Academic Journals Database
Disseminating quality controlled scientific knowledge

Addressing the impact of environmental uncertainty in plankton model calibration with a dedicated software system: the Marine Model Optimization Testbed (MarMOT)

ADD TO MY LIST
 
Author(s): J. C. P. Hemmings | P. G. Challenor

Journal: Geoscientific Model Development Discussions
ISSN 1991-9611

Volume: 4;
Issue: 3;
Start page: 1941;
Date: 2011;
VIEW PDF   PDF DOWNLOAD PDF   Download PDF Original page

ABSTRACT
A wide variety of different marine plankton system models have been coupled with ocean circulation models, with the aim of understanding and predicting aspects of environmental change. However, an ability to make reliable inferences about real-world processes from the model behaviour demands a quantitative understanding of model error that remains elusive. Assessment of coupled model output is inhibited by relatively limited observing system coverage of biogeochemical components. Any direct assessment of the plankton model is further inhibited by uncertainty in the physical state. Furthermore, comparative evaluation of plankton models on the basis of their design is inhibited by the sensitivity of their dynamics to many adjustable parameters. The Marine Model Optimization Testbed is a new software tool designed for rigorous analysis of plankton models in a multi-site 1-D framework, in particular to address uncertainty issues in model assessment. A flexible user interface ensures its suitability to more general inter-comparison, sensitivity and uncertainty analyses, including model comparison at the level of individual processes, and to state estimation for specific locations. The principal features of MarMOT are described and its application to model calibration is demonstrated by way of a set of twin experiments, in which synthetic observations are assimilated in an attempt to recover the true parameter values of a known system. The experimental aim is to investigate the effect of different misfit weighting schemes on parameter recovery in the presence of error in the plankton model's environmental input data. Simulated errors are derived from statistical characterizations of the mixed layer depth, the horizontal flux divergences of the biogeochemical tracers and the initial state. Plausible patterns of uncertainty in these data are shown to produce strong temporal and spatial variability in the expected simulation error over an annual cycle, indicating differences in the significance attributable to model-data misfits at different data points. An inverse scheme using ensemble-based estimates of the simulation error variance to allow for this environment error performs well compared with weighting schemes used in previous plankton model calibration studies. The efficacy of the new scheme in real-world applications will depend on the quality of statistical characterizations of the input data. Practical approaches towards developing reliable characterizations are discussed.
Affiliate Program      Why do you need a reservation system?