P10 A statistical approach to generating perturbed physics ensemble forecasts using the WRF model

Shen, Chenwei, Qy Duan, College of Land Surface System Science & Sustainability Faculty of Geographical Science Beijing Normal University

The WRF model can be regarded as millions of different models contained in one framework. For WRF3.7.1, the potential number of different parameterization scheme combinations is over 2 million. How to select the suitable parameterization scheme combinations for a particular location? This study reports a way that can automatically select the most suitable scheme combinations to reduce the forecast uncertainty. This study used Hypercube sampling, Tukey Honest Significance Difference (Tukey-HSD) and variance analysis as the screening methods to identify a set of parameterization scheme combinations (perturbed physics ensemble) for short-term summer precipitation forecast in the Greater Beijing area. The screening is based on the performance criteria, Equitable Threat Score and BIAS. After several rounds of screening, 23 screened scheme combinations, together with the control one, formed a 24 members perturbed physics ensemble. The generated ensemble was then validated with several skill metrics, Ranked Probability Score, Brier Score and Reliability Operating Characteristic curves. It was found that performance of perturbed physics ensemble is better than the deterministic forecast used currently by Beijing Institute of Urban Meteorology for the all rainfall intensities.