“Everybody talks about the weather, but nobody does anything about it” – until now that is. Using weather derivatives, risk managers are beginning to respond to the cold spells, heat waves, droughts or flooding that can otherwise eat into their bottom line.
The growth of the weather-derivatives market lends potentially great value to information about the past, present and future state of the climate in a particular region. The Risk Prediction Initiative (RPI), a business-research partnership of the Bermuda Biological Station for Research, is working to maximise the value of climate information and forecasts by creating a partnership between climate scientists and the businesses that are involved in the nascent weather derivatives market.
During a recent weather derivatives workshop, the RPI encouraged climatologists, energy providers and (re) insurers to identify the types of climate research that can most benefit derivative traders. According to RPI science programme manager Rick Murnane, the participants worked “to separate fact from fiction, sense from nonsense in the seasonal prediction of temperature and precipitation anomalies and heating- and cooling-degree days.”
Accurate historical data concerning the normal climate experienced in a city or broader region is a prerequisite for any weather-derivatives contract, as these data provide the “climate normals” on which weather derivatives trades are based.
Yet obtaining an accurate climate normal at a given site is a challenge. Says New Jersey State Climatologist and Rutgers University Professor David Robinson: “Gathering accurate climatic data is difficult. Anyone planning to collect data is aware that inaccuracies and biases may result from problems at any stage of the acquisition and archiving process. Simply put, you must ‘know' your data.”
“Knowing your data” requires recognition of the main sources of error and bias in climatic information: data records of insufficient duration, inaccurate recording instruments, inconsistent observing methods and changes in weather station's location or surroundings.
The shorter the climate record, the more likely it is to be biased by variability such as a multi-year drought. According to the US National Weather Service, historical records should be at least 30 years long for use in generating climate normals. This allows data analysts to recognise and smooth any short term weather variations that might otherwise bias a shorter time series.
However, shorter climate records are also less likely to be biased by changes in instrumentation, observation methods, location, surroundings or a long term trend in temperature. Thus, the normalising period used in degree-day contracts is sometimes shorter than 30 years, particularly in areas with a significant temperature trend.
Accurate instruments are required to ensure reliable measurements and to minimise “data drift.” Data drift occurs when a weather instrument such as a thermometer gradually drifts out of calibration. If not detected and remedied at an early stage, data drift can produce a spurious trend in climate.
The use of consistent methods is also crucial. For instance, a shift in the time at which observations are made can significantly affect the record of daily maximum temperatures exceeding an extreme threshold at a given weather station. Observers at most stations record data once a day (most commonly at 07:00 hours, 16:00 hours or midnight), using a thermometer that shows both the current temperature and the maximum temperature recorded since the last observation. Daily temperatures usually peak in the afternoon around 14:00 hours.
Now imagine a scenario of two consecutive calendar days in which the maximum daily temperature on the first calendar day passes 85° F at 14:00 hours and remains above 85° F until 16:30. On the second calendar day, the temperature peaks at only 81° F. If this scenario took place at a station that employs a 16:00-hour observation time, the resulting climate data would indicate the occurrence of two consecutive days with a high temperature above the 85° extreme threshold, as the thermometer would carry the maximum temperature value from the observation time reading on the first calendar day forward to the second.
Yet a station using a midnight observation time would only record a single day with an extreme maximum daily temperature. Over time, such observation-time biases can accumulate to significant levels. Switching the observation time at a station can generate discontinuities of more than 1°F in the daily maximum-temperature record.
A final requirement for “knowing your data” is awareness of any changes in a weather station's location or surroundings. The sprawl of urban heat into previously rural areas has introduced a warming trend into many climate records around the world. This “urban heat island” effect has been removed from data sets used in global-change studies, to expose more clearly any signal of global warming. But what climate-change scientists consider “noise,” derivative traders may well consider “signal.” It makes little difference to the parties in a cooling-degree-day contract weather if an observed rise in mean daily maximum temperatures is due to urban heat or global warming. Either one will make homeowners more likely to turn on their air conditioners.
As the use of weather derivatives increases the value of climate data, efforts to minimise data-collection errors and biases are increasing as well. A case in point is the continuing shift by the US National Weather Services to its automated surface observing system (ASOS). ASOS comprises a network of 900 fully automated airport weather stations that update observations every minute, 24 hours a day, every day of the year. It more than doubles the number of full-time surface weather observing locations in the United States.
Unfortunately, the data on which most weather derivatives trades are now based must rely on historical climate records that predate ASOS. Moreover, the shift to ASOS can itself introduce bias into climate records, as it resulted in many station relocations.
Of course, the ultimate prerequisite to knowing your climate data is the availability of descriptive records that document the history of a weather station, its instrumentation and any changes in its surroundings. Without this type of “metadata,” there is no way to know whether an observed climate trend, or the value of a climate normal, results from an inaccurate instrument, a change in observation time or urbanisation.
While viable methods exist to adjust the effects of non-climatic biases on monthly mean temperature and precipitation values, efforts to evaluate and adjust such biases on extreme and daily climate data remain at the forefront of research.
Art DeGaetano, of the Northeast Regional Climate Center at Cornell University, has developed new statistical procedures to identify and adjust daily climate records that have been biased by changes in observation time or by urban sprawl. According to Dr DeGaetano, the method can account for 90% of the bias introduced by a shift in observation time. The method involves comparing data from the station in question with data from a network of surrounding stations. Testing with artificially biased data sets shows that the method identifies discontinuities of ±1° in about 75% of the cases.
Dr DeGaetano's new method and methods previously developed to “clean” monthly mean temperature and precipitation records require metadata for their implementation. Yet biases in a climate record can also be introduced by non-documented changes such as an improperly maintained instruments. Dr DeGaetano is now working on statistical techniques to recognise and rectify these types of data problems as well.
Reliable historical data for generating accurate climate normals is the first prerequisite for any weather-derivatives contract. Another crucial component is access to a suite of seasonal forecasts that predict how the weather during the contract period will compare to long-term mean climate conditions. A suite of differing forecasts provides the uncertainty that underlies any transfer of risk. If both parties to a contract have access to only one forecast, or if all forecasts have perfect skill, the outcome would be known in advance and a deal would be moot.
El Niño and La Niña are the best understood and most used predictors of seasonal climate variability in the US. Years of research have led to the capability to forecast these events and their effects on rainfall and temperature as long as a year in advance. During the 1997-98 El Niño, a persistent storm track enhanced storminess in California and the southeast, while northern states experienced a milder winter and fewer heating-degree days. During the current La Niña, a different pattern of storminess has been forecast and observed. The Pacific northwest has experienced extreme storminess and an increase in heating-degree days, while the eastern US has recorded a decrease in heating-degree days.
But El Niño-related climate variability is only half the picture. Other climate cycles such as the North Atlantic and North Pacific oscillations have comparable impacts on US climate and may also affect Europe.
According to Ants Leetmaa, director of Climate Prediction Center of the National Oceanic and Atmospheric Administration (NOAA), new research indicates a strong seasonality and regionality to these effects. “Research is underway to understand the physical origins of the variability,” says Dr Leetmaa. “This is a prerequisite for developing a forecast that incorporates these effects.” Also important is the development of faster and higher-resolution computer models.
In addition to concerns about accurate data and access to useful seasonal forecasts, derivative traders must contend with the possibility that human-induced global warming may be shifting underlying climate means and the probability of extreme events. This is the murkiest component of the derivative trader's triad of concern.
The observed global warming of 0.3 to 0.6°C since the late 19th century has not been uniform in space or time. Analysis of US daily data back to the late 1940s reveals significant regional and seasonal variations in maximum and minimum temperature. Temperatures in the northeastern US exhibit a warming trend during the last century, whereas temperatures in the southeastern US have cooled. Also, about two-thirds of the global temperature increase over the last century has occurred during the past 30 to 40 years.
If the global warming trend continues, an enhanced hydrological cycle should theoretically lead to increased rainfall rates and more flooding. Available evidence confirmed this hypothesis, with measured increases in total precipitation in the US, Canada, Australia, Russia, and China during the 20th century. This precipitation increase reflects an increase in heavy rainfall events and a decrease in moderate events.
Some efforts have been made to generate indices that would allow for “fingerprinting” of trends in extreme events, but according to Mike Crowe of NOAA's National Climatic Center: “The relationship between climate extremes and the response of degree-day totals over a season has yet to be thoroughly explored.”
The analysis of extreme events and the development of climate-extreme indices remain challenging research areas. So, too, do efforts to develop new methods of normalising climate data and identifying predictable correlations between these data and climate cycles such as El Niño or the North Atlantic oscillation. Funding from the RPI is now driving this research forward.