Stochastic Plume Simulations for the Fukushima

Accident and Deep Water Horizon Oil Spill

Emanuel Coelho1

G.Peggion1, C.Rowley2, P. Hogan2

1 University of New Orleans (resident at Naval Research Laboratory) 2 Naval Research laboratory

EGU-Vienna Austria, April 2012

2

COMPLEX Littoral Ocean Dispersion Dynamics

and limited knowledge of the exact amounts of released

radiation makes difficult predicting OCEAN IMPACTS

Tides

River inflow and

E-P

Air-Sea Fluxes ?

ATMOSPHERIC and COASTAL FORCING

OCEAN RESPONSE

Atmospheric Dispersion ?

Surface Heat Flux

Wind

Ocean Dispersion ?

Local currents and shear

Space-time Variability

Coupled Mixing Layer

Wind Waves

Coastline Coastal Topography

Bathymetry

M-S/O

M-S/A

OCEAN

IMPACT ?

RADIATION

INPUT

FROM

UNITS ?

3

COMPLEX Littoral Ocean Dispersion Dynamics

and unknown amounts of oil at surface and subsurface

makes difficult predicting OCEAN IMPACT

Tides

River inflow and

E-P

Surface and Biology

interaction ?

ATMOSPHERIC and COASTAL FORCING

OCEAN RESPONSE

Surface Heat Flux

Wind

Ocean Dispersion ?

Local currents and shear

Space-time Variability

Coupled Mixing Layer

Wind Waves

Coastline Coastal Topography

Bathymetry

M-S/O

M-S/A

OCEAN

IMPACT ?

OIL INPUT

FROM

DEEPWATER

WELL ?

Oil reaction, dispersion and

collection ?

4

4

Assimilation:

next ensemble

initial state

prior ensemble forecast

X0

B(X0)

Xa

B(Xa)

observations

Xf,new

B(Xf,new)

Assimilation:

next ensemble

initial state

new ensemble forecast

forecast control state forecast error variance forecast covariance

sonic layer depth (SLD) covariance of T,S on grid

with forecast SLD in box

Forecast Perturbation observations

STOCHASTIC FORECASTS: ERROR VARIANCES – e.g. ACCOUNTS FOR FORECAST QUALITY INFORMATION

ERROR COVARIANCE INFORMATION – e.g. ALLOWS CASCADING AND

EXTRAPOLATING UNCERTAINTY/ERROS TO HIGHER END PRODUCTS AND DATA ASSIMILATION

Forecast Perturbation

5

East Coast of Japan set-up:

• 2km resolution nest including tides and local data assimilation

• 32 Member ensemble

• Area initialized on model day March 1, 2011

• Run up to present

• Running daily on R&D computers

• 48 hour T,S,U,V ensemble predictions

• Particles released at constant rate starting March 11, 2011

• Ensemble dispersion RAC produced daily on R&D computers

Ensemble Prediction of Dispersion Probability Distribution

6

Example of Ensemble tracks

for Jun, 20 2011 multiple layers, initial states and atmospheric forcing produce

different sets of possible tracks in the upper ocean

SURFACE 25m Depth

7

Risk Assessment Code (RAC) defines a

stochastic plume by weighting Severity (i.e.

amounts on the same order as at source)

along with Probability of presence of

traceable amounts

Approach:

• Inject hourly tracers on each ensemble

member starting from the plant location

• Predict tracers trajectories and

accumulate through forecasts

• Ensemble trajectories are used to

predict a probability distribution of tracer

concentrations at each grid point

RISK ASSESSEMENT CODE PLUME

Objective: Identify areas in the ocean where traceable amounts of

radioactive water could be found, assuming a constant release

Prediction of plume advection

(and dispersion) from a fixed

source over periods larger

than 24 hours can lead to

substantial errors that

continue to grow in

subsequent days

8

Ensemble Prediction of

Dispersion Probability

Distributions

RAC example from the real-

time run

Data delivered in netcdf format and pre-

formatted maps

9

Measurements Reported

March, 28 2011

(RAC map refers to 2011032700)

IMPACT Rules:

very high <= 5km

high <= 10km

medium <= 15km

minimal <=20km

(sources: TEPCO, JAMSTEC, AIAA)

10

Gulf of Mexico set-up:

• 3km resolution nest including tides and local data assimilation

• 16 Member ensemble (32 member available soon)

• Run from April 1st, 2010 to September, 10th 2010

• Running daily updates on R&D computers

• 72 hour T,S,U,V ensemble predictions

• Particles released at constant rate and advection run with

lower estimates of oil released starting April 20th, 2010

• Ensemble dispersion RAC produced daily on R&D computers

Ensemble Prediction of Dispersion Probability Distribution

11

Oil Map integrating NOAA calibrated

estimates and actual observations

RELO/NCOM simulations on a

3km resolution grid,

assimilating local and remote

ocean data only.

Advection scheme uses an 5

days time scale for the oil and

assumes a Gaussian oil source

based on the lower estimates

for releases every 12 hours.

CONTROL RUN

12

Oil Map integrating NOAA calibrated

estimates and actual observations

RELO/NCOM 16 members

ensemble simulations on a 3km

resolution grid. The

control/central run assimilates

local and remote ocean data

only.

Advection scheme uses an 5

days time scale for the oil and

assumes a Gaussian oil source

based on the lower estimates

for releases every 12 hours.

ENSEMBLE MEAN

13

RISK ASSESSEMENT CODES

for oil concentrations based on

the 16 member RELO/NCOM

ensembles.

RISK ASSESSEMENT

CODES

LAGRANGIAN (tracers dispersion)

EULERIAN (solution ADR eq.)

14

Ensembles simulations allow integrating multiple levels of unknown

inputs and produce valuable planning information

Stochastic Dispersion Forecasting – FUTURE WORK

PLANNING TOOLS include:

• Monitoring priorities

• Shipping areas of Awareness

• Sensitive Coastal areas

• Mitigation Op. Planning

NEAR FUTURE WORK FOLLOWS UNDER THREE PROGRAMS:

- Ocean Tracer Concentration and Plume Simulations

(ONR)

- CARTHE (Gulf of Mexico Consortium, GRI)

- Long Range Forecasting in the Gulf of Mexico (GOMEX-

PPP Phase 2)

15

END Questions ?