Automatic Optimization of Weather Research and Forecasting (WRF) Model Parameters

Qingyun Duan
Beijing Normal University
Noon April 1 in Room 2155

Weather Research and Forecasting (WRF) model is a community numerical weather prediction (NWP) model widely used by many researchers and practitioners in the world.  There are three ways to improve the performance of the WRF model.  The first way is to improve the parameterization schemes for different physical processes, including radiation, cumulous cloud, microphysics, surface and boundary layers and land surface.  The second way is to obtain the most accurate and reliable boundary and initial conditions with new observational technologies and data assimilation techniques.  The third way is to enhance the estimation of parameters present in the WRF model.  This study concerns with the estimation of the parametric uncertainty of the WRF model.  We identified a list of over 20 parameters which are thought to be influential in the performance of the WRF model in forecasting precipitation and temperature.  A global sensitivity analysis is first used to screen out the most sensitive parameters.  Then we explore the use of surrogate modeling approach to identify the optimal parameters for the screened parameters.  The numerical case study is set for the North China domain surrounding the Beijing city area.  The WRF model will be run at 9km spatial resolution.  The forecast lead time is set to 5 days.  We have selected 9 different storms over the summer months (June/July/August) over the 2008-2010 period to study the effect of parametric uncertainty on predictive skill of precipitation and surface temperature forecasts.  This presentation will show the preliminary results from this study.