When Good Weather Isn’t Good Enough: The O2D (Operations to Decisions) Divide

When Good Weather Isn’t Good Enough: The O2D (Operations to Decisions) Divide

Peter P. Neilley and John K. Williams

The Weather Company, An IBM Business, Andover MA

Murphy (1993) noted that weather forecasts only realize value when used to assist in making good decisions. Yet nearly 25 years later, while there are many real-world decision support systems developed for specific weather-impacted scenarios, use of weather forecast information in objectively optimized decision making remains far from widespread. Rather, most routine weather-impacted decisions are made using a subjective mix of experience and intuition. How much fuel should be put on an aircraft facing uncertain weather at the destination? When should an evacuation order be issued for city in the face of a tropical cyclone threat? What route should one take to work given the expected snow? Such choices are routinely made without optimized decision guidance, despite the fact that such guidance is increasingly feasible.

The skill of our foundational weather forecasts continues to improve and presumably so do dependent weather-impacted decisions, regardless of how they are made. However, the foundational forecast data is also getting more complex, e.g., via the emergence of ensembles and forecast probability distributions, such that decision makers often are not able to fully digest or best use the information. As a result of this increasing complexity of forecast data, the gap between the potential value in our forecasts and the actual realized value implies that our science is leaving tremendous value on the table.

We explore three fundamental reasons for the lack of widespread implementation of objective, weather-influenced decision-making:

1. Many decisions are complex and dependent on a myriad of weather and non-weather factors, requiring decision makers to have expertise in multiple physical and social sciences. As a result, no one discipline has assumed general responsibility for maturing applications of the decision sciences to weather-impacted circumstances.
2. There is a lack of generic foundational tools upon which weather-based decision support capabilities can be constructed. As a result, each incarnation of a decision system generally builds soup-to-nuts solutions with little reusability or scalability.
3. Historically, our forecasts have been largely deterministic. Deterministic forecasts lend themselves to simpler decision making, but are not well suited for decision optimization.

We argue that the meteorological sciences should assume ownership of the development of weather-influenced decision support technologies as our community has the most at stake in realizing their value to society. The foundational tools needed include decision catalyzing technologies that are analogous to how a numerical weather prediction model converts observations into forecasts. The development of such infrastructures will enable us to accelerate the closing of the operations-to-decisions divide. Moreover, the moment is ripe for development of such tools given the emergence of Big Data, deep learning and other catalyzing paradigms.