SOUTHERN OSCILLATION EXTREMES RECONSTRUCTED FROM TREE RINGS OF THE SIERRA-MADRE OCCIDENTAL AND SOUTHERN GREAT-PLAINS

The El Nino-Southern oscillation (ENSO) is correlated with climate and tree growth over northern Mexico and the southern Great Plains of the USA. Warm events favor moist-cool conditions from October through Mar ch (event years 0 and +1), and subsequent tree growth (year +1) in the region tends to be above average. The opposite climate and tree growt h conditions prevail with less consistency during cold events. ENSO-se nsitive tree-ring chronologies from this region were selected to devel op two reconstructions of the Southern Oscillation index (SOI) back to 1699. For the first reconstruction, a multiple regression-based calib ration equation between prewhitened and regionally averaged tree-ring data from Mexico and Oklahoma and a prewhitened winter (DJF) SO index during the period 1900-71 were used to estimate the winter SOI for eac h year from 1699 to 1971. The tree-ring predictors account for 41% of the winter SOI variance from 1900 to 1971, and the reconstructed SO in dices are significantly correlated with independent winter SO indices available from 1866 to 1899. Because correlation analyses indicate tha t SO extremes have a much stronger influence on climate and tree growt h over Mexico and the southern United States than near-normal SO indic es, a second reconstruction of just winter SOI extremes was also devel oped. Discriminant function analysis based on the growth anomalies at eight individual tree-ring sites located in Mexico, Texas, and Oklahom a was used to classify 56 years from 1699 to 1965 into two opposite wi nter SOI ''extremes'' (i.e., greater-than-or-equal-to 0.5 or less-than -or-equal-to -0.5). Using a high posterior probability of membership i n either extreme category (P greater-than-or-equal-to 0.65), and ignor ing classifications of near normal winter SO indices, the discriminant functions identified 27 years as winter SOI extremes during the calib ration period from 1866 to 1965 (56 extremes actually occurred; reliab ility = 48%). However, 5 of the 27 classified extremes were actually n ear-normal indices (post agreement = 81%). With a reasonably stationar y relationship between Mexico/Southern Plains climate and winter SOI, validation tests indicate that the reliability of the classification-b ased reconstruction is about 50%, with a post agreement of at least 70 %. These calibration and validation results suggest that the 56 extrem es classified from 1699 to 1965 represent about half of the true numbe r of extremes during this 267-year period, and that each reconstructed extreme has up to a 70% chance of representing a true winter SOI extr eme. The most accurate estimates of past winter SOI extremes may be ac hieved in those years when the regression and classification methods o f reconstruction agree, but comparisons with the instrumental data ind icate that evidence for a past extreme cannot be disregarded when base d on only one method. Both reconstructions indicate an increase in the frequency of winter SOI extremes after ca. 1850. Because the regressi on and classification errors are randomly distributed through time, th ese and other reconstructed changes in event frequency may reflect rea l changes in the extratropical influence of the SO over Mexico and the southern United States, if not in the SO itself.