An uncertainty assessment approach for an isotope

FACULTY OF SCIENCE
An uncertainty assessment approach for an
isotope-enabled carbon cycling model of a
sub-tropical estuary
AUSTRALIAN COASTAL OCEAN MODELLING AND OBSERVATIONS WORKSHOP (ACOMO) 2014
Sri Adiyanti1), Matthew R. Hipsey1), Isaac Santos2), Damien T. Maher2), Bradley Eyre2)
ARC
Linkage
Project
in collaboration
with CCB-SCU
& MBRC
The University of Western Australia
MOTIVATION


Large spatiotemporal variability in water quality of contributing environments &
biogeochemical processes along freshwater-marine continuum.
This variability are not captured with routine monitoring programs.
Needs a 3-D Hydrodynamic-Biogeochemistry Modelling
Challenges: Complex, Over-parameterization, Model Equifinality
Our modelling framework designed to:
(1) reduce model equifinality : application of an “isotope-enabled”
biogeochemical model.
(2) approximate uncertainty on model parameters : application of isotope model
within a Bayesian Hierarchical Framework (BHF).
The University of Western Australia
MODEL EQUIFINALITY
“Wide ranges of parameter values subject to complex multivariate relationships
that result in plausible observed behaviours and produce equivalently accurate
predictions” (Arhonditsis, et al., 2008)
“Different input factors (model structures/parameter sets) being acceptable
simulators of the natural system” (Beven, 1993)
Right results for the wrong reason
Right result for the right reason
The University of Western Australia
UNCERTAINTY or VARIATIONS?
 Sources of Uncertainty:
 Model structural assumptions
 Inputs (flows, loads & meteorology)
 Model parameter values -->50 parameters
 Error / bias etc. in observed data
 Sources of variations: Intrinsic behaviour of the system
The University of Western Australia
FRAMEWORK
1.
Estimate water retention time;
2.
Auto-calibration using Markov
Chain Monte Carlo (MCMC)-based
search algorithm with a steady
state 1-D mixing model;
3.
Define model parameter
“posteriors” (ie: most likely
parameter range);
4.
Run fully dynamic
3-D estuary model,
including isotopes using
parameters from 3 …
The University of Western Australia
Study Site
Caboolture River Est
Moreton Bay
8
The University of Western Australia
Identify, but not
Included yet!!
The University of Western Australia
full spatial domain
Coupled hydrodynamic-biogeochemical model system
Nonspatial
driver
0D
GLM
General
Lake
Model
1D
GOTM
General Ocean
Turbulence
Model
1D
GETM
General
Estuarine
Transport Model
3D
MOM4
Modular
Ocean
Model
3D
TUFLOW FV
Unstructured
mesh model
3D
ROMS
Regional
Ocean
Modelling
3D
Application Programming Interface
Framework for Aquatic Biogeochemical Models
Application Programming Interface
local point in space
Biogeochemical & Ecological Modules
Fasham et
al. 1990
NPZD
Fennel &
Neumann
(1996)
Simple
tracer
module
Benthic
dynamics
Carbonate
Buffering
& pH
ERSEM
Mnemiopsi
s
ERGOM
Neumann
et al.
(2002)
Phytoplankton
feedbacks
on
hydrodynamic
s
AED
AED
Aquatic
AED
Eco-AED
dynamics
aed_isotopes
C/O/N
Other
models
etc…
Courtesy: Bolding & Bruggeman
The University of Western Australia
COMPUTATIONAL DOMAIN
The University of Western Australia
Hydrodynamic Model Assessment
The University of Western Australia
Model
Field
Hydrodynamic Model
Assessment
Salinity Contour
Model
Field
Model
Field
The University of Western Australia
FRAMEWORK
1.Estimate water retention time;
2. Auto-calibration using Markov Chain
Monte Carlo (MCMC)-based search
algorithm with a steady state 1-D mixing
model;
3.Define model parameter “posteriors” (ie:
most likely parameter range);
4.Run fully dynamic 3-D estuary model,
including isotopes using parameters from 3
…
The University of Western Australia
C ISOTOPE-ENABLED MODEL
Field: CR1 (Dec 2011) – CR5 (May 2012)
The University of Western Australia
C-ISOTOPE MODEL PARAMETERS
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
mineralisation rate (rmin)
fractionation of 13CDOC during mineralisation (αDOC,min)
mortality rate (rmor)
carbon uptake rate (Cupt)
fraction of detritus in POC (fdet)
fraction of DOC release due to mortality (fdoc)
decomposition rate (rdecom)
release from sediment or resuspension rate (rres)
The University of Western Australia
1-D MIXING MODEL
C, mmol/m3
fx =(Ss-Sx)/Ss
River
Ocean
fx
The University of Western Australia
Biogeochemical reaction -augmented mixing model
 CDOC(x) = CDOCmix(fx) - CDOCDIC(τ) + CDOC,sedDOC,wat(τ) + CPOCDOC,wat(τ)
 CDIC(x) = CDICmix(fx) + CDOCDIC(τ) - CDICPOC(τ) + CPOCDIC(τ)

13C
DOC(x)

13C (x)
DIC
= 13CDOCmix(fx) - 13CDOCDIC(τ) + 13CDOC,sedDOC(τ) +13 CPOC,DOC(τ)
= 13CDICmix(fx) +13CDOCDIC(τ) - 13CDICPOC(τ) + 13CPOCDIC(τ)
τ = retention time at point x, calculated for each cruise from
TUFLOW-FV simulation
Other parameters, α, θmin, kCO2, kmin, FDOC, μGTH etc….
The University of Western Australia
Conservative vs (Conservative Mixing + Source/Loss Terms)
example: CR1 (14 Dec 2011)
95% ci for model parameter uncertainty
95% confidence interval (ci)
≈ (µ+1.96SD)
Low (<1 m3/s) to
medium
(<50 m3/s) flow
River
95% ci for observations uncertainty
Ocean
Conservative mixing
Conservative mixing plus
Source-Loss term
Field
The University of Western Australia
FRAMEWORK
1.Estimate water retention time;
2. Auto-calibration using Markov Chain
Monte Carlo (MCMC)-based search
algorithm with a steady state 1-D mixing
model;
3.Define model parameter “posteriors” (ie:
most likely parameter range);
4.Run fully dynamic 3-D estuary model,
including isotopes using parameters from 3
…
The University of Western Australia
Example: Parameters Posteriors for CR1
The University of Western Australia
FRAMEWORK
1.Estimate water retention time;
2. Auto-calibration using Markov Chain
Monte Carlo (MCMC)-based search
algorithm with a steady state 1-D mixing
model;
3.Define model parameter “posteriors”
(ie: most likely parameter range);
4. Run fully dynamic 3-D estuary model,
including isotopes using parameters from
3…
The University of Western Australia
Combination of 8-parameter posteriors
 Combination of 8-parameter posteriors for CR1-CR5 (5 sets):
(1) µ
(2) µ+σ
(3) µ- σ
(4) min
(5) max
(6) median
 producing 30 sets of model parameters combinations
 30 sets of model outputs, from which the model uncertainty was presented.
The University of Western Australia
15 Dec 2011
DOC
CR1
CR1
DOC
µM
µM
DIC
µM
µM
DIC
µM
µM
Field vs 3-D and 1-D Model CR1 & CR2: DIC & DOC
Caboolture South STP ?
20 Jan 2012
CR2
CR2
The University of Western Australia
Field vs 3-D and 1-D Model CR1 & CR2 δ13CDIC & δ13CDIC
‰
δ13CDIC
‰
Burpengary East STP
CR1
‰
δ13CDIC
CR2
δ13CDOC
CR1
‰
Burpengary East STP
δ13CDOC
CR2
The University of Western Australia
Field
Model
The University of Western Australia
CARBON BUDGET
The University of Western Australia
CONCLUSION:
 The framework is flexible, i.e. parameters and variables/model data
likelihood can be updated simultaneously to produce a better
representative of posteriors.
 It can be used to identify the source of the bulk of simulation
uncertainty: hydrodynamic or biogeochemical processes?.
WHAT’S NEXT?
 Quantify the significance of “Identified but not yet included” discharges:
(1) Caboolture STP and Burpengary East STP discharges.
(2) Inclusion Submarine Groundwater Discharge.
(3) Stormwater Runoff Discharge downstream.
The University of Western Australia
ACKNOWLEDGMENTS:





Rayner Haase & Jessica Mowat of Moreton Bay Regional Council (MBRC)
Perrine Mangion (CCB-SCU)
Australian Research Council (LP110200975).
Brendan Bush and Casper Boon of Aquatic Ecodynamics SEE-UWA
BMT-WBM (http://www.bmtwbm.com.au/) and to FABM project team
(http://fabm.sourceforge.net/)
 MCMC: http://helios.fmi.fi/~lainema/mcmc.
The University of Western Australia