Applying a dual optimization method to quantify carbon fluxes: recent progress in carbon flux inversion

The widely performed Bayesian synthesis inversion method (BSIM) utilizes prior carbon flux and atmospheric carbon dioxide observations to optimize the unknown flux. The prior flux is usually computed from ecological models with large biases. The BSIM is useful in solving the problem of insufficient data, but it will increase the inaccuracies in the estimates caused by the biased prior flux. In this study, we propose a dual optimization method (DOM) to introduce a set of scaling factors as new state variables to correct for the prior flux according to information on plant functional types. The DOM estimates the scaling factors and carbon flux simultaneously by minimizing the cost function. The statistical properties of the DOM, which compare favorably with the BSIM, are provided in this article. We tested the DOM through simulation experiments which represent a true ecosystem. The results, according to the root mean squared error, show that the DOM has a higher accuracy than the BSIM in flux estimates.