WSP05759 - Laserfiche WebLink

Home Browse Search

OD3~?S ( c 6 improved, unaccountable shifts in critical events occurred due to the effect of mov- mg averages. In more recen[ work, lagged tree-ring variables and multi-variate statistical techniques have been used to determine the sign and magnitude of coefficems used to transfer ring-width variances and different lags in[o an estimate of specific cli- matic variables (Stockton, 1971). Box-Jenkins modeling (Box and Jenkins, 1976) can also be used to determine the appropriate persistence model by successive test- ing. Models range from a single autoregressive (AR) or moving average (MA) to more complex autoregressive moving average (ARMA). Meko (1981) evaluated both approaches and found that each had advantages and disadvantages depending on data conditions. Where tree-ring series had considerable autocorrelation, the Box-Jenkins approach provided a generally flaller response than the lagged predic- tor method. This evenness of response sacrificed accuracy at longer wave lengths by underestimating the amplitude of broad swings from the mean. The Box-Jenkins method is preferable if the ;;pecrrum of the long-term reconsrrucrion i~fJ~st because the method anempts to minimize the distortion of persistence. In contrast, the lagged predictor method ~ccepts frequency distortion as a trade-off forj!igher total variance explained. -- Landwher and Matalas (1986) used the stochastic properties of tree-ring data to test the desirability of using infinite memory processes in hydrologic simu- lations. Based on examination of 72 dendrochronological records, they found that fractional gaussian processes (infinite memory) were not the moSt appropriate mod- els for simulating persistence in the tree-ring data and, by inference, in hydrologic records.