A Prototype Convective-Scale Weather Analysis and Forecast System for the Warn-on-Forecast Research Project

A real-time analysis and forecasting system for convective-scale weather has been developed as one of the two prototypes for the Warn-on-Forecast (WoF) project led by the National Severe Storms Laboratory (NSSL). This system is based on a three-dimensional variational (3DVAR) data assimilation system. Another prototype system is designed for an ensemble-based approach, which is tested separately. The goal of the WoF project is to provide physically-consistent gridded analyses and numerical guidance to help forecasters make real-time warning decisions in a timely manner. The system of concern consists of three components: 1) a numerical forecasting model, 2) a convective-scale data analysis system and 3) a real-time control system. For the preliminary version of this system, the forecast component is the latest version of the ARW core within the WRF model. The data analysis system uses a three-dimensional variational (3DVAR) method that is parallelized and optimized specifically for assimilating high density observations, including both radar and satellite data, as well as other conventional observations. A flexible real-time control system glues variants of workflows for analysis cycles, forecast cycles and post-processing together. The control system is written in Python language. It provides flexibility by separating the machine-dependent job management with the workflow configurations for analysis and forecast so that the whole system can be migrated to any HPC platform with only minor changes to its specific job scheduler. Furthermore, the run-time parameters for analysis and forecast are provided through template files in Fortran namelist format, which is configurable by meteorological scientists without concern about the workload management on a specific platform.

This system has been tested during the 2016 Hazardous Weather Testbed (HWT) Spring Experiment period on a Cray XE6m computer at NSSL. The analysis and forecasting system uses 36 compute nodes (1,152 cores) on the Cray. The system is designed to be weather adaptive,  automatically detecting severe local hazardous weather events and positioning high-resolution analysis and forecast grids optimally on that basis. Alternatively, the timing and positioning of high-resolution forecast grids can be user specified.  The 3DVAR analysis system is optimized specifically for high density observations in a parallel computing environment through MPI so that it can incorporate available mesoscale forecasts, radar observations, satellite data, and traditional observations to perform a rapid analysis. Two types of products are generated: a series of analyses and a series of forecasts. During the 2016 Spring Experiment, the high-resolution analysis and forecast domain had 1.3 km grid spacing and 480 X 480 grid points, on which analyses were performed every 5 minutes and 3-h forecasts were launched every 30 mins.  The detailed implementation of the system and its performance during the 2016 Spring Experiment period will be reported in the conference.