How is the model calibrated?
Finding the correct parameter values in such a complex model is a daunting task for a researcher. So, we have developed a computer-based algorithm which uses a custom Model Performance Index (MPI). Basically, the computer does a number of “eyeballing” runs comparing the available data and boundary or frequency conditions (e.g. maximum values or seasonal cycles) with the model’s output for different sets of parameters. The parameter values are not selected randomly (as in a Monte Carlo technique) since the chance of finding the “good” points in such a big parameter space would be too small. Rather, the algorithm improves its search strategy by analyzing the results of each change and running statistical analyses at every step. This allows us to find acceptable parameter combinations in a reasonable time, much better than any human operator or any random algorithm could do. Additionally, the data gathered to improve the search strategy are of great value for improving the knowledge of th
