A Framework for Convective-Scale Observing System Simulation Experiments Using Ensembles Weather and Forecasting

Convection-allowing models are used to predict the evolution of severe weather phenomena in the atmosphere. These models are sensitive to errors in the numerical environment used to initialize their forecast integration. Data assimilation methods can help overcome these errors but are often limited by sparse observational coverage. Numerous novel observation platforms are currently available that promise to close these coverage gaps, but they have not yet been widely assimilated into weather models. Observing system simulation experiments are often used to determine how best to assimilate these novel observations in space and time by using synthetic observations from a high-resolution “nature run.” However, a single deterministic nature run does not provide a measure of the flow’s intrinsic predictability within the chosen modeling system. We present a framework to provide insight into this predictability by using an ensemble of nature runs. The ensemble provides a range of likely outcomes for storm evolution and an upper limit against which forecasts are verified. We applied this framework to two events from Oklahoma in 2023: a quasi-linear convective system in February and a supercell case in April. The intrinsic predictability of the nature-run ensemble was used to calibrate each case and to verify the forecast ensembles. Results showed that the intrinsic predictability suggested by the nature-run ensemble was both field and case dependent. This framework may help guide future studies by giving researchers a better understanding of what is possible for a chosen flow problem and modeling system and how best to include and arrange novel observations. Grant no. NA21OAR4320204 Grant no. NA23OAR4590406