Sensitivity of Optimal Extension of Observation Networks to Model Transport

Prabir K. Patra, S. Maksyutov, D. Baker, P. Bousquet, L. Bruhwiler, Y-H. Chen, P. Ciais, A. S. Denning, S. Fan, I. Y. Fung, M. Gloor, K. R. Gurney, M. Heimann, K. Higuchi, J. John, R. M.Law, T. Maki, P. Peylin, M. Prather, B. Pak, P. J. Rayner, J. L. Sarmiento, S. Taguchi, T. Takahashi, C-W. Yuen

Tellus, 55B (2) 498-511, 2003.


Optimal extensions of the surface CO$_2$ observation network have been studied using 15 global transport models and a time-independent inverse model. The objective function, regional average CO$_2$ flux estimate uncertainty, is minimized for the TransCom-3 (level 1) set up. An ensemble model calculation shows that the regional average CO$_2$ flux uncertainties could be reduced to about 0.36, 0.32, 0.28 and 0.26 Gt C/yr per region, from about 0.53 Gt C/yr per region corresponding to the basic network, after adding of 5, 10, 15 and 20 optimally located stations, respectively. The optimal station locations are mostly found in continental South America and Africa. The distribution of flux uncertainty reduction efficiency per new station tends to become uniform with the network extension. We show that the multimodel approach to network design converges if a large enough extension is considered; about 20 stations in this inverse model framework. The gain in the flux uncertainty reduction for first few stations depends on the properties of atmospheric transport, and is nearly proportional to the model simulated local emission's signal in the surface layer. In addition, it is seen that the simulated spatial and temporal variability of CO$_2$ concentration has significant influence on distributing the optimal stations as well as determining the regional flux estimate uncertainty.