Verification of 10-Meter Wind Forecasts from NSSL-WRF in Predicting Severe Wind-Producing MCSs

Predicting convective winds associated with mesoscale convective systems (MCSs) remains a major challenge for operational severe weather forecasters. To assess the performance of the NSSL-WRF in forecasting severe-wind-producing MCSs between 2012 and 2014, a climatology of these MCSs was developed. Severe-wind-producing MCSs were first manually identified by finding swaths of severe wind reports caused by MCSs through inspection of radar reflectivity structure to ensure proper convective mode. To objectively identify severe-wind-producing MCSs using an object-based approach, storm reports were filtered based on nearby radar reflectivity, wind speed, and whether the report was a measured or estimated gust. Reports were converted to spatial probabilities via Gaussian smoothing so that objects could be identified. Objects were identified by testing various minimum intensity and area thresholds to determine which thresholds most accurately matched the manually identified severe-wind-producing MCSs. Objects identified based on radar-filtered storm reports most accurately matched manually identified severe-wind-producing MCSs. This allowed for development of an object-based climatology of severe-wind-producing MCSs. This climatology shows a maximum of severe-wind-producing MCSs near the Ohio River Valley with another relative maximum on the Georgia-Alabama border. All identified severe-wind-producing MCSs occurred east of the Rocky Mountains. Severe-wind-producing MCSs occurred most often in June and least often in November.

Daily maximum 10-m wind forecasts for the 24 hours beginning at 12Z (i.e., f12-f36) were generated from 0000 UTC NSSL-WRF hourly maximum 10-m wind fields. The same smoothing applied to storm reports was also applied to various forecast daily maximum 10-m wind thresholds between 15 kt and 58 kt. The same intensity and size thresholds were applied to both forecast and observation fields in identifying objects. Forecasts were then verified both on a grid-point-by-grid-point basis and on an object-matching basis. Object-matching utilizes a fuzzy logic algorithm to match forecast and observed objects. Across all wind speed thresholds, the 10-m wind field has a critical success index around 0.15 when using object-based verification, which is an improvement over grid-point verification. Lower wind speed thresholds over-forecast severe-wind-producing MCSs and approach a probability of detection (POD) near 100% for very low wind speed thresholds. As wind speed thresholds increase, the POD decreases sharply without much improvement in the false alarm ratio (FAR). For very high wind speed thresholds, very few events are forecast, so both POD and FAR are low. This suggests that 10-m winds are not a great proxy for identifying severe-wind-producing MCSs, though it does provide a baseline for existing diagnostics in forecasting severe-wind-producing MCSs. Modeled 10-m winds will be filtered based on simulated reflectivity in a similar manner to the filtering method applied to observations. Verification results of the filtered 10-m wind forecasts will also be presented.