The long view Meteorologists don't always get tomorrow's forecast right, so how can they predict an entire season in advance? Jeffrey Schultz* analyses the capabilities of long-range weather forecasting

The effectiveness of long-range weather forecasting is an interesting area to explore, as the answer tells us a lot about the capabilities of science today, as well as important advances made in recent years. While conversation can sometimes turn to a recent forecast error ("they said it was going to rain today!"), in reality, short-term forecasting has made significant improvements over the last few decades.

Short-term forecasts, which predict daily weather events over the next several days, have increased in accuracy because of advances in computer technology, better understanding of atmospheric processes and an increased observational network consisting of radar, satellites and automated equipment. For example, forecasting skill of the fifth day has more than doubled since the late 1970s, and projections of major storms on day five are now as skillful as day three forecasts were ten years ago.¹

In the realm of long-range seasonal forecasting, discoveries made in the 1980s and 1990s regarding our understanding of atmospheric and oceanic interactions have made it possible to forecast an entire season several months in advance. The major difference between long- and short-term forecasting is that long-term forecasting does not attempt to predict day-to-day weather events over the course of a season; instead the forecasts determine whether the overall whole of the season, say total precipitation or average temperature, will be above or below normal. Another important difference is that long-range forecasts are given in probabilistic ranges for the season.

Forecasting exact amounts for a season, or using resolutions down to a day, are not scientifically sound. Instead, probability distributions, giving the chances for all possible outcomes, are appropriate. The American Meteorological Society's Policy Statement on Weather Analysis and Forecasting says: "No verifiable skill exists or is likely to exist for forecasting day-to-day weather changes beyond two weeks. Claims to the contrary should be viewed with scepticism." The Center for Ocean-Land-Atmosphere Studies (Cola) agrees: "The sequence of [daily] weather events cannot be predicted precisely beyond 1-2 weeks, [but] the atmospheric circulation and precipitation, averaged for an entire season, are predictable."

One should be wary of secret and 'proprietary' methodologies, and check with government and academic sources for the validity of extraordinary forecasting claims.

Forecasts can be made for areas as large as states or regions, or can be made site-specific, down to one weather station in a city. Critical days forecasts, such as the number of days the maximum temperature will exceed a threshold, and multivariable forecasts, such as the number of days the minimum temperature and snowfall exceeds their thresholds have been done. End-users of these forecasts include industries from retail to agriculture to energy.

A meteorologist starts the seasonal long-range forecasting procedure with historical climate records, or a climatological database. Climate data provides the all-important foundation from which a forecast is whittled. The data should be as long, complete and as 'clean' as possible. Adjustments for warming or cooling trends, station moves and equipment changes all help clean abnormalities. It would be quite dangerous to base a forecast on 10, 20, or 30 years of data if a very recent change occurred. For example, when the Sioux Falls, SD weather equipment was upgraded, it was also repositioned, moving a quarter of a mile closer to the edge of a river bank. The new site was much cooler, and often enveloped by fog. Without this knowledge, forecasts based on the older climate record will be too warm.

Changes in equipment need to be accounted for. It is very important to note, however, that when compared to natural climate variability and the margin of error in long-range forecasting, most data cleansing and adjustments that are made will be insignificant.

Next, the status of ENSO (El Niño - Southern Oscillation), and other oceanic patterns and atmospheric teleconnections, should be accounted for. The phase, strength, timing and longevity are all factors that affect the impact of an ENSO event on atmospheric circulation. The persistence of certain teleconnections, such as the North Atlantic Oscillation (NAO), can cause blocking patterns that last anywhere from a few weeks to several months. Soil moisture conditions also play a role during spring and summer. Dry soil allows a positive feedback to set up, where solar radiation goes completely into heating the ground and thus the air, rather than partially into moisture, which would absorb solar radiation through evaporation. Another positive feedback that would be a factor for winter temperatures would be an expansive snow cover across Canada and the northern US, which would re-enforce already chilly air masses as they head southward.

A meteorologist next turns to computer models, both statistical and dynamical in nature. Statistical models incorporate the evolution of atmospheric circulation patterns, sea surface temperatures, and land surface temperature and precipitation over the preceding year. Long-term warming/cooling and wet/dry trends are accounted for, including multi-year regimes. Analogue year data, which compare similarities of past seasons to the current and upcoming season, are also an essential tool. It is important not to get carried away with statistical links; one can find a correlation between anything. Temperatures in New York, for example, can be linked to El Niño activity or the Yankees' home record. Understanding the physical reasoning behind an apparent link, along with proper use of statistics, will help weed out occurrences resulting from chance. This is where we examine dynamical, or physical, models.

Dynamical models simulate the interaction between ocean, land, and atmosphere, and are heavily influenced by ENSO phases. They are based on the laws of physics, and how the atmosphere transfers heat and energy within large-scale circulation patterns and storms. These models have improved through years of research and discoveries by academia, government and industrial meteorologists. They will only get better and more accurate as we continue to research and unlock more secrets of the atmosphere and its circulation.

Although long-range weather forecasting is relatively new, it is useable now. Long-range forecasts are not as accurate every season as short-term forecasting is every day, but they definitely do have skill, and using them over a period of time will prove to be better than going on historical data alone. Better education and understanding of the capabilities of this science is necessary for wider acceptance and use of long-range weather forecasting. And over time, companies who use these forecasts will be able to better manage risk, reduce volatility and maximise gains.

*Jeffrey D. Schultz is chief climatologist at Weather 2000, based in New York (www.weather2000.com)

¹ American Meteorological Society's Policy Statement on Weather Analysis and Forecasting, Bulletin of the American Meteorological Society, 79, 2161-2163

COPYRIGHT NOTICE: Reproduced from Weather Derivatives, an FOW publication © Metal Bulletin Plc 2000 All rights reserved. No part of this publication (text, data or graphic) may be reproduced, stored in a data retrieval system, or transmitted, in any form whatsoever or by any means (electronic, mechanical, photocopying, recording or otherwise) without obtaining Metal Bulletin plc's prior written consent. http://www.fow.com/