The Use of Adaptive (targeted) Observations in Operational Numerical Weather Forecasting

Over the past 40 years there have been significant improvements in weather forecasting. These improvements are primarily due to (1) improved model physics and increased numerical grid resolution made possible by ever-increasing computational power, and (2) improved model initialization made possible by the use of satellite-derived remotely sensed data. In spite of these improvements, however, we are still not able to consistently and accurately forecast some of the most complex nonlinear diabatic mesoscale phenomena, such as propagating tropical mesoscale convective systems/cloud clusters, tropical storms, and intense extratropical storms. These phenomena develop over very fine spatial scales of motion and temporal periods and are dependent on convection for their existence. Poor observations of convection, boundary layer dynamics, and the larger scale pre-convective environment are often the cause of these substandard simulations and thus require improved observational data density and numerical forecast grid resolution. This paper performs a set of Observing System Simulation Experiments (OSSE). The objective of the OSSE experiments is to demonstrate that an adaptive (targeted) observational strategy can improve forecast accuracy over existing more conventional observational strategies in terms of enhancing the initial conditions and subsequent accuracy of the simulations of a numerical weather prediction model. For the proof of this concept, hurricane Floyd (1999) is chosen as a test case. The set of experiments starts from a baseline high-resolution forecast of hurricane Floyd using the Operational Multiscale Environment model with Grid Adaptivity (OMEGA). This baseline run serves as the truth set for the OSSE under a “perfect model” assumption. From the baseline run, atmospheric vertical profiles were extracted to simulate “pseudo-observations” using different adaptive strategies. These data extracts were used to create new coarse-resolution forecasts of hurricane Floyd that were then compared against the both baseline and real atmospheric observations. In general, the experiments show that additional adaptive observations in sensitive areas can help to reduce hurricane forecast errors significantly from a Numerical Weather Prediction (NWP) model.