"Nesting and coupling of "Nesting and coupling of physical and biological physical and biological

models “models “

Albert J. HermannAlbert J. HermannUniversity of Washington JISAOUniversity of Washington JISAO

NOAA/PMELNOAA/PMEL

Collaborators: Dale Haidvogel, Sarah Hinckley, Collaborators: Dale Haidvogel, Sarah Hinckley, Elizabeth Dobbins, Phyllis Stabeno, Kate Hedstrom, Elizabeth Dobbins, Phyllis Stabeno, Kate Hedstrom,

Enrique Curchitser, Dave Musgrave, Georgina Enrique Curchitser, Dave Musgrave, Georgina Blamey, Bern MegreyBlamey, Bern Megrey

OUTLINEOUTLINE

• General commentsGeneral comments

• Physical BCsPhysical BCs

• Biophysical methods Biophysical methods – pollock in the Gulf of Alaskapollock in the Gulf of Alaska– salmon in the NE Pacific salmon in the NE Pacific

• Conclusions and points of contentionConclusions and points of contention

What should coupled, regional, What should coupled, regional, biophysical models include?biophysical models include?

• Mesoscale features and variability Mesoscale features and variability – eddies eddies – coastal trapped wavescoastal trapped waves– frontsfronts

• Multiple trophic levelsMultiple trophic levels

• Tidal effects Tidal effects – vertical migration interacts with tides -> mean vertical migration interacts with tides -> mean

advection of organismsadvection of organisms– mixing produces fronts, supplies nutrients mixing produces fronts, supplies nutrients

The Challenge of Coupled The Challenge of Coupled Regional ModelsRegional Models

• Zymurgy's First Law of Evolving Zymurgy's First Law of Evolving System DynamicsSystem Dynamics““Once you open a can of worms, the Once you open a can of worms, the only way to re-can them is to use a only way to re-can them is to use a larger can”larger can”

• Long history of attempted solutions to Long history of attempted solutions to the BC problem (some involving larger the BC problem (some involving larger cans)cans)

Robust approaches to the Robust approaches to the physical BC problem do existphysical BC problem do exist• Palma and Matano (2000) surveyPalma and Matano (2000) survey

– relaxation-radiation hybrid had “best overall relaxation-radiation hybrid had “best overall performance”performance”

• Marchesiello et al. (2001)Marchesiello et al. (2001)– nudge weakly for outgoing informationnudge weakly for outgoing information– nudge strongly for incoming informationnudge strongly for incoming information– apply sponge near boundaryapply sponge near boundary– allow oblique radiationallow oblique radiation

• Nudging is suboptimal, but simple and Nudging is suboptimal, but simple and robustrobust

Tidal and Subtidal DynamicsTidal and Subtidal Dynamics

• Need different BCs, butNeed different BCs, but can can exist peacefully exist peacefully together in one modeltogether in one model

• Output data can be tricky; beware of aliasing Output data can be tricky; beware of aliasing the tidal signalthe tidal signal

• Flather (2D) and Marchesiello (3D) solution:Flather (2D) and Marchesiello (3D) solution:– Flather: add/remove water at the free wave speed Flather: add/remove water at the free wave speed

to match specified SSH at boundary. Nice for 2D to match specified SSH at boundary. Nice for 2D tidestides

– Marchesiello: radiate with selective nudging. Nice Marchesiello: radiate with selective nudging. Nice for everything else for everything else

• (Telescoped solution; worked but had issues)(Telescoped solution; worked but had issues)

Biophysical MethodsBiophysical Methods

• Pollock in the Northern Gulf of AlaskaPollock in the Northern Gulf of Alaska

• Salmon in the Northeast PacificSalmon in the Northeast Pacific

Coupling Scheme used for Pollock Coupling Scheme used for Pollock StudiesStudies

Spatially Explicit IBM: follow Spatially Explicit IBM: follow individuals in 4 dimensionsindividuals in 4 dimensions

Known issues of IBMsKnown issues of IBMs

• Particles disperse; need to reseed the populationParticles disperse; need to reseed the population

• Two-way interaction with other species (e.g. the Two-way interaction with other species (e.g. the NPZ model) can be trickyNPZ model) can be tricky– could get unnaturally patchy prey fieldcould get unnaturally patchy prey field

• Single-point Lagrangian statistics can be misleadingSingle-point Lagrangian statistics can be misleading– spatially variable random walk -> preferentially accumulate spatially variable random walk -> preferentially accumulate

particles in areas with low dispersion (proper algorithms avoid particles in areas with low dispersion (proper algorithms avoid this artifact)this artifact)

– However, However, activeactive particles (e.g. swimming larvae) may in reality particles (e.g. swimming larvae) may in reality accumulate in areas with slower swim speed (not an artifact).accumulate in areas with slower swim speed (not an artifact).

So, why not use Eulerian So, why not use Eulerian approach for everything?approach for everything?• Need a huge number of variables (time of last feeding, Need a huge number of variables (time of last feeding,

what eaten, etc) at every gridpoint to track complicated what eaten, etc) at every gridpoint to track complicated “history” in Eulerian format“history” in Eulerian format

• Eulerian is always looking at the local average individual Eulerian is always looking at the local average individual (which is, on average, dead) and the local average (which is, on average, dead) and the local average attributes (which are misleading because of nonlinear attributes (which are misleading because of nonlinear interactions among species)interactions among species)

• Harder to encode complicated behaviors in Eulerian formatHarder to encode complicated behaviors in Eulerian format

• In principle, can go back and forth between NPZ and IBM In principle, can go back and forth between NPZ and IBM (Eulerian and Lagrangian) techniques (has been done for (Eulerian and Lagrangian) techniques (has been done for point dipsersal problems)point dipsersal problems)

Biophysical MethodsBiophysical Methods

• Pollock in the Northern Gulf of AlaskaPollock in the Northern Gulf of Alaska

• Salmon in the Northeast PacificSalmon in the Northeast Pacific

Nested Biophysical Models for GLOBEC:

NCEP/MM5 -> ROMS/NPZ -> IBM

GLOBEC NPZ GLOBEC NPZ model for the model for the

CGOA CGOA

(S. Hinckley)(S. Hinckley)

Width of arrow represents N flux

Model NestingModel Nesting

• Really a form of data assimilation, larger Really a form of data assimilation, larger model is “data”model is “data”

• Works best if the surface forcing is the Works best if the surface forcing is the same across all scales – o/w easy to get same across all scales – o/w easy to get discontinuities (and associated rim discontinuities (and associated rim currents) across boundariescurrents) across boundaries

• Here, use Marchesiello/Flather BC to feed Here, use Marchesiello/Flather BC to feed from coarser to finer gridfrom coarser to finer grid

• Tides (from a tidal model) applied only at Tides (from a tidal model) applied only at smaller scalessmaller scales

NESTED CIRCULATION MODEL DOMAINSNested Model Domains

Pameters of the Physical Models

NPAC Model SSH

NESTED MODELS

SSH DOY 255

CGOA domain - color

NEP domain – b&w

AK

CA

CGOA MODEL

SSH DOY 255

Sitka AK

Prince William Sound

Winds from MM5 (black) and NCEP Winds from MM5 (black) and NCEP (white)(white)

Flux through Shelikof Strait Black=ADCP data Red=model

MM5

NCEP

Conclusions/PreferencesConclusions/Preferences

1.1. Avoid spatial boundaries in dynamically Avoid spatial boundaries in dynamically active areasactive areas

2.2. Avoid spatial boundaries between Avoid spatial boundaries between different biological communities different biological communities

3.3. Consider the different time scales of Consider the different time scales of different variables when setting BCsdifferent variables when setting BCs

4.4. Two-way coupling is Two-way coupling is notnot necessarily necessarily better than one-way nestingbetter than one-way nesting

5.5. Visualization matters!Visualization matters!

2. Avoid spatial boundaries 2. Avoid spatial boundaries between different communities between different communities (BCs are tricky at the (BCs are tricky at the ecotoneecotone) )

• Gradient in community structure/limiting nutrient Gradient in community structure/limiting nutrient can produce artifacts as reactive materials cross can produce artifacts as reactive materials cross the boundary and seek new “equilibrium”the boundary and seek new “equilibrium”– Example 1: deep ocean is Fe limited, coastal Example 1: deep ocean is Fe limited, coastal

ocean is not -> spurious ocean is not -> spurious bloom at boundarybloom at boundary as offshore as offshore water moves onshorewater moves onshore

• Impedance change across boundary (different Impedance change across boundary (different boxes for different system) may lead to boxes for different system) may lead to discontinuity in biological variablesdiscontinuity in biological variables

Nested Biophysical Models for GLOBEC:

NCEP/MM5 -> ROMS/NPZ -> IBM

A big bloom occurred at the edge of the coastal NPZ model!

Onshore flow

Solution: add Fe variable, develop single NPZ model spanning both regions

6. Visualization matters!6. Visualization matters!

• Easy to miss incorrect features in a 3D field if limited Easy to miss incorrect features in a 3D field if limited to 2D visualsto 2D visuals

• High-end visualization becomes especially useful High-end visualization becomes especially useful when bio variables are added –e.g. where is a nutrient when bio variables are added –e.g. where is a nutrient source and who is using itsource and who is using it

• Many attributes of biophysical models are truly 3D Many attributes of biophysical models are truly 3D (e.g. patchy); perturbation signals, mixing(e.g. patchy); perturbation signals, mixing

• Visualize spatial paths from physical model, IBM in 3DVisualize spatial paths from physical model, IBM in 3D

3D view of particles in ROMS Gulf of Alaska simulation (use red/blue glasses for stereo 3D effect)

ALASKA

Cook Inlet

Alaska

Salinity Isosurface (32.6 psu)