Ryan Torn - Applying ensemble-based data assimilation techniques to understand the predictability and dynamics of mesoscale weather systems
Understanding the sources and evolution of numerical weather prediction errors is critical to improving forecasts of various atmospheric phenomenon. Errors can originate from two primary sources: the model initial conditions (i.e., the analysis), or errors in the model formulation (i.e., model error). Typically, initial conditions are produced via data assimilation, whereby a short-term model forecast is combined with observations to produce a best estimate of the atmospheric state. Improvements to the initial conditions can be obtained by either adjusting how observations impact the model state, or taking observations in regions of significant error growth.
Ryan Torn's research focuses on applying ensemble-based data assimilation techniques, such as the ensemble Kalman filter (EnKF) to understand the predictability and dynamics of mesoscale weather systems. These ensemble-based techniques have some advantages over operationally-used variational methods because it uses an ensemble of short-term forecasts to compute background error statistics, which determine the relative weighting of observations and how to spread observation information onto the model state variables; variational methods use quasi-fixed background error statistics based on long-term averages. Furthermore, the EnKF provides a set of equally-likely analyses, which can be used ensemble forecasting.
Forecast sensitivity analysis provides an objective means of evaluating how changes to an initial condition affect a forecast and where to gather additional observations to reduce forecast errors. Most previous sensitivity studies have used the adjoint of a linearized forecast model to determine how initial condition changes impact a forecast metric. Adjoints suffer from a number of difficulties including coding, linearity and moist processes. Ensemble-based sensitivity analysis provides an attractive alternative because it uses the statistics of an ensemble of analyses and forecasts from an EnKF system for this calculation and thus does not suffer from some of the adjoint’s limitations. Furthermore, this method is able to combine data assimilation and sensitivity analysis together in a consistent manner.
Since arriving at NCAR, Ryan has been working on applying these ensemble-based techniques to understand how initial condition errors can impact tropical cyclone track and intensity forecasts. It has been hypothesized that better storm-scale tropical cyclone analyses will lead to improved intensity forecasts.
This figure shows the sensitivity of a 48 hour forecast of Hurricane Katrina's minimum sea-level pressure to the analysis of the zonal wind component averaged between 250 and 850 hPa (shading) at 00 UTC 25 August 2005 as estimated from an EnKF system Ryan has been working with. The contours on this plot are the ensemble-mean analysis of the vertically-averaged zonal wind. At this time, the forecast of minimum sea-level pressure is sensitive to the winds associated with the tropical cyclone and the wind over the Gulf of Mexico. These regions indicate where additional observations could have benefited the forecast. Future work will focus on applying a similar analysis to other tropical cyclones to understand why some tropical cyclones are highly predictable, while others are not.
ASP Spotlight May 2008
For more ASP spotlights click here http://www.asp.ucar.edu/spotlight/archive.jsp