The Impacts of Using Mixed Physics in the Community Leveraged Unified Ensemble Weather and Forecasting

A well-known problem in high-resolution ensembles has been a lack of sufficient spread among members. Modelers often have used mixed physics to increase spread, but this can introduce problems including computational expense, clustering of members, and members that are not all equally skillful. Thus, a detailed examination of the impacts of using mixed physics is important. The present study uses two years of Community Leveraged Unified Ensemble (CLUE) output to isolate the impact of mixed physics in 36-h forecasts made using a convection-permitting ensemble with 3-km horizontal grid spacing. One 10-member subset of the CLUE used only perturbed initial conditions (ICs) and lateral boundary conditions (LBCs) while another 10-member ensemble used the same mixed ICs and LBCs but also introduced mixed physics. The cases examined occurred during NOAA’s Hazardous Weather Testbed Spring Forecast Experiments in 2016 and 2017. Traditional gridpoint metrics applied to each member and the ensemble as a whole, along with object-based verification statistics for all members, were computed for composite reflectivity and 1- and 3-h accumulated precipitation using the Model Evaluation Tools (MET) software package. It is found that the mixed physics increases variability substantially among the ensemble members, more so for reflectivity than precipitation, such that the envelope of members is more likely to encompass the observations. However, the increased variability is mostly due to the introduction of both substantial high biases in members using one microphysical scheme, and low biases in other schemes. Overall ensemble skill is not substantially different from the ensemble using a single physics package. 2019 OAR (Oceanic and Atmospheric Research) ESRL (Earth System Research Laboratory) GSL (Global Systems Laboratory) Submitted https://doi.org/10.1175/WAF-D-18-0197.1 Other 1954

高分辨率集合预报领域长期存在一项公认难题：集合成员间离散度（spread）不足。预报研究者常通过混合物理参数化方案（mixed physics）提升离散度，但该方法会引发诸多问题，包括计算成本高企、成员聚类以及成员预报技巧参差不齐。因此，深入剖析混合物理参数化方案的影响具有重要意义。本研究依托两年的社区统一耦合集合预报（Community Leveraged Unified Ensemble, CLUE）输出数据，针对水平网格间距为3公里的对流允许型集合预报（convection-permitting ensemble）的36小时预报，剥离出混合物理参数化方案的单独影响。其中一组10成员子集仅采用扰动初始条件（initial conditions, ICs）与侧边界条件（lateral boundary conditions, LBCs）；另一组10成员集合则采用相同的扰动初始与侧边界条件，同时引入混合物理参数化方案。本研究涉及的个例均取自2016年与2017年美国国家海洋和大气管理局（National Oceanic and Atmospheric Administration, NOAA）危险天气试验台春季预报试验项目。研究采用模式评估工具（Model Evaluation Tools, MET）软件包，针对组合反射率、1小时及3小时累积降水量，计算了适用于各成员及整体集合的传统格点检验指标（gridpoint metrics），同时也计算了所有成员的基于对象的验证（object-based verification）统计量。结果表明，混合物理参数化方案可显著提升集合成员间的变异性（variability），且该提升在反射率场上的表现较降水量场更为突出，使得集合成员值域更大概率覆盖观测值。不过，这种变异性的提升主要源于两类偏差：采用某类微物理方案（microphysical scheme）的成员普遍存在显著正偏差，而采用其他微物理方案的成员则普遍存在负偏差。整体而言，采用混合物理方案的集合，其整体预报技巧与采用单一物理参数化方案包（physics package）的集合并无显著差异。本研究由美国海洋与大气研究（Oceanic and Atmospheric Research, OAR）地球系统研究实验室（Earth System Research Laboratory, ESRL）全球系统实验室（Global Systems Laboratory, GSL）于2019年提交，DOI: 10.1175/WAF-D-18-0197.1，其他相关信息：1954。