computational modelling of the plankton ecosystem
DESCRIPTION
Computational Modelling of the Plankton Ecosystem. Tony Field. Contributors. John Woods Roger Wiley Tony Field Silvana Vallerga. Wes Hinsley Matteo Sinnerchia Jeremy Cope Mohammad Raza. Angelo Maggiore Reza Adams Samir Al-Battran Massoud Aref Wolfgang Barkmann Tim Barrell - PowerPoint PPT PresentationTRANSCRIPT
Computational Modelling of the Plankton Ecosystem
Tony Field
Angelo Maggiore
Reza Adams
Samir Al-Battran
Massoud Aref
Wolfgang Barkmann
Tim Barrell
Alan Brice
Matt Booth
Cheng-Hua Chang
James Duggin
Lucas Partridge
Simon Smith
Kevin Stratford
Sarah Talbot
Jana Tharmaratnam
Dave Turner
Stephen Warren
Uli Wolf
Pak-Wing Fok
Sam Gratrix
Chris Harris
Chris Hurt
Ben Jefferys
Cheng-Chien Liu
Katrina Lythgoe
Camille Maclet
Enrique Nogueira
Darren Osborne
John WoodsRoger WileyTony FieldSilvana Vallerga
Wes HinsleyMatteo SinnerchiaJeremy CopeMohammad Raza
Contributors
Goals“Improved scientific understanding”
– Stability– Climate– Toxic blooms– Disease– Fisheries– Environmental managementetc.
Phytoplankton
• Fix atmospheric carbon via photosynthesis– Basis of the “carbon pump”– Influence atmosphere and climate
• Form the base of the food chain– Key to fisheries ‘recruitment’
• Nasty surprises– Red tides
Diatom
Zooplankton
• Grazers– Dominant consumers of phytoplankton
• Transporters– Eat at one depth, excrete at another
• Nasty surprises– Cholera
Copepod
Modelling the Ecosystem
• Traditionally population-based– Coupled ODEs– Each defines one continuous “compartment”
• The seminal model is Fasham’s (1990)– Compartments are Phyto, Zoo, Bact, Detritus,
Nit, Amm, DON– The “currency” is nitrogen (mmol N/m^3)…
= Growth – Grazing – Mortality – Mixing/dilution
M
Pthm
Pk
PGPrNnNPMt
dt
dP ))((),,,,()1(
21
1
M
Zth
Zk
ZGGG
dt
dZ )(
2
22
332211
M
Bthm
Zk
ZBGU
dt
dB ))((
2
22
32
= Grazing – Predation – Dilution
= Uptake – Grazing – Mortality – Mixing/dilution
Etc.
BUT… Demography is not an artefact of nature!
• Individual-based Models (IBMs)– Primitive phenotypic equations based on
experimental observation– Models compute individuals’ trajectories– Bio-chemical/bio-optical feedback from
individuals– Demography is an emergent property– Sound scientific basis– Demonstrably stable over a
wide range of parameterisations
• IBMs are expensive– interactions per time step
• A compromise: Lagrangian Ensemble (LE)– Agents represent subpopulations– Interactions between agents and fields
)( 2nO
Concentration
Populationsin
i
sin
TP
Agents Ingestion
The 1D LE Metamodel• Models a single “water column”
(mesocosm)– Horizontal correlations >> vertical
• Mesocosm advected by ocean circulation• External forcing by sun and atmosphere
– Equations for solar elevation– ERA40 data for atmosphere (from ECMWF)– Other ‘scenario’ factors, e.g. ApCO2, Fe…– Initial conditions from NOAA world ocean
atlas
• Cannot assume ‘homogeneity’
WindSunCloud
Mixing layer
Laminar flow
Turbocline
Temperature
Thermocline
Irradiance
Turbidity (bio-optical feedback)
Surface
+Upwelling(ignored)
• LE Metamodel Encapsulated by the Virtual Ecology Workbench (VEW)
• Supports building of new models and archiving of old models
• End users: ecologists, students…– Configure & integrate models; analyse data…
• Model designers: biological oceanographers– Define “agent” behaviour and chemistry via
primitive equations
The “VEW”
VEW Components• Data viewer
• Model designer
• Planktonica
• Compiler
• Species builder
• Scenario builder
• Controller
• Particle manager
• Output controller
• Run manager
• LiveSim
• VEWAnalyser
• VEWDocumenter
• The VEW assumes a stratified virtual mesocosm…
…
1
2
L
3
…Physics Biology
• Agents move between layers (by swimming, turbulence, gravity, …)
• “Visible” model ingredients…
Layer i
Irradiance
ChemistryPhysics (fixed)
vI
Temp T
Ambient concC
MLDepth
poolCingestC
uptakeCz
Other vars
User-defined, each with specified units
User-defined Cx defined automatically for each C
Globals
t timestep
Agents and Biodiversity• New agent types (Functional Groups, FGs)
may be defined in the VEW
• FGs may have several life cycle stages1 2 3 4
• FGs can be parameterised to define species
• Behaviour in each stage specified by rules- Each rule modifies agent’s internal state- Special functions control LE integration- Essentially a “single-assignment” language
The Curtain “API”
• Nutrient uptake/release– uptake(c,x) – requests uptake of of
chemical c from ambient environment– release(c,x) - similarly
These affect internal and external
• Ingestion (food web)– ingest( ) – both arguments are vectors
gx
poolC concC
rV ,
Encodes agent/field interaction, nutrient budgeting, logging, emergence…
• change(s) – changes the stage to s• pchange(p,s) – changes the stage to s with
probability p.
p
1-p
ss’
s
User view
ss’
s
Behind the curtain
Pop n
Pop mean np
Pop mean n(1-p)
• divide(m) – duplicate the agent m-fold
m
1
…
Pop n Pop mn
User view Behind the curtain
• create(s,m,as) – creates m new instances of the agent, each in stage s with internal state set by list of assignments as
User view Behind the curtain
Pop n
Pop mn
Pop n
vs’
v’s
v’s
v’s
…
0
1
2
m
v’s
v’ = internal state set by as
vs’
vs’
vs’
Some examples (from “WB”)Diatom energetics…
)(),)(min(
)(7.03.0
)(3600
)(
max
1
1)/(2
2
JEtEEEE
JhT
TRE
JheIrkE
Wmtt
IIII
respphoto
rLresp
IIvFphoto
a
mvmm
mv
Imported(reusable)
Some examples (from “WB”)Diatom energetics…
)(),)(min(
)(7.03.0
)(3600
)(
max
1
1)/(2
2
JEtEEEE
JhT
TRE
JheIrkE
Wmtt
IIII
respphoto
rLresp
IIvFphoto
a
mvmm
mv
Some examples (from “WB”)Diatom energetics…
)(),)(min(
)(7.03.0
)(3600
)(
max
1
1)/(2
2
JEtEEEE
JhT
TRE
JheIrkE
Wmtt
IIII
respphoto
rLresp
IIvFphoto
a
mvmm
mv
Some examples (from “WB”)Diatom energetics…
)(),)(min(
)(7.03.0
)(3600
)(
max
1
1)/(2
2
JEtEEEE
JhT
TRE
JheIrkE
Wmtt
IIII
respphoto
rLresp
IIvFphoto
a
mvmm
mv
0
50
100
150
200
250
300
350
0 4 8 12 16Time (hrs)
W/m
^2
0
10
20
30
40
50
60
Ep
ho
to (
J/h
* 1
0^
6)
IvImEphoto
Diatom nutrient uptake…
),(then0if
),(then0if
)(,1min
)(
max
1
AtdAsuptakedN
NtdNsuptakedN
tdN
ANNs
gNhkAA
Au
kN
NudN
poolpool
conc
concA
Nconc
concN
Modifies droop ‘pools’by uptake from environment
concA
concNdN spoolNpoolA
Death and decay…)(then0if DeadchangeE
Then, in the ‘Dead’ stage…
),(
),(
tNNNrelease
tAAArelease
poolrem
poolrem
Bacterial remineralisation(from droop pool to environment)
Diatom reproduction…
)2(then)(andif divideNANEE divpoolpooldiv
Copepod ingestion of diatoms…
Imin
min
maxGT
grazed
grazedingest
ingestmin
kPP
PPf
Rzz
fFWP
RPR
RPPP
*
)*(
,),10max(
)(integratemin
),min(
)0elsethen*if*,(ingest
2
15
max
Prey vector Vector expression
Sums over trajectory
Function of ‘satiation’
Copepod ageing and reproduction…
)(thenif AdultchangeGC
tAgeAge
maxpool
0
0
5.0with
R
GC
A
S
minpool
r
),(thenif nNewborncreateAgeAge
G
GCn
rep
min
maxpool
Then, when Adult…
Some ResultsTaken from 6-year integrations at the Azores (41’N, 27’W)
Concluding remarks• First attempt at LE modelling environment
– Scientifically sound basis– Stable ecosystems (=> prediction)– Results already inform scientific understanding
• BIG advantages– Robust, well engineered, domain-specific– Many really useful tools– Modular, re-usable components– Static checking (consistency, units, types…)– Good housekeeping (budgeting, emergence…)
Q: Is “computational model” right?
A: Possibly not…– Think IBMs; LE is behind the scenes– Model food webs properly (A eats B!)– Derive interaction rates– Introduce stochastic time delays (currently
history-based)– Fix the “language”– Turbulent advection
etc.
Warning!• Interdisciplinary projects are slippery beasts
– Avoid offering a programming ‘service’!– Stick to focused science/computing agenda
• BUT, interesting CS can emerge, e.g.Languages
Type systems
Type mutation (Fickle)
Memory management
Meshing (AFEM)
Parallelisation
Code optimisation
Radiation transport
etc.
A Quick Tour of the VEW
• Ambient grazing/predation “rate”
If T is time to next interaction and target interactions areWhat is niITP n ,...,1),,( ?
nII ,...,1