/
Toward seasonal to multi-annual marine biogeochemical prediction using GFDL’s Earth Toward seasonal to multi-annual marine biogeochemical prediction using GFDL’s Earth

Toward seasonal to multi-annual marine biogeochemical prediction using GFDL’s Earth - PowerPoint Presentation

briana-ranney
briana-ranney . @briana-ranney
Follow
346 views
Uploaded On 2019-12-09

Toward seasonal to multi-annual marine biogeochemical prediction using GFDL’s Earth - PPT Presentation

Toward seasonal to multiannual marine biogeochemical prediction using GFDLs Earth System Model Jongyeon Park Charles A Stock John P Dunne Xiaosong Yang Anthony Rosati Jasmin G John Shaoqing ID: 769775

prediction bgc retrospective data bgc prediction data retrospective initialization intro ocean model obs skill temp mol chl atmos assimilation

Share:

Link:

Embed:

Download Presentation from below link

Download Presentation The PPT/PDF document "Toward seasonal to multi-annual marine b..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

Toward seasonal to multi-annual marine biogeochemical prediction using GFDL’s Earth System Model Jong-yeon Park, Charles A. Stock, John P. Dunne, Xiaosong Yang, Anthony Rosati, Jasmin G. John, Shaoqing ZhangNOAA-GFDL / Princeton University (Biogeochemistry, Ecosystems, and Climate Group)

Physical climate model ( Atmosphere+ocean+land): understanding + predictionEarth system model ( Atmosphere+ocean+land + chemical + ecologibal component coupled with )Prediction. Carbon feedback study: Park et al. 2015 Seasonal –multi-year to decadal prediction with ESM?  Initialization problem Deserei Need for marine BGC prediction Lethal threats to marine ecosystem Marine ecosystem is non-stationary, shifting to a new productivity state ( Wayte 2013, Mollmann et al. 2015, Auzijonyte et al. 2016)  Dynamic ecosystem prediction is required for adaptive marine resource mgmt. From XL Catlin Seaview survey “Ocean acidification” “Harmful algae bloom” From NOAA-GLERL From World Wildlife Fund “Over fishing - poor fishery mgmt ” Ocean carbon cycle monitoring/assessment : Basis of marine food webs : Nearly 50% carbon fixation at global scale Intro – BGC Initialization data – Retrospective Prediction “Hypoxia”

ESM, tool for global BGC prediction Tommasi et al. 2017 stock bImproved fishery management through seasonal prediction of SSTPrediction from ESM can provide climate-informed marine resource management. - e.g. Tommasi et al. 2017 : Seasonal climate prediction can be used to improve marine resource management. : Higher average catch and stock biomass of sardines using future SST information. Intro – BGC Initialization data – Retrospective Prediction

Challenges in BGC prediction using ESM Model uncertainties (physics and BGC) GFDL-ESM2M-COBALT Mean surface chlorophyll s GFDL-ESM2M-COBALT SST [K/K] CHL [mg/m 3 /K] SST/Chlorophyll during ENSO SST Chl Intro – BGC Initialization data – Retrospective Prediction

Challenges in BGC prediction using ESM Model uncertainties (physics and BGC)BGC Initialization problem - Relatively sparse global-scale observations of subsurface BGC - Large number of BGC tracers ( > 30 tracers) - Large positive skewness of BGC variables - Degradation of BGC when integrating with physical data assimilation Only a few studies on BGC predictability using ESM - Seferian et al. 2014, Chikamoto et al. 2015 - BGC predictability up to few years depending on the regions Intro – BGC Initialization data – Retrospective Prediction “Limited to potential predictability / simple SST nudging

Exp 1: Control run forced with Retrospective Atmos forcingExp 2: Exp1 + relaxing surface/3D temp & salinity Exp 3: ECDA run with BGC model onRoadmap of BGC Predictability Project Provide a experimental basis for global BGC prediction system Develop BGC prediction system integrated with joint Bio- Phy assim Goal Q1. Which approaches are viable? Q2. How do viable approaches improve BGC? Limits of BGC predictability (Fernando et al. 2018) Initial assessment of BGC predictive skill Produce BGC retrospective Perform retrospective forecasts Step 1 Assess BGC predictability (Statistical and perfect model frameworks) Assess elements of existing ESM models to reproduce observed Predicting variability across scales Proof-of-concept of global BGC Prediction Q1. Which approaches are viabl e? Q2. How do viable approaches improve BGC? Optimal initialization strategy Builds on GFDL’s seasonal prediction system Seasonal to multi-annual BGC Prediction Intro – BGC Initialization data – Retrospective Prediction Assess prediction skill (model- obs ) (Optimal initialization strategy)

COBALT : The Carbon, Ocean Biogeochemistry and Lower Trophics planktonic ecosystem model33 tracers (3 phytoplankton groups, 3 zooplankton groups, free-living bacteria, organic matter, C, N, P, Si, ....)Light, temperature, nutrient limitationsCoupled with physical ocean model (MOM4, 1deg resolution) GFDL’s marine biogeochemistry model Stock et al. 2014 Intro – BGC Initialization data – Retrospective Prediction

Simulated global BGC patterns Stock et al. 2014 [mg/m 3 ] [ mol /kg (×10 -6 ) ] [ mol /kg (×10-4)] OBS MOM-COBALT CORE NO 3 (surface) O 2 (600-200m) CHL (surface) Intro – BGC Initialization data – Retrospective Prediction

NO 3_eqO2_eq Simulated equatorial subsurface BGC[mol/kg (×10-4)] [ mol /kg (×10 -6) ] MOM-COBALT CORE Intro – BGC Initialization data – Retrospective Prediction OBS

Monthly anomaly correlation (model vs. OBS) Sep1997-dec2007 Chlorophyll correlation skill Obs Model Corr = 0.78 CHL [mg/m 3 ] Intro – BGC Initialization data – Retrospective Prediction

How can we estimate Retrospective global BGC? Intro – BGC Initialization data – Retrospective Prediction Produce BGC retrospective Perform retrospective forecasts Assess prediction skill (model- obs ) (Optimal initialization strategy)

GFDL’s ECDA (Ensemble Coupled Data Assimilation)Seasonal to decadal global climate predictionGFDL-CM2.1 Physical assimilation - Atmos : u, v, temp (6 hourly NCEP2) - Ocean: Temperature and Salinity (XBT, MBT, CTD, OSD, MRB, gtspp, argo, AVHRR SST)Ensemble Kalman filter (12 ensemble members used)  ECDA + COBALT - Run period: 1991 - 2016- Assumption: improved physical field better represents BGC Data Assimilation System + BGC model fully integrated with prediction system Intro – BGC Initialization data – Retrospective Prediction

Experiments Experiments DescriptionObservation standard errorsCTRL No ocean data assimilation Atmos : 0.1 m/s (Wind), 0.1 K (Temp) DA-BGC Baseline assimilation run Atmos : 1 m/s (Wind), 1 K (Temp) Ocean: 0.5 K (Temp), 0.1 g/kg (Salinity) A series of modified DA-BGC Changing atmosphere/ocean data constraint Atmos : 1–0.1 m/s (Wind), 1–0.1 K (Temp) Ocean: 0.5–106 K (Temp), 0.1–2×106 (Salinity) Observation errors Atmos: 1 m s-1 (Wind), 1 K (Temp)Ocean: 0.5 K (Temp), 0.1 g kg-1 (Salinity)Atmos: 0.1 m s-1 (Wind), 0.1 K (Temp)Ocean: 0.5 K (Temp), 0.1 g kg-1 (Salinity)Atmos: 0.1 m s-1 (Wind), 0.1 K (Temp)*Ocean_eq: 100 K (Temp), 20 g kg-1 (Salinity) Atmos : 0.1 m s -1 (Wind), 0.1 K (Temp)     Data Assimilation “Assimilation Increment” Model estimate Intro – BGC Initialization data – Retrospective Prediction

DA-BGC NO3 (ECDA-COBALT) NO3 (OBS) NO 3 (surface) O 2 (600-200m) CHL (surface) OBS Degraded BGC fields in the tropics CTRL [mg/m 3 ] [ mol /kg (×10 -6 ) ] [ mol /kg (×10 -4 ) ] Intro – BGC Initialization data – Retrospective Prediction

NO 3_eqO2_eq Degraded subsurface BGC at the Equator DA-BGC CTRL OBS [ mol /kg (×10 -4 ) ] [ mol /kg (×10 -6 ) ] Intro – BGC Initialization data – Retrospective Prediction

Degraded CHL correlation skill Monthly anomaly correlation (DA-BGC vs. OBS) Sep1997-dec2015 Intro – BGC Initialization data – Retrospective Prediction

[m/s (×10-6)] W[m/s (×10-6)]Spurious velocity problem DA-BGC CTRL + Residual (Bio + diffusion + nonlinear high frequency advection)           (Bio + diffusion + nonlinear high frequency advection) [ mol /kg/s (×10 -12 ) ] Intro – BGC Initialization data – Retrospective Prediction

  Zonal pressure gradient Wind stress Near surface zonal momentum equation [m/s (×10 -6 ) ] W Spurious velocity problem Common feature in ocean assimilation system (e.g. Burgers et al. 2002, Xie and Zhu 2007, Waters et al. 2016) Due to lack of balance in assimilation increment Less harmful to eq. physics compared to gains from DA, but big obstacle toward BGC prediction [m/s (×10 -6 ) ] DA-BGC CTRL Intro – BGC Initialization data – Retrospective Prediction

What is the viable approach to address the problem? Intro – BGC Initialization data – Retrospective Prediction

Strong trade winds bias at the Equator [m/s] NO ATM assim (default) DA- BGC_Atm U 10m DA-BGC NCEP2 Intro – BGC Initialization data – Retrospective Prediction

DA-BGC (Wind/ Tempobs_err= 1.0)DA-BGC (Stronger ATM const) Strong atmosphere data constraint helps, but W more atmos obs info assimilated [m/s (×10 -6 )] Strong data constraint (Wind/ Temp obs_err= 0.5 ) Stronger data constraint(Wind/Temp obs_err= 0.1 ) NO 3 [ mol /kg (×10 -6)] Intro – BGC Initialization data – Retrospective Prediction

CHL correlation skill still not recovered Monthly anomaly correlation (model vs. OBS)Sep1997-dec2015 Intro – BGC Initialization data – Retrospective Prediction

100 5 Temperature obs standard error (°C)Less observations assimilated Linear latitudinal ramp [m/s (×10 -6 )] [ mol /kg (×10 -6 ) ] Less obs info assimilated NO 3_ eq bias W T obs_err = 0.5 T obs_err = 5 T obs_err=100 CTRL 10N 10S Eq 0.5 Intro – BGC Initialization data – Retrospective Prediction Weak equatorial ocean data constraint

Tradeoff between BGC bias & Phy biasCompared withEN4 (Temp)WOA (NO3)  Optimal run: Strong Atmos data constraint, Weak Equatorial (10S-10N) Ocean data constraint Intro – BGC Initialization data – Retrospective Prediction Less ocean observation info More ocean observation info

Does this approach improve BGC field compared to the non-assimilative run (CTRL)? Intro – BGC Initialization data – Retrospective Prediction

[mg/m 3 ][mol/kg (×10-6)] [ mol /kg (×10 -4)] Modified DA improves global BGC simulation NO 3 (surface) O 2 (600-200m) CHL(surface) CTRL OBS DA- BGC_opt O 2 Intro – BGC Initialization data – Retrospective Prediction

Modified DA improves subsurface BGC simulation NO3_eq O2_eq DA- BGC_opt CTRL [ mol /kg (×10 -4 ) ] [ mol /kg (×10 -6 ) ] OBS Intro – BGC Initialization data – Retrospective Prediction

O 2_eqbias CTRL - OBSModified DA improves subsurface BGC simulationNO3_eqbias DA- BGC_opt - OBS [ mol /kg (×10 -4 ) ] [ mol /kg (×10-6)] Intro – BGC Initialization data – Retrospective Prediction

DA- BGC_opt CHL correlation skill DA-BGC_opt - CTRLCTRL Intro – BGC Initialization data – Retrospective Prediction

Global Map Eq. Section Modified DA improves BGC simulations Intro – BGC Initialization data – Retrospective Prediction

Intro – BGC Initialization data – Retrospective Prediction Produce BGC retrospective Perform retrospective forecasts Assess prediction skill (model- obs ) (Optimal initialization strategy) Preliminary results

1991 Retrospective prediction1992 199320162014· · · · · · · · · 2015 Targeting seasonal-to-multi-annual prediction 2-yr-long, 12-ensemble prediction run started every months Prediction skill assessment : Anomaly Correlation Coeff . (ACC) : Lead-time-dependent monthly-mean drift removed 1 JAN 1 Feb 1 Mar ×12 1 Apr ×12 ×12 ×12 ×12 Intro – BGC Initialization data – Retrospective Prediction

Lead time: 1-3 mon 4-6 mon 7-12 mon 13-24 mon BGC prediction skill (global) Prediction skill: anomaly correlation Mean prediction skill Maximum prediction skill Intro – BGC Initialization data – Retrospective Prediction Chlorophyll anomaly correlation (Model vs. OBS)

Lead time: 1-3 mon Initialization month Forecast Lead timeBGC prediction skill (Regional) Intro – BGC Initialization data – Retrospective Prediction

Intro – BGC Initialization data – Retrospective Prediction ENSO, source of BGC prediction skill ENSO Years (97, 98, 99, 00, 03, 08, 10, 11, 12) Non-ENSO Years

Intro – BGC Initialization data – Retrospective Prediction Potential utility for annual fish catch prediction http://marineregions.org Large marine ecosystems (LMEs) LMEs selection criteria: 1. Fish catch is dominated by bottom-up forcing (SST, CHL) 2. Model can predict SST & CHL. 3. Predicted SST & CHL explain reported fish catch. Reported annual fish catch data in LMEs (LMEs account for 95% of global fish catch)

Gulf of Alaska California Current Agulhas CurrentFish Catch Prediction SkillLead time: 1-2 year Lead time: 0-1 year Potential utility for annual fish catch prediction Intro – BGC Initialization data – Retrospective Prediction Bottom-up forcing dominant? Predictable SST & CHL? Predicted SST & CHL explain fish catch?

Summary How to fix the spurious velocity problem?Initialization strategy: integrating BGC model with physical data assimilation : Significant BGC biases along the equator due to spurious upwelling. : Strong atmosphere data constraint helps, but not enough. : Eliminating spurious velocities further required enforcing stricter fidelity to the internal model dynamics over ocean data constraintsRetrospective Prediction : Significant BGC biases introduced along the Equator The BGC bias is related to the vertical velocity field which is caused by wind bias in Atmos model and by momentum imbalance from DA Produce BGC retrospective Perform retrospective forecasts Assess potential predictability & prediction skill Strategy: Integrating BGC model with physical data assimilation . Substantial BGC bias along the equator due to spurious upwelling . Addressed by enforcing stricter fidelity to model dynamics over ocean data constraints. Improved BGC with optimally constrained run. Up to 2 year of BGC prediction skill. ENSO, a key source of prediction skill. Potential utility for fish catch prediction is promising! “Thank you for your attention”

Back up

CHL SST

Predictability Prediction skillSST

Predictability Prediction skillCHL

[°C/day] Off-equatorial data constraints Strong data constraintWeak data constraint

a bd c[°C/day]

a bc [°C/day]

Mechanistic understanding of BGC predictability - Diagnose missing/insufficient processes in the modelClosing the prediction skill gap Intro – BGC Initialization data – Retrospective Prediction Phytoplankton Improvement of ESM - e.g. coupling of atmosphere-ocean aerosols : Improved chlorophyll correlation when using AM4 Iron deposition (0.38  0.64 in NINO3 region )

ECDA-COBALT (Atmos+Ocean+BGC)NCEP 4xdaily dataOcean Observations1994-2015(12 ensembles)Exp 3. ECDA-COBALT run - Fully integrated BGC with physical AssimilationClosing the prediction skill gapBiogeochemical assimilation - Ocean biogeochemistry profile data (Biogeochemical Argo floats) Intro – BGC Initialization data – Retrospective Prediction