Abstract ID: 601

A procedure for estimation of data induced uncertainty in model predictions in the LBA domain.

By using one year of data collected at the Santarem Km 67 location available before 2006 and the corrected data for the same location made available for the LBA model intercomparison project we investigate a Bayesian procedure for quantification of the model prediction uncertainty associated with the errors in the data. The procedure follows a series of ensemble simulations and a hierarchical approach with Markov Chain Monte Carlo based methods to approximate the parameter distributions associated with different levels of data uncertainty. Several possible levels of data errors are analyzed under different distribution assumptions. At the same time we use several general sensitivity analysis methods to obtain a different kind of uncertainty estimates and compare the overall behavior of the model when driven by forcing data with reduced (by an unknown quantity) uncertainty. We apply the procedures to the Simple Biosphere Model version 3 (SiB3). The procedure presented is general and can be used with any other model and model type.

Session:  Carbon - The LBA Model-Intercomparison Project.

Presentation Type:  Poster