Much work has been done investigating the influence of Tropical Pacific sea surface temperatures (SSTs) on climate, in particular the seasonal influence of the El Nino-Southern Oscillation (ENSO) phenomena. Over some land areas and seasons, tropical SST and ENSO have been shown to have a substantial effect. However, investigation of SST influence on monthly time scales, even using near global SST data, has so far produced somewhat limited correlations and predictive skill over much of the earth's surface, especially when precipitation is considered.
Here we show substantial evidence of SST influence over the majority of the U.S. during much of the year. Using monthly SST anomalies (SSTA) from one or more ocean regions and a combination of statistical methods, we find relatively high SSTA-precipitation correlations, i.e., multiple (bias-corrected) correlation coefficients about 0.2 - 0.4, and 0.3 - 0.6, on monthly and 3 month time scales, respectively, particularly from October through June. These results suggest that substantial prediction of precipitation and climate on a 1 to few months lead time is a possibility. We compare correlations and predictions based on the much of the Pacific basin (from about 30S to 56N), the Gulf of Mexico, Atlantic (from about 25S to 66N), and the "North Atlantic Oscillation" region just south of the tip of Greenland (45N to 61N and 25W to 48W). We note a number of correlations which appear to be due to atmospheric effects. This work was limited to the U.S. for simplicity; however, similar methods would be expected to be applicable to areas which have comparable (or longer) precipitation data bases. Previous work by others suggests correlations of comparable or larger magnitude can be obtained.
Precipitation was chosen for investigation (and method development) because of its typically difficult statistical qualities even on a monthly time scale (high noise and severely non-Gaussian distributions) and its influence on agricultural yield. Methodological experimentation was needed. A significant step was developing a method to transform precipitation distributions into Gaussian form. This allowed Principle Component multiple regression to be used with some confidence. We found that cross-validation would likely miss significant results. Methodology was checked using Monte Carlo simulations