Weather and the calculated risk: exploiting forecast uncertainty for operational risk management

Air & Space Power Journal, Spring, 2008 by F. Anthony Eckel, Jeffrey G. Cunningham, Dale E. Hetke

[ILLUSTRATION OMITTED]

Air Force SMART Operations 21 (AFSO21) has prompted a fresh look at ways to improve combat capability, including enhancing the decision-making process. (1) Highly effective and efficient operations require optimal decision making in situations that involve risk of unfavorable outcomes. Such risk exists due to uncertainty in decision inputs. operators routinely face a variety of inexact inputs, such as intelligence reports on enemy strength, projections on available logistics, and performance of weapon systems. this article explains how the uncertainty in one such decision input--the weather forecast--can be used within the principles of operational risk management (ORM) to improve combat capability by applying a new advancement called "ensemble forecasting."

Typically, Department of Defense (DOD) missions with weather vulnerabilities consider a single weather forecast, thus largely ignoring forecast uncertainty, which can often prove significant. Focusing attention on a single forecast leads to nonoptimal decisions. (2) Failure to consider an objective description of the potential forecast error leaves an operator overly vulnerable to costly mistakes and the wasting of resources--a situation analogous to betting on a horse race without considering each horse's projected odds of winning.

Clearly, the absolute best information for weather-related decisions would indeed be a consistently perfect deterministic forecast (i.e., a single-valued prediction for a weather phenomenon). Unfortunately, deterministic forecasting is anything but perfect. Forecast skill varies greatly due to the challenge of predicting the incredibly complex atmospheric system that contains nonlinear hydrodynamic, thermodynamic, radiation, chemical, and physical interactions. in fact, it is incredible that we can predict the atmosphere at all. (3)

The inherent uncertainty of the weather can be described with a "stochastic forecast," which expresses a distribution or range of possibilities that defines the potential error in the deterministic forecast and that can come in many different forms and from many different sources. For example, weather climatology (i.e., seasonal conditions) is normally given stochastically, such as the average, minimum, and maximum monthly expected rainfall at a location.

The idea of including uncertainty as part of a forecast is nothing new. (4) People recognized the potential value for applying stochastic forecasts within Air Force operations as early as the 1960s, but we have yet to capitalize upon it. (5) today's forecasting remains primarily deterministic because (1) application of deterministic weather for decision making is straightforward, (2) benefits from and methods of applying stochastic forecasts are not widely understood, and (3) production of robust stochastic forecasts for short-term forecasts (up to a few days) has not been practical or affordable. However, since advancements in science and technology currently support production of stochastic forecasts, now is the time for the Air Force to learn and pursue the advantages of this technique.

Production and Application of Stochastic Forecasts

The primary tool for meteorologists for the past 40 years has been computer-based, numerical weather prediction (NWP) modeling. Weather observations are analyzed and then fed into a complex algorithm that simulates atmospheric behavior over time to generate a single, modeled forecast that has a varying degree of accuracy. NWP models, run at meteorological prediction centers, cover domains of various size (from city to global), resolution (from a few kilometers to hundreds of kilometers), and lengths of time into the future (from a few hours to weeks) to meet specific needs.

Producing just one deterministic (single solution) forecast in an NWP model requires performing trillions of calculations very quickly to process the data in time to be useful. this production involves extremely powerful, expensive supercomputers. A typical operational NWP model uses computer hardware worth about $100,000 to $1,000,000, depending upon the model configuration (domain size/ resolution, run-time requirements, etc.).

Computers have now advanced to the point that running NWP models in a stochastic (multisolution) mode, using ensemble forecasting, has become cost-effective. in this type of forecasting, the NWP uncertainty is quantified by running the model many times (typically 20-30 individual solutions), with slight changes to the information fed to each model run as well as adjustments to the model's inner workings. this generates a spectrum of forecasts in which each forecast is a valid possibility; together, they yield an objective, stochastic weather forecast.

Additional processing to generate an ensemble forecast requires roughly an order of magnitude (10 times) more computer power, with an equivalently higher cost, compared to deterministic NWP. this article takes a DOD perspective on how the benefits of using ensemble forecasts greatly exceed the cost of their production. Across the weather-support community, the benefits for improving users' decision making have fueled extensive research and development over the past 20 years. meteorological prediction centers worldwide are currently generating ensemble-based stochastic forecasts for their customers: