An ensemble approach to weather forecasting

A potential India-U.S. partnership could lead to better forecasting through collaboration.

During President Obama’s recent visit with officials in India, one of the more interesting topics discussed addressed potential partnerships between the Indian Meteorological Department and U.S.-based forecasting agencies and corporations. I was asked for my opinions on this topic by a New Delhi-based correspondent. Below you’ll find my abbreviated reply:


The primary beneficiary will first become apparent for end users in India, but the gains will extend beyond the borders to virtually any country/entity with a financial or social exposure to raw material prices.

Science progresses through the open exchange of information, and while on one hand the applied benefits of scientific research and the resultant applications can be commercialized and therefore protected, the open nature of collaborative agreements such as the India/U.S. item discussed can also be a great benefit to a much larger segment of society.

Many public and private weather groups have developed and refined techniques to develop monthly and seasonal weather forecasts. Forecasting, and in particular long-range forecasting, often relies on a blend of common physical/fluid dynamic principles coupled with a variety of closely guarded mathematical approaches. As such, while the basic scientific principles underlying the development of a forecast may be generally the same regardless of the source of the forecast, each public or private forecasting group puts their own spin on the forecast, where they try to separate themselves from their competition.

However, there is strength in numbers. So from the point of view of the end user, a forecast can gain a higher confidence if it is in more agreement with other reputable forecasts.

For many years, the Indian Met Department (IMD) was the only source for the agricultural sector when it came time to develop plans around the annual monsoon. For growers, these decisions include seed variety, plant/harvest dates, quantity of pesticides, etc., which carry a heavy financial burden, and oftentimes the decisions are made based on one (IMD) forecast. In recent years, the IMD has had a less than spectacular record in their seasonal rainfall forecast during the monsoon, so this serves as a time where other approaches can and should be considered.

The result of a potential collaboration via partnering with NOAA or other non-Indian weather groups can only enhance the IMD’s own forecasting process. By exchanging some methods, the IMD can learn about where the strengths and weaknesses lie within their own methodology. In the long term this is a very successful methodology.

The result will then be a better forecast for not only the IMD’s primary customers (the agribiz community), but others who also have an exposure to fluctuation in prices of important raw materials sourced from India. Further, better forecasting and monitoring techniques that are jointly developed will serve to provide more price transparency in the futures markets of related commodities, and minimizing price volatility is good for both producer and consumer.

There is no single forecast group than can develop a long range weather outlook that is correct 100 percent of the time. Taking an “ensemble approach,” where results from several different forecasts are used to guide and continuously refine a seasonal outlook, is a safer way to approach the weather risk associated with an upcoming season.

Related:

tags: , ,
  • Alex Tolley

    “Taking an “ensemble approach,” where results from several different forecasts are used to guide and continuously refine a seasonal outlook, is a safer way to approach the weather risk associated with an upcoming season.”

    Can you expand on this more? At best, this would add bounds to likely forecasts. Are you saying that the “best” forecast methodology changes with time and thus will be one of a number of methods used? Isn’t weather forecasting 1 week ahead still not better than 60% accurate? If so, how will a number of different methods with 60% +/- accuracy be that much more useful?

  • http://www.weathertrends360.com/ Michael Ferrari

    You are correct in noting that the typical point forecast is not very accurate beyond 1 week. Actually, the skill level drops off significantly after around day 5; the limits of science can really only develop a somewhat reliable short range forecast using traditional methods out to about two weeks – after that chaos introduces too much error.

    However, the techniques used in developing seasonal forecasts are very different than just extending the short range forecast out for a few more days; as such, skill in seasonal assessments tends to be significantly higher than 60%.

  • John

    In India everything going wrong is currently blamed on the monsoon. Yesterday I asked why the printer (which was already in repair before the monsoon started) was still not back. I got the same answer, used for every delay in these days: “Sorry. …you know: the rains…”.
    If the monsoon forecasts could be improved, not only farmers would benefit, the whole Indian economy would get a boost!