Forecasting & Data

The vast majority of weather observations are sourced from government-run weather stations.

These weather stations typically follow World Meteorological Organization (WMO) standards for meteorological observations.  These standards establish a scientific benchmark for the quality of instrumentation, station siting, and observing practices.

In order for a weather station to be eligible for weather hedging (to be used as the official source of data) it must meet basic requirements for data accuracy and reliability.  In addition to the basics, these stations must have a long historical record for pricing purposes (typically 10+ years of data) as well as reliable ongoing feeds for tracking and eventually settling a transaction.

Within the past few years the use of Gridded Data has become increasingly common in the market.

Gridded data is a catchall phrase that describes datasets sourced from meteorological reanalysis products to satellite-derived information.  Through a variety of methods, observations from various sources are combined and added to a regularly spaced grid.  Gridded datasets can provide weather data for elements and locations where no observed data is available.

As with all datasets, it is important to understand the limitations of these datasets before using them for hedging purposes.

To the surprise of most non-meteorologists, meteorological datasets are fraught with errors and missing reports.

In order to resolve these problems, market vendors will offer cleaned data products.  Cleaned data is data that has been processed to detect and replace erroneous observations and fill missing values.  Cleaning methodologies vary from company-to-company but will typically involve a combination of statistical analysis coupled with meteorologist oversite.

The goal of data cleaning is to provide a dataset that is ready for use in this market.

In addition to the missing and erroneous observations (corrected by data cleaning), some datasets suffer from discontinuities.

A discontinuity is a change in the observations resulting from a sudden and continuous event that has a consistent impact upon the data. Events such as station moves, instrument changes, and changes to the local environment (immediately local) can impact observations.

A classic example of a discontinuity would be building of a new runway in close proximity to a weather station. The additional pavement will often impact observations by increasing both the daily maximum and minimum temperatures. The goal of data recalibration is to correct for these events effectively bringing the dataset in line with the current observing state of the weather station.

Recalibrated Data offers a more accurate view when pricing weather hedgers.

Note, discontinuities are different from long-term gradual changes in observations (typically hotter, colder, wetter, dryer) often the result of regional or global climate trends.

Settlement Data is a data that has been processed by independent third parties for the specific purpose of settling a weather hedge.

These datasets have undergone an exhaustive review to detect erroneous observations due to:

• Instrument failure
• Instrument drift
• Observation tampering

The method used to detect and correct data errors as well as fill missing observations will be detailed as part of a Settlement Data specification.  The scope of the specification will vary from transaction-to-transaction with larger, data complicated, transactions going as far as necessitating independent weather station installation.

With that said, more often than not, transactions will simply adhere to industry best practice (statistical review with meteorologist oversite).

A weather forecast is a prediction of what will happen in the future.

Forecasts are used by Hedgers, Speculators, and Risk Takers to assess the value of a weather risk contract.

While there are a vast amount of weather models run by entities throughout the world, the most commonly known ones are:

• The Global Forecast System (GFS) from the US
• The European Center for Medium-Range Weather Forecasts (ECMWF) from Europe

Both groups provide an operational weather model and an ensemble weather model, leveraging global numerical weather prediction (NWP).

Weather models disseminate their data in grids, allowing for forecasts at any point of interest.

Ensemble weather models average multiple lower-resolution model members that are each run with slightly different initial conditions, allowing for different solutions. These ensemble forecasts allow for users to analyze possible solutions different from what the operational models show – or improve or degrade confidence in the respective operational model run.

While long-range climate models can provide weather outlooks spanning weeks to months, it is just as important to understand their limitations.

These models are skillful, but generally only for an overall idea or theme. For example, if it is 1 May and you want to know the weather for New York City in August, these models can indicate whether they project the month to be warmer or cooler than normal. While it is beyond the state of the science to project-specific temperatures for specific days that far in advance, there is skill in projecting the direction of the temperature anomaly.

The image below is a monthly temperature anomaly composite.

Using that, one would gauge that the month will see above normal temperature anomalies throughout much of North America, with the strongest focus in the Midwest and Northeast and north into Canada. While it would be ideal to know specific temperatures, these forecasts can still be incredibly useful if a warm or cool month – rather than weather on specific days – would have impacts on your interests.

In addition to weather models powered by computers, there are multiple companies that provide their own weather forecasts, independent of any model.

While weather models are a critical tool in forecasting weather, they are not perfect. Meteorologists that analyze the output from these models understand the ongoing biases, issues, and limitations in certain situations – especially when it comes to extreme events.

Using that knowledge, vendors are able to provide skilled forecasts for any location of interest, typically with better skill than any particular model.