observations’ from 40 random locations each step. Ocean Sciences Meeting 24 Feb 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.

Assimilate ‘observations’ from 40 random locations each step. Ocean Sciences Meeting 24 Feb 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.

Assimilate ‘observations’ from 40 random locations each step. Ocean Sciences Meeting 24 Feb 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.

Assimilate ‘observations’ from 40 random locations each step. Ocean Sciences Meeting 24 Feb 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.

Assimilate ‘observations’ from 40 random locations each step. Ocean Sciences Meeting 24 Feb This isn’t working very well. Ensemble spread is reduced, but..., Ensemble is inconsistent with truth most places. Confident and wrong.Confident and right.

Some Error Sources in Ensemble Filters Ocean Sciences Meeting 24 Feb Model error 2. Obs. operator error; Representativeness 3. Observation error 4. Sampling Error; Gaussian Assumption 5. Sampling Error; Assuming Linear Statistical Relation 46

Observations impact unrelated state variables through sampling error. Ocean Sciences Meeting 24 Feb 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

Lorenz-96 Assimilation with localization of observation impact. Ocean Sciences Meeting 24 Feb

Lorenz-96 Assimilation with localization of observation impact. Ocean Sciences Meeting 24 Feb

Lorenz-96 Assimilation with localization of observation impact. Ocean Sciences Meeting 24 Feb

Lorenz-96 Assimilation with localization of observation impact. Ocean Sciences Meeting 24 Feb

Lorenz-96 Assimilation with localization of observation impact. Ocean Sciences Meeting 24 Feb

Lorenz-96 Assimilation with localization of observation impact. Ocean Sciences Meeting 24 Feb

Lorenz-96 Assimilation with localization of observation impact. Ocean Sciences Meeting 24 Feb

Lorenz-96 Assimilation with localization of observation impact. Ocean Sciences Meeting 24 Feb

Lorenz-96 Assimilation with localization of observation impact. Ocean Sciences Meeting 24 Feb

Lorenz-96 Assimilation with localization of observation impact. Ocean Sciences Meeting 24 Feb This ensemble is more consistent with truth.

Localization computed by empirical offline computation. Ocean Sciences Meeting 24 Feb

Some Error Sources in Ensemble Filters Ocean Sciences Meeting 24 Feb Model error 2. Obs. operator error; Representativeness 3. Observation error 4. Sampling Error; Gaussian Assumption 5. Sampling Error; Assuming Linear Statistical Relation 59

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb 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.

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. For truth, use F = 8. In assimilating model, use F = 6.

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. For truth, use F = 8. In assimilating model, use F = 6.

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. For truth, use F = 8. In assimilating model, use F = 6.

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. For truth, use F = 8. In assimilating model, use F = 6.

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. For truth, use F = 8. In assimilating model, use F = 6.

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. For truth, use F = 8. In assimilating model, use F = 6.

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. For truth, use F = 8. In assimilating model, use F = 6.

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. For truth, use F = 8. In assimilating model, use F = 6.

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb dXi / dt = (Xi+1 - Xi-2)Xi-1 - Xi + F. For truth, use F = 8. In assimilating model, use F = 6.

Assimilating in the presence of simulated model error. Ocean Sciences Meeting 24 Feb 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.

Reduce confidence in model prior to deal with model error Ocean Sciences Meeting 24 Feb Use inflation. Simply increase prior ensemble variance for each state variable. Adaptive algorithms use observations to guide this.

Assimilating with Inflation in presence of model error. Ocean Sciences Meeting 24 Feb Inflation is a function of state variable and time. Automatically selected by adaptive inflation algorithm.

Assimilating with Inflation in presence of model error. Ocean Sciences Meeting 24 Feb Inflation is a function of state variable and time. Automatically selected by adaptive inflation algorithm.

Assimilating with Inflation in presence of model error. Ocean Sciences Meeting 24 Feb Inflation is a function of state variable and time. Automatically selected by adaptive inflation algorithm.

Assimilating with Inflation in presence of model error. Ocean Sciences Meeting 24 Feb Inflation is a function of state variable and time. Automatically selected by adaptive inflation algorithm.

Assimilating with Inflation in presence of model error. Ocean Sciences Meeting 24 Feb Inflation is a function of state variable and time. Automatically selected by adaptive inflation algorithm.

Assimilating with Inflation in presence of model error. Ocean Sciences Meeting 24 Feb Inflation is a function of state variable and time. Automatically selected by adaptive inflation algorithm.

Assimilating with Inflation in presence of model error. Ocean Sciences Meeting 24 Feb Inflation is a function of state variable and time. Automatically selected by adaptive inflation algorithm.

Assimilating with Inflation in presence of model error. Ocean Sciences Meeting 24 Feb Inflation is a function of state variable and time. Automatically selected by adaptive inflation algorithm.

Assimilating with Inflation in presence of model error. Ocean Sciences Meeting 24 Feb Inflation is a function of state variable and time. Automatically selected by adaptive inflation algorithm.

Assimilating with Inflation in presence of model error. Ocean Sciences Meeting 24 Feb 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.

Ocean Sciences Meeting 24 Feb  (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 82

83 Obs DART POP Coupler 2D forcing 3D restart 3D state 2D forcing from CAM assimilation DATM Ocean Sciences Meeting 24 Feb Current Ocean (POP) Assimilation

FLOAT_SALINITY FLOAT_TEMPERATURE DRIFTER_TEMPERATURE33963 MOORING_SALINITY MOORING_TEMPERATURE BOTTLE_SALINITY BOTTLE_TEMPERATURE CTD_SALINITY CTD_TEMPERATURE STD_SALINITY 674 STD_TEMPERATURE 677 XCTD_SALINITY 3328 XCTD_TEMPERATURE 5790 MBT_TEMPERATURE XBT_TEMPERATURE APB_TEMPERATURE World Ocean Database T,S observation counts These 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

Coupled Free Run POP forced by observed atmosphere (hindcast) Ensemble Assimilation 48 POP oceans Forced by 48 CAM reanalyses Physical Space: 1998/1999 SST Anomaly from HadOI-SST 85Ocean Sciences Meeting 24 Feb. 2012

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 Regarded to be accurate. Free run of POP, the warm water is too far North.

time = 10days time = 20days time = 30days time = 60daystime = 50days time = 40days 87 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 Temperature Challenges in correcting position of Kuroshio. 60-day assimilation starting from model climatology on 1 January Ocean Sciences Meeting 24 Feb. 2012

time = 10days time = 20daystime = 30days 88 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 time = 60daystime = 50days time = 40days Latitude Temperature Ocean Sciences Meeting 24 Feb Green dashed line is posterior at previous time, Blue dashed line is prior at current time, Ensembles are thin lines.

time = 10days time = 20daystime = 30days time = 60daystime = 50days time = 40days day assimilation starting from model climatology on 1 January 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 Temperature Challenges in correcting position of Kuroshio. Ocean 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 90

Ocean Sciences Meeting 24 Feb Code to implement all of the algorithms discussed are freely available from: 91