Uncertainty in Carbon and Water Budgets Due to Processing and Partitioning Choices.

There has been significant work in the last decade to quantify the uncertainty in budgets of carbon and water from eddy covariance measurements of mass and energy fluxes. Much of this has concentrated on identifying the sources of random and systematic errors in the flux measurements and estimating their contributions to the total uncertainty. However, uncertainty due to these errors is usually estimated using data processing path containing a single implementation of the algorithms.

Several applications for the quality control and gap filling of fluxes and the partitioning of Net Ecosystem Exchange (NEE) are now available to the flux community. Examples include the open-source FluxNet, REddyProc, PyFluxPro and DINGO packages plus a commercial application, Tovi, from LI-COR Biosciences. While all of these packages share common approaches, details of the algorithm implementation differ and this introduces another source of uncertainty to carbon and water budgets that is due to the choice of processing package. The situation is analogous to the processing of high-frequency turbulence data where initial differences between multiple packages have now been largely resolved by careful intercomparison.

Previous work has compared results from 4 open-source packages and found that the range of NEE, Ecosystem Respiration (ER) and Gross Primary Productivity (GPP) values from the different processing packages was up to twice as large as the uncertainty due to the u*-threshold distribution estimated by the Change Point Detection approach. We extend this work to include a more complete assessment of random and systematic errors and add the results from the Tovi package. The results allow users to compare their current processing path with the alternatives and suggest careful intercomparison of the available packages is required to reduce this source of uncertainty to acceptable limits.