Complementarity of flux- and biometric-based data to constrain parameters in a terrestrial carbon model

Abstract

To improve models for accurate projections, data assimilation, an emerging statistical approach to combine models with data, have recently been developed to probe initial conditions, parameters, data content, response functions and model uncertainties. Quantifying how many information contents are contained in different data streams is essential to predict future states of ecosystems and the climate. This study uses a data assimilation approach to examine the information contents contained in flux- and biometric-based data to constrain parameters in a terrestrial carbon (C) model, which includes canopy photosynthesis and vegetation–soil C transfer submodels. Three assimilation experiments were constructed with either net ecosystem exchange (NEE) data only or biometric data only [including foliage and woody biomass, litterfall, soil organic C (SOC) and soil respiration], or both NEE and biometric data to constrain model parameters by a probabilistic inversion application. The results showed that NEE data mainly constrained parameters associated with gross primary production (GPP) and ecosystem respiration (RE) but were almost invalid for C transfer coefficients, while biometric data were more effective in constraining C transfer coefficients than other parameters. NEE and biometric data constrained about 26% (6) and 30% (7) of a total of 23 parameters, respectively, but their combined application constrained about 61% (14) of all parameters. The complementarity of NEE and biometric data was obvious in constraining most of parameters. The poor constraint by only NEE or biometric data was probably attributable to either the lack of long-term C dynamic data or errors from measurements. Overall, our results suggest that flux- and biometric-based data, containing different processes in ecosystem C dynamics, have different capacities to constrain parameters related to photosynthesis and C transfer coefficients, respectively. Multiple data sources could also reduce uncertainties in parameter estimation if these data sources contain complementary information.