In a variety of disciplines large, numerical simulations have become a fundamental scientific tool. A key problem is how to inform or update such simulations in real time with large numbers of noisy observations, especially when many of the predicted variables are unobserved or the observed quantities bear a complex relation to the predicted variables. In principle, Bayesian methods provide a solution to this state-estimation problem, but evolving and updating the required probability distributions are problematic in practice, as the most straightforward approaches require computations of overwhelming size.
These collaborative investigators will address these issues through the use of novel ensemble-based or Monte Carlo approaches and within the context of numerical weather prediction (NWP). Weather prediction is a challenging test of any approach to state-estimation, as operational models for the continental United States will soon have of the order of 108 degrees of freedom and ingest an observational data stream of more than a terabyte per day. The Principal Investigators' application of ensemble state-estimation techniques to NWP is motivated by recent success in test problems with simulated observations, ranging from the prediction of isolated thunderstorms in a cloud model to global atmospheric flow in a general circulation model, and by potential advantages over existing operational data assimilation schemes. In particular, ensemble-based techniques directly estimate the uncertainty of the prior prediction and thereby avoid the assumption of stationary, isotropic forecast uncertainty made in most existing schemes. The benefits of this direct estimation will also likely increase as next-generation of NWP models reach resolutions of about 1 km and the use of remotely-sensed observations, such as from the operational network of Doppler radars, increases at those scales. Thus, this research will lay the foundation for a significant step forward in weather forecasting, especially at the scales where most severe and disruptive weather occurs.
The proposed work will be carried out within the context of the Weather Research and Forecasting (WRF) model, which is a next-generation NWP model designed for use at the horizontal resolutions of 1-10 km. The WRF model will be employed in operational weather forecasting and also will be supported for use by the research community. Use of WRF multiplies the educational benefits of this project beyond the direct involvement of students and postdoctoral researchers and provides a clear path to the implementation of results to improve routine weather forecasts.
The team assembled within this group Information and Technology Research project includes leaders in ensemble assimilation techniques as well as members with expertise in numerical modeling, ensemble forecasting, and the interpretation of Doppler radar observations. The project will be coordinated through joint supervision of graduate students and postdoctoral fellows, joint publications and annual workshops. In addition, common software will be used in all the research, thus facilitating the transfer of methodologies and expertise within the group.
Successful completion of this research potentially will provide significantly improved capabilities in weather numerical models. These improvements will allow advances to be made in the forecasting of a variety of weather phenomena.
