Model-based data assessment for terrestrial carbon processes: implications for sampling strategy in FACE experiments

The value of different types of data in the estimation of different carbon transfer parameters is investigated. A carbon accounting model is used with different observation operators to generate data. The effectiveness of the inversion is assessed by observing relative errors of estimators and likelihood ratios. It is demonstrated that for an observation operator that relative errors vary widely with the sample test problems. An effective strategy to test types of data is to test the effectiveness of corresponding observation operators on an ensemble of sample problems for which parameters are selected from the space of admissible parameters. The selection is carried out under the assumption that the test parameters themselves are random variables uniformly distributed over the space of admissible parameters.