Improving Winter Storm Forecasts With Observing System Simulation Experiments (osses). Part I: An Idealized Case Study of Three U.S. Storms

Abstract

Severe weather events can have a significant impact on local communities because of the loss of life and property. Forecast busts associated with high-impact weather events have been attributed to initial condition errors over data-sparse regions, such as the Pacific Ocean. Numerous flight campaigns have found that targeted observations over these areas can improve forecasts. To better understand the impacts of measurement type and sampling domains on forecast performance, observing system simulation experiments are performed using the National Centers for Environmental Prediction Global Forecast System (GFS) with hybrid 3DEnVar data assimilation and the ECMWF T511 nature run. First, three types of simulated perfect dropsonde observations (temperature, specific humidity, and wind) are assimilated into the GFS over a large idealized sampling domain covering the Pacific Ocean. For the three winter storms studied, forecast error was found to be significantly reduced with all three types of measurements providing the most benefit (%–15% reduction in error). Instances when forecasts are not improved are investigated and concluded to be due to challenging meteorological structures, such as cutoff lows and interactions with atmospheric structures outside the sampling domain. Second, simulated dropsondes are assimilated over sensitive areas and flight tracks established using the ensemble transform sensitivity (ETS) technique. For all three winter storms, forecast error is reduced up to 5%, which is less than that found using an idealized domain. These results suggest that targeted observations over the Pacific Ocean may provide a small improvement to winter storm forecasts over the United States.