A number of businesses use forecasts of extreme weather as a tool to mitigate risk or determine trading opportunities. Given that even the best forecasts are not entirely accurate, these businesses must work to understand the inherent uncertainties of weather forecasts and make decisions in spite of them.
Weather forecasting remains an inexact science despite many advances in recent years. The atmosphere is fundamentally a chaotic system, according to theoretical studies. They've proven that a tiny change in conditions can rapidly amplify with time--the so-called "butterfly effect."
Working with data from areas with few observations, or observations that are in error, can lead to modeling errors that grow with time. Imperfect model equations or those without enough resolution can also lead to errors that amplify over the course of a forecast.
For example, many atmospheric models currently run with horizontal grid spacing of 4 kilometers to 12 kilometers. Clouds or even small thunderstorms might be smaller than that resolution, so the model wouldn't effectively capture them. Should these small clouds organize into a larger complex of thunderstorms, the model would miss that too.
Besides modeling error, another major issue confronts businesses dealing in weather risk. While there are issues with lack of observations, especially in sparsely populated land areas and at sea, the larger practical problem in meteorology is the opposite: the overwhelming amount of raw data available--from surface observations at airports, to satellite and radar imagery, to gigabytes of computer forecast model output.
Meteorologists and forecast users need to distill this data into useful information using graphical displays, and understand the limitations of the models and analysis tools. For most situations, a meteorologist or business user is faced with forecasts from a variety of sources, each one different from the others.
The difficult part of the task is deciding which of the forecasts is the most appropriate.Perhaps it's some combination of several. This is generally done by keeping in mind the accuracy of the various models in previous similar situations, and remembering their biases. Some models tend to overforecast precipitation, for instance, or underforecast cold outbreaks.
Users of forecasts can also be best served by getting interpretation of forecasts from meteorologists, rather than just numerical outputs. Meteorologists can express the confidence level of a forecast and what other possible scenarios would be. The users can then decide what their risk tolerance is and choose whether to act on the forecasted weather.
Model ensembles are one way to better quantify weather forecast uncertainty. Over the last decade, increases in computer power have allowed operational weather centers to begin running many copies, or an ensemble, of the same model for the same forecast time. Within the ensemble, each copy is based on slightly different conditions and evolves differently. When the number of ensemble copies is large enough, on the order of 10 to 100 copies, the range of difference among them can be used to suggest the confidence level of the forecast, and what the various possible scenarios are.
As an example, consider hurricane track forecasting. Experts look at the output tracks from as many models as possible. These can be the same weather model run many times with slightly different initial conditions, or a variety of completely different models. The confidence level of the overall forecast can be understood by examining the tracks from all of these models. If they cluster toward a single solution, one can assign a much higher confidence level to the forecast than if they fan out into many different tracks. The forecasts might also cluster around two likely outcomes. That would mean that the average of all outputs is not the useful forecast; rather, the likely solution is one of those two possibilities.
A similar principle is used for temperature forecasts. On a day with a 20 percent chance of a thunderstorm, 80 of 100 model ensemble runs might forecast late afternoon temperatures between 93 degrees F and 97 degrees F, but the other 20 would show the effects of the thunderstorm with temperatures near 80 degrees F. Energy traders use this type of information in assessing expected electricity demand.
In many situations, decisions based on weather forecasts are used for trading in the financial markets. Traders can be successful in aggregate by trading on probabilistic forecast information, despite occasional forecast busts. The degree of uncertainty also gives an indication of the range of possible outcomes, so that traders can understand their downside risk. An energy trader could position himself to best profit by a city's temperature reaching 98 degrees F, but he or she could risk less capital on a day with 30 percent chance of a thunderstorm (and much lower high temperature) than on a day with a 10 percent chance.
Another market play can occur when a hurricane threatens to pass through the field of drilling platforms in the Gulf of Mexico. Even a near-miss by a strong hurricane will be disruptive to natural gas supplies as rigs are evacuated for the safety of personnel. An actual strike would obviously also inflict damage and lead to larger and longer disruptions. Natural gas traders use hurricane track forecasts to gauge the risk that the storm will affect the areas of the Gulf where natural gas is produced, given that disruptions to natural gas supply would cause price spikes.
Clearly, tolerances are different when forecasts are used for making safety decisions, such in coastal evacuations ahead of hurricanes or for the aviation industry. These users must frequently act to protect against the worst-case conditions, even if they are less likely.
BUSINESSES AT HAND
Though many businesses are affected by the weather, the heaviest users of weather forecasting services are those with the ability to react to the forecasts: either to avoid losses or gain business advantage.
The energy markets are highly reactive to weather conditions--in particular electricity and natural gas, which have volatile prices due to the difficulty of keeping reserve capacity. On the demand side, high summer temperatures increase electricity usage for air conditioning, while low winter temperatures drive high natural gas usage for heating. On the supply side, hurricanes threatening natural gas production and petroleum refining capacity in the northwest Gulf of Mexico can drive prices higher.
The insurance and reinsurance markets are most interested in weather conditions that lead to losses over a wide area. So hurricanes and mid-latitude wind storms are the largest hazards of interest. Seasonal hurricane forecasts have shown some skill at predicting the number of Atlantic storms in the upcoming season - this information can be useful when an insurance company is considering reinsurance.
The daily schedules of commercial airlines are highly dependent on accurate weather forecasts. Adverse conditions at a major airport that delay, divert or cancel flights cause ripple effects of problems throughout the system. Hazardous conditions in-flight along busy flight routes lead to air-traffic-control decisions to avoid the affected areas, leading to congestion and delays on the remaining routes.
Business users who understand the strengths and limitations of weather forecasting will have an advantage over their competitors. For high-impact weather forecasts, a clear sense of the confidence level of the forecasts is crucial, the interpretation of which is best provided by consultation with human forecasters. Once the forecasts and their certainty are clear, businesses can take action to reduce losses or take advantage of favorable situations.
WILLIAM RAMSTROM is a meteorologist in forecasting research and development at WSI Corp. His professional experience includes numerical weather prediction, operational forecasting for aviation clients and several years in the financial-services industry.
October 15, 2007
Copyright 2007© LRP Publications