Intro: Participants, motivation Methods: ( i ) Models, (ii) observations, (iii) skill metrics

• Intro: Participants, motivation

• Methods: (i) Models, (ii) observations, (iii) skill metrics

• Results (i): What is the relative hydrodynamic skill of these CB models?

• Results (ii): What is the relative dissolved oxygen skill ofthese CB models?

• Summary and ongoing work

Estuarine Hypoxia Modeling

OUTLINE

Carl Friedrichs (VIMS) and the Estuarine Hypoxia Team

Presented at U.S.IOOS/SURA Modeling Testbed All-Hands Annual MeetingWashington, DC, June 22, 2011


Carl Friedrichs (VIMS) and the Estuarine Hypoxia Team

Federal partners (*unfunded partner)• David Green* (NOAA-NWS) – Transition to operations at NWS• Lyon Lanerole (NOAA-CSDL) – Transition to operations at CSDL; CBOFS2• Lewis Linker* (EPA), Carl Cerco* (USACE) – Transition to operations at EPA; CH3D, CE-ICM• Doug Wilson* (NOAA-NCBO) – Integration w/observing systems at NCBO/IOOSNon-federal partners• Marjorie Friedrichs, Aaron Bever (VIMS) – Metric development and model skill assessment• Ming Li, Yun Li (UMCES) – UMCES-ROMS hydrodynamic model• Wen Long, Raleigh Hood (UMCES) – ChesROMS with NPZD water quality model • Scott Peckham (UC-Boulder) – Running multiple ROMS models on a single HPC cluster• Malcolm Scully (ODU) – ChesROMS with 1 term oxygen respiration model• Kevin Sellner (CRC) – Academic-agency liason; facilitator for model comparison• Jian Shen (VIMS) – SELFE, FVCOM, EFDC models• John Wilkin, Julia Levin (Rutgers) – ROMS-Espresso + 7 other MAB hydrodynamic models


Motivation for Studying Estuarine Hypoxia

(VIMS, ScienceDaily)

1) Help enable operational short-term (≤ weeks) estuarine hypoxia forecasts (NOAA-CSDL, NOAA-NCEP)

(UMCES, Coastal Trends)

2) Help enable long-term restoration scenario (≥ years) estuarine hypoxia forecasts (EPA-CBP)

Mid-summer oxygen differences across the pycnocline mirror the salinity difference across the pycnocline for mid-Chesapeake Bay stations, 1949-1980.

(Flemer et al. 1983)

Motivation (cont.): 3) Evaluate the ability of multiple models to hindcast salinity stratification and dissolved oxygen, given the paradigm that salinity stratification must be modeled well to predict estuarine hypoxia well.







OUTLINE


Methods (i) Models: 5 Hydrodynamic Models (so far)

(& J. Wiggert/J. Xu, USM/NOAA-CSDL)

o ICM: CBP model; complex biologyo bgc: NPZD-type biogeochemical modelo 1eqn: Simple one equation respiration (includes SOD)o 1term-DD: depth-dependent net respiration

(not a function of x, y, temperature, nutrients…)o 1term: Constant net respiration

Methods (i) Models (cont.): 5 Dissolved Oxygen Models (so far)

o ICM: CBP model; complex biologyo bgc: NPZD-type biogeochemical modelo 1eqn: Simple one equation respiration (includes SOD)o 1term-DD: depth-dependent net respiration

(not a function of x, y, temperature, nutrients…)o 1term: Constant net respiration

Methods (i) Models (cont.): 5 Dissolved Oxygen Models (so far)

o CH3D + ICMo EFDC + 1eqn, 1termo CBOFS2 + 1term, 1term+DD o ChesROMS + 1term, 1term+DD, bgc

Methods (i) Models (cont.): 8 Multiple combinations (so far)

Map of Late July 2004

Observed Dissolved Oxygen [mg/L]

~ 40 EPA Chesapeake Bay stationsEach sampled ~ 20 times in 2004

Temperature, Salinity, Dissolved Oxygen

Data set for model skill assessment:

(http://earthobservatory.nasa.gov/Features/ChesapeakeBay)

Methods (ii) observations: S and DO from Up to 40 CBP station locations

- How do we compare these models to data and to each other?- e.g., Is ChesROMS 1-Term (very simple) or CH3D-ICM (very complex) more accurate?- e.g., Which is modeled better, dissolved oxygen (DO) or salinity stratification (dS/dz)?

(from A. Bever)

Methods (iii) Skill Metrics: Target diagram

(modified from M. Friedrichs)

Dimensionless version of plot normalizes by standard deviation of observations







OUTLINE


unbiasedRMSD

[°C]

bias [°C]

unbiasedRMSD[psu]

bias [psu]

unbiasedRMSD

[psu/m]

bias [psu/m]

unbiasedRMSD

[m]

bias [m]

(a) Bottom Temperature (b) Bottom

Salinity

(c) Stratificationat pycnocline (d) Depth of

pycnocline

Outer circle in each case = error from simply using mean of all data

Inner circle in (a) & (b) = errorfrom CH3D model

Results (i): Hydrodynamic Model Comparison

- All models do very well hind-casting temperature.

- All do well hind-casting bottom salinity with CH3D and EFDC doing best.

- Stratification is a challenge for all the models.

- All underestimate strength and variability of stratification with CH3D and EFDC doing slightly better.

- CH3D and ChesROMS do slightly better than others for pycnocline depth, with CH3D too deep, and the others too shallow.

- All underestimate variability of pycnocline depth.(from A. Bever, M. Friedrichs)

ChesROMS

EFDC

UMCES-ROMS

CH3D

CBOFS2

Results (i) Hydrodynamics: Temporal variability of stratification at 40 stations

Mean salinity of individualstations

[psu]

- Model behavior for stratification is similar in terms of temporal variation of error at individual stations

(from A. Bever, M. Friedrichs)

Used 4 models to test sensitivity of hydrodynamic skill to:

o Vertical grid resolution (CBOFS2)o Freshwater inflow (CBOFS2; EFDC)o Vertical advection scheme (CBOFS2)o Atmospheric forcing – winds (ChesROMS; EFDC)o Horizontal grid resolution (UMCES-ROMS)o Coastal boundary condition (UMCES-ROMS)o 2004 vs. 2005 (in progress)

Results (i) Hydrodynamics (cont.): Sensitivity to model refinement

-0.2-0.4-0.6-0.8

0.2

-0.2

-0.2

CBOFS2

CBOFS2 model pycnocline depth is insensitive to: vertical grid resolution, vertical advection scheme and freshwater river input

Results (i) Hydrodynamics (cont.): Sensitivity of pycnocline depth


Stratification at pycnocline is not sensitive to horizontal grid resolution or changes in atmospheric forcing. (Stratification is still always underestimated)

CH3D, EFDC

ROMS

Stratification

Results (i) Hydrodynamics (cont.): Sensitivity of stratification at pycnocline


Bottom salinity IS sensitive to horizontal grid resolution

High horiz res

Low horiz res

Bottom Salinity

Results (i) Hydrodynamics (cont.): Sensitivity of bottom salinity








OUTLINE


Results (ii): Dissolved Oxygen Model Comparison

- Simple models reproduce dissolved oxygen (DO) and hypoxic volume about as well as more complex models.- All models reproduce DO better than they reproduce stratification.- A five-model average does better than any one model alone.


Results (ii): Dissolved Oxygen Model Comparison (cont.)


Average of these 5 models is better than any single model.

Results (ii): Dissolved Oxygen Model Comparison (cont.)Hy

poxi

c Vo

lum

e in

km

3

1-term DO model ICM (most complex model)

1-term DO model underestimates high DO and overestimates low DO: high not high enough, low not low enough

Total RMSD = 0.9 ± 0.1 Total RMSD = 0.9 ± 0.1

ChesROMS-1term (simplest model)

Results (ii) Dissolved Oxygen: Temporal variability of bottom DO at 40 stations

Mean bottom DO

individualstations[mg/L]


(by M. Scully)

Results (ii) Dissolved Oxygen: Top-to-Bottom DS and Bottom DO in Central Chesapeake Bay

ChesROMS-1term model

- All models reproduce DO better than they reproduce stratification.- So if stratification is not controlling DO, what is?

(by M. Scully)

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Date in 2004

Hypo

xic

Volu

me

in k

m3

20

10

0

Base Case

(by M. Scully)


Results (ii) (cont.): Effect of Physical Forcing on Dissolved Oxygen

Seasonal changes in hypoxia are not a function of seasonal changes in freshwater.


Date in 2004

Hypo

xic

Volu

me

in k

m3

20

10

0

Base Case

Freshwater river input constant

(by M. Scully)(by M. Scully)



Seasonal changes in hypoxia may be largely due to seasonal changes in wind.


Date in 2004

Hypo

xic

Volu

me

in k

m3

20

10

0

Base CaseJuly wind year-round





Date in 2004

Hypo

xic

Volu

me

in k

m3

20

10

0

Base Case

January wind year-round


Seasonal changes in hypoxia may be largely due to seasonal changes in wind.







• Accomplishments, scientific summary and ongoing work


OUTLINE


• Model-model and model-data comparisons for 5 hydrodynamic and 8 combined hydrodynamic-dissolved oxygen models (so far).

• Model forcings, grids, output, and observed data sets transferred to Testbed CI Team.• Progress on transitioning of findings to NOAA:

-- Team member L. Lanerolle (NOAA-CSDL) has incorporated our “1term” hypoxia formulation within the research version of NOAA-CSDL’s Chesapeake Bay Operational Forecast System.

-- A meeting between NOAA-NCEP and Estuarine Hypoxia Team PIs (coordinated by Team member D. Green, NOAA-NWS) is scheduled for July 5-6, 2011, to further hammer out transition steps for moving the CBOFS hypoxia model to NCEP.

• Progress on transitioning of findings to EPA:

-- Estuarine Hypoxia Team helped organize a June 9-10, 2011, workshop to provide advice to the EPA Chesapeake Bay Program on future estuarine modeling strategies in support of Federally mandated environmental restoration.

-- At the workshop, it was proposed that the Estuarine Hypoxia Team play a major role in establishing a Chesapeake Modeling Laboratory for developing and testing agency models (as recommended by a new National Academy of Sciences report to EPA).

ACCOMPLISHMENTS

• Available models generally have similar skill in terms of hydrodynamic quantities• All the models underestimate strength and variability of salinity stratification.• No significant improvement in hydrodynamic model skill due to refinements in:

– Horizontal/vertical resolution, atmospheric forcing, freshwater input, ocean forcing.

• In terms of DO/hypoxia, simple constant net respiration rate models reproduce seasonal cycle about as well as complex models.

• Models reproduce the seasonal DO/hypoxia better than seasonal stratification.• Seasonal cycle in DO/hypoxia is due more to wind speed and direction than to

seasonal cycle in freshwater input, stratification, nutrient input or respiration.– Note: This does not mean than inter-annual variation in nutrient input/respiration is unimportant.

• Averaging output from multiple models provides better hypoxia hindcasts than relying on any individual model alone.

SCIENTIFIC SUMMARY

• Available models generally have similar skill in terms of hydrodynamic quantities• All the models underestimate strength and variability of salinity stratification.• No significant improvement in hydrodynamic model skill due to refinements in:

– Horizontal/vertical resolution, atmospheric forcing, freshwater input, ocean forcing.

• In terms of DO/hypoxia, simple constant net respiration rate models reproduce seasonal cycle about as well as complex models.

• Models reproduce the seasonal DO/hypoxia better than seasonal stratification.• Seasonal cycle in DO/hypoxia is due more to wind speed and direction than to

seasonal cycle in freshwater input, stratification, nutrient input or respiration.– Note: This does not mean than inter-annual variation in nutrient input/respiration is unimportant.

• Averaging output from multiple models provides better hypoxia hindcasts than relying on any individual model alone.

• Why are the errors from multiple hydrodynamic models so similar, and why do they all underpredict salinity stratification?

SCIENTIFIC SUMMARY

From Warner et al. 2005, Ocean Modeling

Turbulence Models Have Lots of Coefficients!!They must be selected carefully!!

(slide from M. Scully)

Minimum TKE setting

Kzmin = 5×10-7 TKEmin=7.6×10-6

Kzmin = 5×10-8 TKEmin=7.6×10-6

Kzmin = 5×10-7 TKEmin=7.6×10-8

Importance of setting appropriate minimum value for TKE

Only when only the minimum value of TKE is sufficiently low, can model achieve the specified background diffusivity!!


Importance of setting appropriate minimum value for TKE


Proposed 2nd Year Efforts

1) Continue to work closely with NOAA-CSDL and NOAA-NCEP to transition our findings for use in short-term (≤ weeks) hypoxia forecast models used by NOAA.

2) Continue to work closely with the EPA Chesapeake Bay Program to transition our findings for use in scenario (≥ years) hypoxia forecast models used by EPA.

3) Further explore the model properties (including canonical settings) that lead to the inability of hydrodynamic models to capture the observed intensity of density stratification.

4) More fully include unstructured grid models in the Year 2 estuarine hydrodynamics and hypoxia intercomparison.

5) Expand the parameter space of model runs to include additional degrees of biological model complexity as well as coordinated, idealized sensitivity runs across multiple models.

6) Run hindcasts encompassing entire CB monitoring history (~30 years) to better understand controls on inter-annual variability in hypoxia.


Intro: Participants, motivation Methods: ( i ) Models, (ii) observations, (iii) skill metrics

Documents

dissolved oxygen models

estuarine hypoxia vims

mab hydrodynamic models

multiple roms models

ability of multiple

model domain

relative hydrodynamic

skill metrics results