Data Assimilation Meetings at Reading

Date Meeting type Speakers
18 July 2012 DARC seminar John Hemmings (NOC, Southampton)
Constraining Plankton Dynamics for Earth System Models using Ocean Data
Marine plankton are fundamental to the ocean carbon cycle, contributing to ocean uptake of CO2 by exporting carbon from the sunlit layer to the ocean interior and sequestering carbon in the deep ocean. A need to understand and predict the role of these organisms in environmental change has led to the development of ocean biogeochemical circulation models with a wide range of different plankton system representations. There is high diversity in these plankton sub-models, both in terms of their structure and process parameterizations, and they are inherently non-linear with many adjustable parameters. Sensitivity to their physical forcing combined with a lack of observational coverage of many key system components makes it difficult to assess candidate models objectively and presents particular challenges for data assimilation.
Inverse methods have been used widely to address parameter uncertainty, typically exploiting data from a small number of ocean time-series sites. However, results are inevitably compromised by error in the local forcing data and neglect of horizontal processes or their inadequate representation. Temporal and spatial variability in the expected simulation error attributable to uncertainty in these environmental drivers has implications for the relative significance of model-data differences at different observation points. This problem is investigated with the Marine Model Optimization Testbed (MarMOT), an innovative software tool designed for the comprehensive analysis of candidate plankton models. A new inverse approach that includes an explicit treatment of quantified uncertainty in a model's environmental input data has been developed. Twin experiments with synthetic data show the method to have strong potential when compared with existing schemes. In real-world applications, its efficacy will depend on the quality of the input characterizations. A major effort in uncertainty quantification will be needed and some of the practical issues are discussed.

Page navigation