Optimal Temporal Frequency of NSSL Phased-Array Radar Observations for an Experimental Warn-on-Forecast System

A potential replacement candidate for the current U.S. National Weather Service operational weather-surveillance-Doppler radar (WSR-88D) network is the phased-array radar (PAR) system. Current WSR-88Ds take ~5 min to produce a full volumetric scan of atmospheric phenomena, whereas PAR technology can scan the same phenomena every ~1 min. How this increase in temporal frequency of radar observations might affect the National Severe Storms Laboratory’s (NSSL) Warn-on-Forecast system (WoFS), which is a frequently-cycled storm-scale ensemble data assimilation and forecast system for severe convective weather, is mostly unclear. Since radar data assimilation is critical for the WoFS, this study’s goal is to determine the optimal temporal frequency of PAR observations for WoFS hazardous weather prediction.

To conduct this study, we use the 31 May 2013 central and eastern Oklahoma tornado and flash flood event. NSSL’s National Weather Radar Testbed PAR started scanning this event more than an hour before the first (and strongest) tornado developed near El Reno, OK and scanned most of the tornadic supercell’s evolution. Several PAR data assimilation experiments are conducted using a model domain with 1-km horizontal grid spacing to produce analyses and very short-term forecasts of the El Reno supercell. For one set of experiments, PAR reflectivity and radial velocity observations are synchronously assimilated from 2145 UTC to 2300 UTC on 31 May 2013 every 1-, 3-, 5-, and 15-min using an ensemble square root-filter algorithm (EnSRF) with 36 ensemble members. For another experiment, 1-min PAR volumetric data are asynchronously assimilated with a 5-min cycling interval during the same data assimilation period. Finally, an experiment is conducted using a combination of 1- and 15-min cycling intervals to investigate the potential use of adaptive cycling intervals for the WoFS. For each experiment, ensemble forecasts are initialized every 15 min from 2200 UTC to 2300 UTC and ran until 0000 UTC on 1 June 2013. Forecasts of 2-km AMSL reflectivity and 2–5-km updraft helicity are subjectively evaluated and objectively verified using spatial and object-based techniques. Results from these experiments indicate that assimilating more frequent PAR observations can lead to more accurate analyses and forecasts of the El Reno supercell at longer lead times.