Forecasting for decision support

Eva Regnier - NPS Monterey
Monday, November 2, 2009, 4:00 pm, Engineering 2 Building, Room 180
Hosted by Assistant Professor Pascale Garaud
Applied Mathematics & Statistics

Abstract

Tropical Cyclones (TCs) are high-impact phenomena, and therefore the
U.S. Navy (among others) is willing to incur costs in the tens of
millions or higher on the basis of a forecast to mitigate TC impact.
Recent advances in probabilistic forecasting support risk-based
balancing of costs and mitigation potential. However, they do not
support dynamic optimization of decisions such as rational setting of TC
conditions of readiness, optimal-track ship-routing, and rollback of the
shuttle (and its successor Constellation rockets) from the launch pad.
The goal of this research is to construct Markov models of tropical
cyclones that match the probabilistic forecasting performance of
simulation-based operational forecast products and can be used to
support subjective and automated dynamic decisions. In order to exploit
the highly valid predictors such as numerical guidance and
satellite-derived parameters, it is necessary to explore a large set of
candidate variables. Therefore automated discovery techniques are being
adapted and evaluated for selecting variables and partitioning the state
space to create high-performing Markov models.