This paper describes the lessons learned from the implementation of a regional ensemble data assimilation and forecast system during the intensive observing period of the Remote sensing of Electrification, Lightning, And Mesoscale/microscale Processes with Adaptive Ground Observations (RELAMPAGO) field campaign (central Argentina, November–December 2018). This system is based on the coupling of the Weather Research and Forecasting (WRF) model and the Local Ensemble Transform Kalman Filter (LETKF). It combines multiple data sources both global and locally available like high-resolution surface networks, AMDAR data from local aircraft flights, soundings, AIRS retrievals, high-resolution GOES-16 wind estimates, and local radar data. Hourly analyses with grid spacing of 10 km are generated along with warm-start 36-h ensemble-forecasts, which are initialized from the rapid refresh analyses every three hours. A preliminary evaluation shows that a forecast error reduction is achieved due to the assimilated observations. However, cold-start forecasts initialized from the Global Forecasting System Analysis slightly outperform the ones initialized from the regional assimilation system discussed in this paper. The system uses a multi-physics approach, focused on the use of different cumulus and planetary boundary layer schemes allowing us to conduct an evaluation of different model configurations over central Argentina. We found that the best combinations for forecasting surface variables differ from the best ones for forecasting precipitation, and that differences among the schemes tend to dominate the forecast ensemble spread for variables like precipitation. Lessons learned from this experimental system are part of the legacy of the RELAMPAGO field campaign for the development of advanced operational data assimilation systems in South America.

本文描述了在带自适应地面观测的电气化、闪电和中尺度/微尺度过程遥感（RELAMPAGO）野外活动（阿根廷中部， 2018 年 11 月至 12 月）。该系统基于天气研究和预测 (WRF) 模型和局部集合变换卡尔曼滤波器 (LETKF) 的耦合。它结合了全球和本地可用的多个数据源，如高分辨率地面网络、来自本地飞机飞行的 AMDAR 数据、探测、AIRS 反演、高分辨率 GOES-16 风估计和本地雷达数据。网格间距为 10 公里的每小时分析与热启动 36 小时集合预测一起生成，每三个小时从快速刷新分析中初始化。初步评估表明，由于同化观测，预报误差减少了。然而，从全球预报系统分析初始化的冷启动预测略优于本文讨论的区域同化系统初始化的预测。该系统使用多物理场方法，侧重于使用不同的积云和行星边界层方案，使我们能够对阿根廷中部的不同模型配置进行评估。我们发现预测地表变量的最佳组合与预测降水的最佳组合不同，并且方案之间的差异往往主导了降水等变量的预测集合分布。