Classification of Hydrological Parameter Sensitivity and Evaluation of Parameter Transferability across 431 US MOPEX Basins
Representing terrestrial processes and their exchanges with the atmosphere, land surface models are important components of Earth system models used to predict climate variations and change. Most land surface models include numerous sub-models, each representing key processes with mathematical equations and model parameters. Optimizing the parameter values may improve model skill in capturing the observed behaviors. With the high-dimensional parameter space, systematic calibration using observations is mission impossible because of high computational demand and limited data to constrain the parameter values. However, this problem can be made more tractable by classifying the complex system into a few relatively homogeneous regions, each responding to changes in model parameters in similar ways. Scientists at Pacific Northwest National Laboratory have developed such a novel classification scheme and demonstrated that the 431 MOPEX basins can be grouped into only 6 unique classes, with the potential that optimal parameters are transferable across basins of the same classes.
Representing terrestrial processes and their exchanges with the atmosphere, land surface models are important components of Earth system models used to predict climate variations and change. Most land surface models include numerous sub-models, each representing key processes with mathematical equations and model parameters. Optimizing the parameter values may improve model skill in capturing the observed behaviors. With the high-dimensional parameter space, systematic calibration using observations is mission impossible because of high computational demand and limited data to constrain the parameter values. However, this problem can be made more tractable by classifying the complex system into a few relatively homogeneous regions, each responding to changes in model parameters in similar ways. Scientists at Pacific Northwest National Laboratory have developed such a novel classification scheme and demonstrated that the 431 MOPEX basins can be grouped into only 6 unique classes, with the potential that optimal parameters are transferable across basins of the same classes.
A research team led by Department of Energy scientists at Pacific Northwest National Laboratory has developed and applied an uncertainty quantification (UQ) framework consisting of importance sampling, exploratory data analyses, and HPC-enabled numerical simulations. They conducted a set of model calibration experiments to develop a classification of river basins, such as optimal model parameters, that are transferable between basins of the same classes. This provides a viable approach for parameter calibration to be possible at a continental scale using data from limited field sites. To develop the classification, they conducted parameter significance tests and assigned sensitivity scores to each parameter/factor of the Community Land Model. This work forms a basis to classify river basins into groups with similar parameter significance patterns. The analysis was applied to 431 relatively pristine basins known as MOPEX basins in the United States, resulting in 6 unique classes. With this classification, model calibration performed using data from representative basins of each class may be transferred to other basins of the same class. This refinement represents a significant methodological advancement in UQ, potentially allowing global parameter calibration to be performed in spite of the numerous data-sparse regions worldwide.