1D relocatable BHM;

1
(No Transcript)
2
Goals
Estimate ocean ecosystem model parameters, and
quantify parameter uncertainty for coastal
domains spanning the North Pacific Ocean
Quantify impacts of climate-scale variability on
coastal ocean ecosystems
Demonstrate the feasibility and advantages of
Bayesian Hierarchical Models (BHM) for large
state-space ocean ecosystems
Objectives

• 1D relocatable BHM
• Data Stage Inputs Regional Obs, Regional
  ROMS output
• Process Model Stage NPZD, NEMURO, Error
  Models, dynamics
• Climate Scale Calculations (1D BHM)
• Data Stage Inputs Pac Boundary Ecosystem
  Climate Project (Di Lorenzo et al.)
• NCAR OGCM, ROMS-Pacific Basin
• 3D Coastal Domain BHM
• Forest of statistically-linked 1D BHM
• Conventional 3D

3
Bayesian Estimation Cartoon
Model for Process of Interest e.g. Phytoplankton
Abundance
mmolN m-3
4
Bayesian Estimation Cartoon
Measurement Error Model e.g. estimates based on
fluorometer readings
mmolN m-3
5
Bayesian Estimation Cartoon
Posterior Distribution Prior updated by
Observation distribution (normalized)
mmolN m-3
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
What do we get from a BHM?

• Distributions
• mode is most likely state, distribution
  (spread) is uncertainty
• animations of posterior mean, uncertainty
  maps, summary fields
• parameter posterior distributions
• are the model parameters identifiable
  given the data?
• partition uncertainty i.e. biological
  components vs. physics
• Conditional Probabilities
• diagnose/compare dependencies (i.e.
  top-down/bottom-up, webs)
• multi-platform (disparate) data stages
• borrowed support from well-known
  distributions to less well-known
• Model and Array Design
• identify next most explanatory term
• identify next most informative observation

15
ROMS-NPZDFe Sea Surface Height Annual Average
CGOA
CCS
CCS dynamical model from C. Edwards and M.
Veneziani Biology implemented by J. Fiechter
WPAC
New WPAC ROMS-NemuroFe and ROMS-NPZDFe
implemented by J. Fiechter and A. Moore
CGOA and WPAC physical boundary conditions from
N. Pacific ROMS due to JAMSTEC E. Curchitser and
E. Di Lorenzo
16
ROMS-NPZD-Fe Surface Chlorophyll Annual Average
(sample data stage
inputs)
CGOA
CCS
WPAC
17
SEAWIFS Chlorophyll 2001 Annual Mean
(use comparison with
ROMS-NPZDFe to estimate error)
CGOA
CCS
WPAC
WPAC
18
Data Stage Input Choices Compare ROMS-NPZD,
ROMS-NEMURO, SeaWIFS
19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
(No Transcript)
23
1D NPZDFe BHM CGOA Initial Results
GLOBEC GAK line data only
Data Influence on
Posterior Mean Trace
Forward integration of Process Model (no
Bayesian estimation)
Mean of posterior distribution from 1D NPZDFe
BHM at one level on inner shelf profile

Observed N concentration at 19.375m on GAK line
(inner shelf)
24
Summary
BHM for 1D-NPZDFe using data from obs
ROMS Preliminary runs exhibit Bayesian learning
and mixing BHM to be validated via sequence of
1D calculations NPZDFe, NPZD, NPZ, NP, N,
P Test issues of uniqueness in solutions
25
Issues
Issues
Data Stage

• Data volume
• Data importance (uncertainty)
• Data timing

Process Model

• Incl. vertical adv. and vertical mixing
• Identify correlated parameters
• Fixed vs. random parameters

26
EXTRAS
27
(No Transcript)
28
(No Transcript)
29
1D NPZDFe BHM CGOA Initial Results
ROMS-NPZDFe data only
Fe limitation vs.
offshore position
T_Fe
k_FeC
FeRR
inner shelf
mid shelf
outer shelf
30
(No Transcript)
31
1D NPZDFe BHM CGOA Initial Results
ROMS-NPZDFe data only
Vm_NO3 convergence vs.
offshore position
Vm_NO3
MCMC iteration trace
inner shelf
mid shelf
outer shelf
Phytoplankton Nitrate Uptake Rate
22K iterations
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)