Data assimilation (DA) for NWP combines a short forecast (“prior”) with current observations to create an analysis, which serves several purposes: accurate initial conditions for long forecasts, boundary conditions for limited area models, input into real-time tactical decision aids, and the initial conditions for the short forecasts that become the prior for the next DA cycle. Observation and forecast error statistics, represented by error covariance matrices, control how much information is drawn from the observations as well as how that information is spread to observation sparse locations. Improving the models of these covariance matrices is the subject of much current work.
The social and economic impacts of hurricanes in the U.S. in the last few years stressed the necessity to better understand the variability, predictability and risk of tropical cyclones. In this talk, I will discuss the current understanding of various aspects of tropical cyclones, from the ability of the current generation of models to make forecasts on sub-seasonal time-scales, to the estimates of hurricane risk in locations with very few occurrences in the historical record.
Abstract: A Geometric View of the Ins and Outs of Simulation-based Forecasting Leonard A Smith Centre for the Analysis of Time Series, LSE Pembroke College, Oxford. Since the 1950’s weather forecasting (on every timescale) has been challenged by the nonlinearity of our models and the limits of our computational power.