Data-model synthesis of ocean acidification mesocosm experiments – establishing a mesocosm modeling workbench
During SOPRAN phases I and II, a series of mesocosm experiments succeeded in generating unique datasets of ocean acidification (OA) effects on biogeochemistry and plankton ecology. Meanwhile, new biogeochemical models have either already been devised or are currently developed within SOPRAN. Having highly topical data together with latest model developments in SOPRAN, a systematic approach is required for an integrated OA data-model analysis. With our data-model syntheses we will specify model uncertainties and will investigate credibility and robustness of model results. We plan cross-validation experiments where a model's ability to explain multiple data sets from independent mesocosm experiments will be assessed. Likewise, we aim at resolving complex interdependencies between different types of observations. We also apply non-linear optimization procedures to estimate those biological- and chemical rates that cannot be measured directly.The major technical objective is to devise and establish a mesocosm modeling workbench that facilitates the use of selected models for simulations of diverse mesocosm experiments. The mesocosm-modeling workbench will fully incorporate experimental data as well as observed physical forcing, like radiation, temperature and salinity. The mesocosm modeling workbench is currently devised for "R" and is planned to include a Flexible Modelling Environment for Inverse Modelling, Sensitivity, Identifiability Monte-Carlo Analysis (FME, Soetaert and Petzold, 2010, Journal of Statistical Software, 33(3), 1-28). We also plan to have the model codes implemented from the repository "Framework for Aquatic Biogeochemical Models" (FABM) (Jorn Bruggeman and Carsten Bolding, 2011, http://www.meece.eu/documents/deliverables/WP2/D2.14.pdf).
Central questions that will be addressed with a mesocosm modeling workbench are:
1) What are the recurrent response signals to OA, given the environmental variations between experiments and within similar CO2-treatments?
2) Are these recurrent (pH-sensitive) response signals significant in model results, given uncertainties in model parameterizations?
Which data are
indispensable to constrain model solutions with respect to effects of OA?
Contact: Markus Schartau (firstname.lastname@example.org)