Estimating Biomass in a Combined SiB and CASA Model Using Data Assimilation

Because direct measurement is not possible, one depends on land surface models to estimate regional carbon fluxes. Weather and climate dominate inter-annual variability in carbon flux, but whether the land surface model produces a long-term carbon source or sink depends sensitively on the assumed initial amount of biomass. Most models assume the initial biomass is in equilibrium with respect to climate, where biomass input from photosynthesis balances biomass losses due to microbial decay. However, biomass is typically not in climate equilibrium because of agriculture, timber harvest, biomass burning, and other external processes, resulting in unacceptable uncertainty in the time-mean simulated fluxes. To improve modeled carbon fluxes, we used data assimilation to estimate initial pool sizes of biomass in the SibCasa model from observed surface fluxes of latent heat, sensible heat, and carbon dioxide from the global flux tower network. SibCasa combines the Simple Biosphere (SiB) biophysical model with the Carnegie-Ames-Stanford Approach (CASA) biogeochemical model to produce a hybrid model capable of estimating net carbon fluxes at a 10-minute time resolution. We use the Maximum Likelihood Ensemble Filter (MLEF) ensemble-based data assimilation technique developed at Colorado State University to calculate optimal estimates of initial pool sizes (and associated uncertainties). The uncertainties are defined in terms of analysis and forecast error covariance matrices, calculated in an ensemble-spanned subspace. We present the SibCasa model and the MLEF technique, and compare estimated initial biomass pool sizes to available observations at various flux tower sites.