CEE Seminar: Tuning Numerical Weather Prediction Models with Uncertainty Quantification Methods
Professor & Chief Scientist
Bejing Normal University
College of Global Change & Earth System Science
Abstract: Hydrologists have been conducting automatic model calibration since the invention of the first digital watershed model. This is not the case for atmospheric models. One reason is that model calibration is regarded by many as an art instead of a science. But a real, practical reason is that calibrating an atmospheric model automatically is simply too challenging. A typical atmospheric model contains a large number of model parameters and simulates many meterological variables. More importantly, it is extremely expensive to run an atmospheric model. It is therefore practically impossible to calibrate an atmospheric model automatically using conventional optimization methods for hydrological model calibration. With recent advances in uncertainty quantification methods for large complex system models, automatic calibration of atmospheric models become possible. In this talk, I demonstrate how UQ methods can be used to calibrate the WRF model automatically. A brief introduction of some key techniques including design of experiments, global sensitivity analysis and surrogate modeling-based optimization will be given first. A case study in which the WRF model is optimized for summer precipitation forecasting in the greater Beijing area will then be presented.
Bio: Qingyun Duan is the chief scientist and a professor at Beijing Normal University (BNU), College of Global Change and Earth System Science. He is a recipient of the Chinese “One-Thousand Talents Program” and a faculty member in the Department of Geographical Science at BNU. He is a fellow of both the American Geophysical Union and American Meteorological Society. He earned his doctorate in hydrology from University of Arizona in 1991 and bachelor’s degree from Wuhan University in 1982. He spent 10 years at NOAA and Lawrence Livermore National Laboratory. His research interests include hydrology and water resources management, hydrological model development and calibration, hydrometeorological ensemble forecasting, soil/vegetation/atmosphere interactions, climate change impacts on water resources, and uncertainty quantification for large complex-system models. He is the developer of several operational land surface hydrological models used in the U.S. National Weather Service. He is author of the wildly used model parameter optimization method, SCE-UA. He led the recent development of uncertainty quantification software platform for large complex-system models – Uncertainty Quantification Python Laboratory (UQ-PyL), and the BNU Hydrological Ensemble Prediction System (BNU-HEPS). Duan has published more than 150 SCI-indexed papers, with more than 8,700 citations and an H-index of 41, according to Institute of Scientific Information (ISI). He is an editor of the Bulletin of American Meteorological Society, Environmental Modelling & Software, and International Journal of Water, and he is a former editor of Water Resources Research. He also is chief editor of The handbook of Hydrometeorological Ensemble Forecasting, published by Springer U.K.