Optimization of Model Parameters and Experimental Designs for a Global Marine Biogeochemical Model

Monday, 15 December 2014
Joscha Reimer, University of Kiel, Kiel, Germany
In geophysics, models are a fundamental key for understanding and predicting. To bring these models closer to reality, insufficiently known model parameters are optimized using measurement data. Usually, these models are computationally very expensive. For this reason, sophisticated parallel algorithms need to be used for their evaluation and optimization.

The measurements required for the optimization of the model parameters are often time-consuming or costly. For this reason, it is desirable that the information content of the obtained measurement results is maximal.

Several conditions under which measurements are carried out are controllable. These conditions are also known as experimental design. This can be, e.g., the point in time, the location or the method of the measurements. These experimental designs can be optimized so that the information content is maximized. Thus, the number of measurements necessary for a certain accuracy of the model parameters and accordingly of the model itself can be significantly reduced.

We optimized the parameters of a global model for phosphate and dissolved organic phosphorus in the ocean using over four million measurement data. The uncertainty in the optimal model parameters and the model itself resulting from uncertainties in the measurements was estimated. Furthermore, experimental designs for additional measurements, consisting of the time and location of the measurements and the tracer to be measured, were optimized. We would like to present the obtained results together with the applied techniques for parameter estimation, uncertainty estimation and optimization of experimental designs.