Results from a multi-year, Results from a multi-year, multi-site AmeriFlux data multi-site AmeriFlux data assimilation with the assimilation with the TRIFFID model TRIFFID model Daniel Ricciuto Daniel Ricciuto Oak Ridge National Laboratory Oak Ridge National Laboratory June 14, 2007 June 14, 2007
22
Embed
Results from a multi-year, multi-site AmeriFlux data assimilation with the TRIFFID model
Results from a multi-year, multi-site AmeriFlux data assimilation with the TRIFFID model. Daniel Ricciuto Oak Ridge National Laboratory June 14, 2007. Carbon cycle uncertainty. C 4 MIP: comparison of 10 coupled “bottom-up” climate/carbon models - PowerPoint PPT Presentation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Results from a multi-year, Results from a multi-year, multi-site AmeriFlux data assimilation multi-site AmeriFlux data assimilation
with the TRIFFID modelwith the TRIFFID model
Daniel RicciutoDaniel RicciutoOak Ridge National LaboratoryOak Ridge National Laboratory
• No statistical assimilation of existing carbon cycle observations.
• Different parametric/structural representation of key feedbacks
• How to incorporate observations in a statistically meaningful way?
Friedlingstein et al. (2006)
Motivation
• Overall goal: to determine predictive uncertainty in CO2 sink strength• Questions assimilation can answer: Given a set of observations,
– How uncertain are model parameters?– Within a set of models, which is the most likely to be correct?
• The problems:– Most CO2 flux observations cover small spatial scales and/or time scales– Are model parameters relevant over multiple spatial scales?
Model predictionsParameters from Ecosystem-scale
observations
Past Future
uncertainty
Forcing and Global constraints
Spatiotemporal scales of current Spatiotemporal scales of current observations and modeling techniquesobservations and modeling techniques
< model gridscale > model gridscaleFIA
Cha
mbe
r flu
x
Tow
er f
lux
Global synthesis inversions
Airborne flux
Biogeochemical modeling
Regional inversions
CO2
Legend:
Observation
Top-down technique
Bottom-up technique
C4MIP
Spatiotemporal scales of current Spatiotemporal scales of current observations and modeling techniquesobservations and modeling techniques
< model gridscale > model gridscaleFIA
Cha
mbe
r flu
x
Tow
er f
lux
Global synthesis inversions
Airborne flux
Biogeochemical modeling
Regional inversions
CO2
Legend:
Observation
Top-down technique
Bottom-up technique
C4MIP
?
Multiple tower data assimilationMultiple tower data assimilation
FIAC
ham
ber
flux
Tow
er f
lux
Global synthesis inversions
Airborne flux
Biogeochemical modeling
Regional inversions
< model gridscale > model gridscale
CO2
Tower sitesTower sitesSite-years analyzed
(37 total)
WLEF: 1997-2004
Harvard: 1992-2003
Howland: 1996-2003
UMBS: 1999-2003
M. Monroe: 1999-2003
Forcing: wind, [CO2], PAR, Tair,
precip
Wood(NL)
Wood(BL)
Leaf(BL)
Root (NL) Root (BL)
GPP (NL,BL)
Leaf(NL) NEE
NPP(NL,BL)
RH
RaRa
Soil carbon (CS)
Simplified TRIFFID ModelSimplified TRIFFID Model• 22 model parameters
- initial soil carbon
- Photosynthesis
- autotrophic respiration
- Phenology
- soil moisture
• 4 PFT-specific
- Tlow, Tupp, nl,
Canopy structure
- initial NL/BL fraction
- initial canopy height
Why TRIFFID?Why TRIFFID?• Dynamic Global Vegetation Model: can be used to predict
- calculates phenology, LAI
• Used in Cox et al. (2000) – strong feedbacks. Are model parameters realistic? Is model structure appropriate?
.
Friedlingstein et al. (2006)
Cox et al. (2000)
Cox et al. (2000)
Friedlingstein et al. (2006)
Data assimilation methodologyData assimilation methodology
• Experiment 1: optimize 5 sites individually (separate)– Are parameters coherent across space
• Experiment 2: optimize 5 sites simultaneously (joint)– Allow soil carbon and leaf nitrogen to vary among sites (adds 8
parameters)– Do these parameters explain cross-site variability?
• This is a nonlinear, nonconvex problem requiring a global optimization algorithm
• Gradient-based techniques will fail!
• Stochastic Evolutionary Ranking Strategy (SRES) – genetic algorithm• Full parametric uncertainty with MCMC underway, no results yet
Convergence diagnosticsConvergence diagnostics
Separate optimization: parameters
Table 0-1: Results of the parameter optimization for each separate optimization at HW (Howland), HV (Harvard), WL (WLEF), MM (Morgan Monroe) and UM (UMBS). Well-constrained parameters are shown in bold, and are defined as values that are within 0.5% of each other in all three SRES runs and are not edge-hitting (upper and lower bounds and parameter units are shown in Table 4-2)
Parameter Published HV HW WL MM UM
CS 10.0 4.81 4.827 9.81 10.7 7.37
Q10H 2.0 1.47 2.394 3.28 1.92 2.65
opt 0.55 0.843 0.717 0.745 0.685 0.764
fac 0.8 1.00 0.320 0.858 1.00 1.00
c 0.30 0.247 0.129 0.240 0.285 0.239
w 0.13 0.00 0.0498 0.0534 0.00 0.00
nl0BL 0.036 0.149 0.250 0.130 0.100 0.112
nl0 NL 0.030 0.140 0.045 0.0430 0.0500 0.0516
BL 0.06 0.0535 0.117 0.0314 0.0498 0.0353
NL 0.06 0.0274 0.021 0.483 0.249 0.340
Q10VM 2.0 2.23 1.74 2.36 2.15 2.04
TlowBL -5.0 1.48 -40.0 -9.70 6.06 -1.83
TlowNL -15.0 3.55 3.41 -1.80 10.0 -20.0
TuppBL 33.0 50.0 50.0 50.0 50.0 50.0
TuppNL 28.0 11.66 50.0 50.0 10.0 49.1
Dc 0.09 0.0192 0.0266 0.0249 0.0371 0.0282
Rg 0.25 0.107 0.0500 0.122 0.0500 0.0500
Rdc 0.015 0.00822 0.0208 0.0183 0.0019 0.00866
Q10RD 2.0 1.49 1.61 2.00 1.100 1.33
Toff 0.0 11.6 0.80 9.30 14.3 11.4
off 0.90 0.678 0.850 0.672 0.172 0.517
p
14.5
18.2 13.4 16.5 20.0 38.7
-(Log L)
Well-constrained parameters:
leaf nitrogen
quantum efficiency
Tlower
phenology (Toff)
Poorly constrained parameters
autotrophic respiration
soil moisture parameters
Difference from published
Coherence across space
Model performance: Model performance: seasonal cycleseasonal cycle
Model performance: Model performance: Cross-site variabilityCross-site variability
• Site-specific parameters
• Soil carbon
• leaf nitrogen
• Joint assimilation captures cross-site variability of site-mean
• parameters
• climate
• forest structure
• Fails to capture interannual variability
Joint optimization parameters
Parameter Published optimized Parameter published optimized CS (HV) 10.0 0.00 BL 0.06 0.0446
Model performance:Model performance:Interannual variabilityInterannual variability
Model limitations
• Why do we fail to reproduce the interannual signal?– Problem with separate and joint optimizations– Limitations of the model structure (hydrology?)– Spatial mismatch: TRIFFID is designed for larger
gridscales– Limitations of the data (biases, uncertainty)
• Questions– Correct subset of model parameters?– Are observed fluxes coherent across space?