The Development of a cyber infrastructure environment for ensemble prediction of hurricanes
College: Liberal Arts and Sciences
Award year: 2002-2003
Accurate prediction of hurricane track, intensity, timing and landfall is both a critical research problem and is of great importance from a societal impact standpoint. Improved predictions will reduce evacuation time and cost, mitigate property damage and save lives. However, research on hurricane track and intensity prediction is a computationally demanding task.
The atmosphere is a chaotic dynamical system. Therefore, any small error in the initial condition will grow with time, eventually leading to a total loss of predictability. In hurricane predictions, errors in observations of initial hurricane position, structure, intensity, and environment are compounded by approximations inherent in numerical model treatment of physical processes, such as precipitation and boundary layer physics. As a result, significant errors currently appear in hurricane track, intensity, timing and landfall location. Presently, operational numerical hurricane forecasting is carried out using a deterministic approach - meaning each model forecast employs a single prediction for a given storm. There is a considerable body of research that this approach has serious limitations and improvements will require a fundamental shift towards a probabilistic approach through the use of ensemble modeling techniques.
Ensemble forecasting (EF) entails many predictions per event to deal with the myriad uncertainties in a numerical weather prediction (NWP) system. A typical NWP system for the hurricane problem must account for the above uncertainties in initial and boundary conditions and model physics. Thoroughly addressing this problem requires making hundreds of forecasts for each event. Therefore, traditional cyber infrastructure environments and computational methodologies are ill-suited to carry out ensemble prediction research. New computational frameworks are needed for the design, execution, processing, mining and visualization of massive numbers of ensemble predictions. An additional consequence of this enormous problem is the need for the development of new tools and techniques for metadata and job management so that supercomputing resources are effectively and efficiently used. In other words, a new cyber infrastructure environment for end-to-end workflow is essential to carry out research on ensemble prediction of hurricanes.
We will work with NCSA scientists and staff members to develop this new cyber infrastructure environment. A portal interface developed by NCSA will be applied to this effort, along with metadata cataloging and mining on the DTF Grid. Specifically, we will work with Jay Alameda on workflow and execution of jobs on the DTF and with Michael Welge on mining the data using the existing D2K environment. The data mining effort will include the development of new objective clustering algorithms for knowledge discovery, which will be imbedded into the D2K environment. This research will be conducted using the WRF model on the TeraGrid. The hurricane prediction problem is not the only one likely to benefit from the new cyber infrastructure environment. The atmospheric sciences community has embraced the ensemble approach for predictions on all scales, from the simulation of thunderstorms to multi-decadal climate simulations. Any techniques and methodologies that will be developed in the proposed project will benefit other researchers engaged in similarly challenging computational problems.