The proposed work will advance scholarship on the statistical post-processing of ensemble weather forecasts. Innovations are promised in several key areas: efficient specification of the forecast problem and training dataset, the possibility of automating predictor selection, the automation of the choice between statistical models, the integration of quality norms derived from the particulars of a given decision problem, the creation of an expert system to define the underlying node structure of the multi-period decision value calculation, and an automated algorithm for analyzing the meteorological and predict and outcome training data to specify the parameters of each decision node. The overall project goal is to develop and code these algorithms, and to prepare a full technology transfer plan for the commercialization of the system.
The proposed work will lay the basis for a wide range of applications in which weather forecasts are fed directly into automated decision systems. The approach envisioned will allow the value of fully probabilistic forecasts to be exploited in scientific, data-driven approaches to risk management across a wide array of weather-sensitive sectors, including energy, insurance, tourism, transportation, aviation, construction, and public safety.
We have made substantial progress in translating existing techniques for the development of statistical forecast models for weather sensitive economic variables from an expert-intensive "cottage industry" into an automated algorithm suitable for wide-scale deployment. Many organizations struggle to respond intelligently to weather uncertainties. Algorithmic, weather-aware decision systems relying on machine-readable forecasts promise economic benefits realized through smarter, more nimble responses to shifting weather forecasts. To adopt such strategies requires that one create, calibrate and maintain statistical models of weather impacts. These models translate general-use weather forecasts into quantitative predictions of impacts on specific activities. Relatively few organizations are prepared currently to muster the mix of resources---historical weather forecast data, database design skills, statistical talent---required to undertake such specialized analytic tasks. The proposed technology development effort strives to lower these barriers. The commercial goal is a service that allows organizations to access the analytic resources they need to model and predict weather's impacts on their operations and finances, without creating their own weather risk analytics capacity. The technology will democratize access to ``smart'' management applications in energy, agriculture, and other weather-sensitive commercial sectors. It also promises to improve society's ability to respond to a range of weather-related challenges in public health and safety. Finally, the work promises to advance scholarship on techniques for blending multiple weather forecasting systems into calibrated predictive models of weather impacts.
