Biome-BGC GLUE 1.3

Biome-BGC is working with a lots of ‘a priori’ unknown and hard to obtain model parameters. Therefore the parameterization is a critical step of using the model. Parameteres can be estimated using inverse calibration techniques based on measurement data, which means that the model is being calibrated. Measurement data have to be collected with respect to the model in order to compare them. Comparison is based on misfit measure (e.g. a sort of likelihood value), which is the function of the difference between observed and modelled data. It is based on Bayesian calibration with Monte Carlo search. Each parameter is varied randomly within their ‘a priori’ range and the model is run several times using variable model parameters. Then the ‘a priori’ distribution is updated with model information (distribution of the likelihood function) to define ‘a posteriori’ density function. From the maximum of the ‘a posteriori’ density function optimal parameter values can be calculated and approved.

GLUE requires a prior execution of a Biome-BGC Monte Carlo Experiment (MCE), that performs an independent parameter variation within ‘a priori’ parameter ranges. Parameters, range of parameter values, output variables and number of randomized repetition has to be set in Biome-BGC MCE workflow, that runs off-line because of the time consuming nature of MCE jobs (usually it takes several days on, for example the EDGeS@home desktop grid). Then one or several GLUE analysis can be launched based on the results of Biome-BGC MCE completed before.

The Biome-BGC Projects Database & Management System was developed to easy prepare, manage and share all of the above mentioned datasets (files), investigations and provide interaction with Taverna workflows. Learn more about BBGCDB at http://ecos.okologia.mta.hu/bbgcdb/.