Limitations of Short-Term Thunderstorm Forecasts from Convection-Allowing Models with 3-km Horizontal Grid Spacing

Abstract Forecasts from a 1-km horizontal grid spacing Warn-on-Forecast System (WoFS-1km) prototype are compared to a 3-km WoFS (WoFS-3km) for 23 days when high-impact events occurred in 2022 and 2023. Composite reflectivity forecasts are verified using object-based verification. The probability of detection (POD) is significantly higher in WoFS-1km than in WoFS-3km for observed objects less than 400 km 2 in area, and false alarm ratio (FAR) is significantly lower in WoFS-1km than in WoFS-3km for forecast objects less than 1500 km 2 in area. This higher POD in WoFS-1km is consistent for all lead times while lower FAR is most pronounced in the first 90–120 min of the forecast. It is shown that improvements in WoFS-1km POD relative to WoFS-3km POD is dependent on object shape as well as size, with significant improvement in WoFS-1km forecasts of storms with a minor axis length less than approximately 24 km. WoFS-1km is found to produce improved POD for area bins representing 67% of the total thunderstorm objects in the dataset. A case study example of WoFS-1km and WoFS-3km forecasts demonstrates the ability of WoFS-1km to better resolve a violent tornado-producing supercell occurring on 10 December 2021. The supercell’s composite reflectivity object area was less than 400 km 2 for several hours, during which WoFS-1km produced improved predictions of the supercell track and maintenance. Once the storm surpassed this area threshold, WoFS-3km produced similar forecasts to WoFS-1km. The statistical findings and qualitative analysis demonstrate benefits of employing reduced horizontal grid spacing for short-term prediction of thunderstorms with convection-allowing models. Significance Statement This study concludes that an ensemble data assimilation and modeling system with 1-km versus 3-km horizontal grid spacing better resolves smaller thunderstorms (<400 km 2 ) while also reducing false alarms of both small and large thunderstorms (of up to 1500 km 2 ). Better resolution of smaller thunderstorms is directly applicable to a high-impact severe event that produced multiple significant tornadoes.