The Impact of Radar Data Assimilation on Short-Term and Next-Day Thunderstorm Forecasts in the 2016 Community Leveraged Unified Ensemble (CLUE)

The impact of radar assimilation in the 2016 Community Leveraged Unified Ensemble (CLUE) is examined. Composite reflectivity forecasts are compared between two 10-member CLUE sub-ensembles that are identical except that one assimilates radar data using a three-dimensional variational technique with cloud analysis (RAD) and the other does not (NORAD). Forecasts are compared using neighborhood and object-based verification metrics. It is found that radar assimilation typically improves the predicted location of composite reflectivity objects for 3–6 hours. However, large differences in thunderstorm coverage persist through 36 hour forecasts, with a low bias in RAD and a high bias in NORAD. These biases are consistent across cases and degrade the quality of forecasts from both ensembles. The contrasting biases in next-day RAD and NORAD composite reflectivity forecasts are influenced by variations in the extent and amplitude of predicted convective available potential energy (CAPE), with NORAD CAPE forecasts consistently predicting upwards of 10,000 more gridpoints (>90,000 km²) with significant CAPE (>1000 J Kg⁻¹) than RAD forecasts. Variation in next-day CAPE forecasts is attributable to the impacts of data assimilation on short-term thunderstorm evolution, in particular large differences between convective cold pool representations in RAD and NORAD.