HRD seminar – Dr. Dusanka Zupanski, Zupanski Consulting, LLC, Fort Collins, CO – 28 March 2016

Dr. Zupanski presented a seminar on “Theoretically advanced, yet computationally efficient data assimilation and forecasting methods”.


We have witnessed several fundamental advancements in data assimilation and forecasting methods in the last decade. Hybrid ensemble-variational data assimilation methods have been developed, utilizing the best properties of the most advanced data assimilation methods: ensemble Kalman filter and variational methods. Satellite radiance assimilation in cloudy conditions has indicated potential for significant improvements in forecasting clouds, precipitation and tropical cyclone rapid intensification. Forecast models have also become more advanced through including more realistic interactions between the atmosphere, land and ocean, via coupled models, resulting in forecast improvements across the coupled processes.

A big challenge that still remains is in the ever-existing requirements for theoretically advanced, yet computationally efficient data assimilation and forecasting methods. Hybrid ensemble-variational methods have already been proven theoretically advanced and computationally efficient and are increasingly being implemented in operational applications across the world.

Employing multi-model ensembles within the hybrid methods could make them even more computationally efficient and also improve their accuracy. In addition, coupled data assimilation has a potential for further forecast improvements, through propagating observed information via coupled forecast error covariance. In our presentation, we will show some examples of multi-model hybrid data assimilation and coupled data assimilation.

Finally, applications of data assimilation in the form of forecast post-processing could also bring significant forecast improvements, requiring negligible computational costs. As it will be explained in our presentation, this approach could extract valuable information from the last-minute observations and propagate it into the future, without the need for rerunning the forecast after data assimilation.

A recording of the presentation is available on the anonymous ftp site:

Screen Shot 2016-03-28 at 1.51.14 PM