Xue, MingLabriola, Jonathan2020-04-302020-04-302020-05https://hdl.handle.net/11244/324158The explicit prediction of deep, moist convection is challenging because small model and initial condition errors rapidly grow and degrade forecast skill. Microphysics schemes employed by convection-allowing models represent a substantial source of model error because microphysical processes are poorly understood and simplifying assumptions must be made to make simulations and forecasts computationally practical. Although data assimilation systems decrease initial condition errors, analysis and forecast skill is sensitive to the experiment design. This dissertation evaluates data assimilation and ensemble forecast system performances at convection-allowing/convection-resolving resolutions, when forecast models employ different multi-moment microphysics parameterization schemes, and the data assimilation configurations are varied. We address the related issues through detailed case studies that provide insights on optimizing the configuration of convection-allowing model forecasts. First, high-resolution hail size forecasts are made for a severe hail event on 19 May 2013 using the Advanced Regional Prediction System (ARPS). Forecasts using the National Severe Storms Laboratory (NSSL) variable density rimed ice double-moment microphysics scheme (referred to as NSSL) exhibit more skill than those using the Milbrandt and Yau double-moment (MY2) or triple-moment (MY3) schemes when verified against radar-derived hail size estimates. Although all three schemes predict severe surface hail coverage with moderate to high skill, MY2 and MY3 forecasts overpredict the maximum hail size. The NSSL scheme uses the two variable density rimed ice categories to generate large, dense hail through the wet growth of graupel. Both the MY2 and MY3 schemes predict hail to be smaller above the 0 °C isotherm because the category is primarily composed of small frozen raindrops; in the melting layer the hail quickly grows because the rimed ice accretes excessive water. MY2 and MY3 forecasts predict the largest hail sizes to be smaller when the accretion water is eliminated beneath the 0 °C isotherm. To improve hailstorm forecast initial conditions, CAPS Ensemble Kalman filter (EnKF) analyses are generated for the 8 May 2017 Colorado severe hail event using either the MY2 or the NSSL scheme in the forecast model. The results of the EnKF analyses are evaluated. With each microphysics scheme two experiments are conducted where reflectivity (Z) observations update either (1) only the hydrometeor mixing ratio or (2) all hydrometeor fields. Experiments that update only hydrometeor mixing ratios can create ensemble error covariances that are unreliable which increases analysis error. Despite improving initial condition estimates, experiments that update all hydrometeor fields underestimate surface hail size, which suggests additional constraint from observations is needed during data assimilation. Correlation patterns between observation prior estimates (e.g., Z) and model state variables are evaluated to determine the impact of hail growth assumptions in the MY and NSSL schemes on the forecast error covariances between microphysical and thermodynamic variables. For the MY2 scheme, Z is negatively correlated with updraft intensity because strong updrafts produce abundant, small hail aloft. The NSSL scheme predicts storm updrafts to produce fewer but larger hailstones aloft, which causes Z and updraft intensity to be positively correlated. Hail production processes also alter the background error covariances for in-cloud air temperature and hydrometeor species. This study documents strong sensitivity of ensemble data assimilation results of hailstorms to the parameterization of microphysical processes, and the need to reduce microphysics parameterization uncertainties. To improve data assimilation configurations for potential operational implementation, EnKF data assimilation experiments based on the operational GSI system employed by the Center for Analysis and Prediction of Storms (CAPS) realtime Spring Forecast Experiments are performed, followed by 6-hour forecasts for a mesoscale convective system (MCS) event on 28-29 May 2017. Experiments are run to evaluate the sensitivity of forecast skill to the configurations of the data assimilation system. Configurations examined include the ensemble initialization and covariance inflation as well as radar observation data thinning, covariance localization radii, observation error settings, and data assimilation frequency. Spin-up ensemble forecast surface temperatures are most skilled when the initial ensemble mean is centered upon the most recent NAM analysis, causing forecasts to predict a strong MCS. Experiments that assimilate radar observations every 5 minutes are better at the placement of high Z values near observed storms but exhibit a substantial decrease in forecast skill initially because of widespread spurious convection. Ensembles that assimilate more observations with less thinning of data or use a larger horizontal covariance localization radius for radar data overpredict the coverage of high Z values due to enhanced spurious convection. Both parameters have modestly positive impacts on forecast skill during the first forecast hour that are quickly lost due to the growth of forecast error. Forecast skill is less sensitive to the ensemble spread inflation factors and observation errors tested during this study. These results provide guidance towards optimizing the GSI EnKF system configuration, for this study the data assimilation configuration employed by the 2019 CAPS Spring Forecast Experiment produces the most skilled forecasts while remaining viable for realtime use.Microphysical Parameterization SchemesEnsemble Forecast VerificationRadar Data AssimilationConvection-Allowing Model ForecastsNumerical weather forecastingWeather forecastingRadar meteorologyHail--Forecasting