A prototype Carbon Cycle Data Assimilation System CCDAS
A prototype Carbon Cycle Data Assimilation System (CCDAS) Inferring interannual variations of vegetationatmosphere CO 2 fluxes Marko Scholze 1, Peter Rayner 2, Wolfgang Knorr#3, Thomas Kaminski 4, Ralf Giering 4 #presenting 1 2 3 4 Fast. Opt
Carbon Cycle Data Assimilation using automatic differentiation Parameters: 58 Biosphere Model: BETHY Fluxes: 800, 000 Atmospheric Transport Model: TM 2 Station Conc. 10, 000 Misfit to Observations 1. Parameter Optimisation: Forward: Parameters –> Misfit Adjoint or Tangent linear: ∂ Misfit / ∂ Parameters 2. Parameter Uncertainties: Hessian: ∂2 Misfit / ∂ Parameters 2 Error covariance=Inverse of Hessian 3. Uncertainty of Diagnostics: Adjoint or Tangent linear Misfit 1 Fast. Opt
CCDAS Setup Assimilated veg. index Satellite CCDAS Step 1 full BETHY Background CO 2 fluxes* Prescribed Phenology Hydrology Assimilated CO 2 + Uncert. CCDAS Step 2 IMBETHY+TM 2 only Photosynthesis, Energy&Carbon Balance Calibrated Params + Uncert. Diagnostics + Uncert. * * ocean: Takahashi et al. (1999), Le. Quere et al. (2000); emissions: Marland et al. (2001), Andres et al. (1996); land use: Houghton et al. (1990)0 Fast. Opt
BETHY (Biosphere Energy-Transfer-Hydrology Scheme) lat, lon = 2 deg • GPP: t=1 h C 3 photosynthesis – Farquhar et al. (1980) C 4 photosynthesis – Collatz et al. (1992) stomata – Knorr (1997) t=1 h • Raut: maintenance respiration = f(Nleaf, T) – Farquhar, Ryan (1991) growth respiration ~ NPP – Ryan (1991) t=1 day • Rhet: fast/slow pool resp. = wk Q 10 T/10 C fast/slow / t fast/slow t slow –> infin. b<1: source average NPP = b average Rhet (at each grid point) b>1: sink Fast. Opt
Concentrations Fast. Opt
Parameters examples: relative error reduction: Fast. Opt
global fluxes Processes 1 Carbon sink: GPP slightly exceeds respiration Carbon source anomaly: drop in GPP exceeds drop in resp El Niño events Carbon sink anomaly: stronger decr. in resp. than GPP Pinatubo eruption La Niña Fast. Opt
normalized CO 2 flux and ENSO Processes 2 lag correlation (low-pass filtered) ENSO and terr. biosph. CO 2: correlation seems strong correlation between Niño-3 SST anomaly and net CO 2 flux shows maximum at 4 months lag, for both El Niño and La Niña states 4 -month lagged: Pinatubo eruption: shows up as largest deviation in the low-pass filtered curve Fast. Opt
Processes 3 El Niño (>+1 s) net CO 2 flux to atm. g. C / (m 2 month) lagged correlation at 99% significance -0. 8 -0. 4 0. 8 Fast. Opt
Carbon Balance Euroflux (1 -26) and other eddy covariance sites* net carbon flux 1980 -2000 g. C / (m 2 year) *from Valentini et al. (2000) and others latitude N Fast. Opt
Conclusions • CCDAS with 58 parameters can already fit 20 years of CO 2 concentration data • Significant reduction of uncertainty for ~13 parameters, some important covariances • terr. biosphere response to climate fluctuations dominated by ENSO and Pinatubo • Can be explained by small perturbations of 3 large fluxes (GPP, Raut, Rhet) Fast. Opt
Outlook • explore more parameter configurations • include fire as a process with uncertainties • include more constraints (isotopes, eddy fluxes) • extend approach to ocean carbon cycle Fast. Opt
- Slides: 12