© 2008, 2009 KnowledgeToTheMax


The model that revolutionized meteorology

In 1978, the weather could be forecasted with statistical significance over no more than one month. In that year, a team set out to try to improve the situation by substituting the principles of reasoning called ultra-optimization for the heuristic methods that had previously been employed in building weather forecasting models. Only 126 years of observational data were available for the combined purposes of constructing the model and conducting an independent validation trial.

In 1980, the team announced its findings (Christensen et al, 1980e). The model which they had constructed forecasted precipitation with statistical significance over twelve to thirty-six months, depending upon circumstances. This amounted to an improvement of a factor of twelve to thirty-six!

The model is based upon three weather patterns that were discovered in the research. They predict a wet or a dry year in the Sierra Nevada, east of Sacramento, California. A wet year signifies precipitation above the multi-year median while a dry year signifies precipitation below the multi-year median.

A conditional inference is associated with each pattern. The patterns and the associated inferences are as follows:


Pattern No. 1:

o   normal or high Pacific Ocean surface temperature 2 summers ago, in the western portion of the ±10° equatorial belt AND,

o   normal or high sea surface temperature 3 springs ago in the northeastern portion of this belt AND,

o   moderate or low precipitation at Nevada City 2 years ago.

Conditional inference No. 1:

     Given pattern No. 1, probability next year is wet = 0.59±0.11.



Pattern No. 2:

o   NOT Pattern No. 1 AND,

o   steady or declining winter sea surface temperature in the equatorial belt last year AND,

o   normal or high sea surface temperatures in the western equatorial belt last year AND,

o   high precipitation at Colfax this year AND,

o   low tree ring growth in Truckee

Conditional inference No. 2:

     Given pattern No. 2, probability next year is wet = 0.22±0.15.



Pattern No. 3:

o   NOT Pattern No.1 or Pattern No. 2 AND

o   low precipitation at Placerville this year AND

o   declining precipitation at Mazatlan (this year relative to last year)

Conditional inference No. 3:

     Given pattern No. 3, probability next year is wet = 0.39±0.10.



In the validation trial, it was found that the model issued statistically significant forecasts at the 94% confidence level. On years that were extremely dry or wet, the confidence level increased to 99%.


An additional result was to discover the significance of the oscillation in the Pacific Ocean known as El Niño for long range weather forecasting.