WMOD00593 - Laserfiche WebLink

Home Browse Search

a relationship between increased December precipi- tation and increased tree growth the following growing season. However, the investigators are aware of some statis- tical parameters which influence any interpretation based on this system. The partial correlation coefficients in any regression have standard errors of varying magnitudes. For variables with high explanation the standard error is relatively low. But as new variables are entered into the regression the overall explanation may increase but the standard error also increases. For example, in the regression with Pagosa Springs climatic data when all 33 variables were entered, 99 percent of the growth variance was accounted fo~ but only the four listed above were valid at the 0.95 level of significance. The reason relates to the standard error. Eventu- ally, as more variables are entered, the standard error is larger than the coefficient itself and the error margin becomes too large. If fewer variables are used in an attempt to maintain low standard errors the overall percentage of explained variance is also reduced. Since these correlations are used in a prediction model, prediction would also be restricted. The question now is whether or not the predicted range of growth will be narrow enough to provide a framework for interpretation. -Stepwise Regression Analysis Stepwise regression analysis for ring width pre- diction were run using two sets of seasonal tempera- ture and precipitation data for Durango. The seasonal aggregation of the monthly dates are shown below. These groupings were defined so as to achieve coherence with regard to the biological processes of the tree growth in the San Juan' area. Set 1 Set 2 Summer (prior) [July August september] October ~ovember December January February] ~MarCh April May ] June ~JUlY August September October Spring-Summer Fall (prior) Winter (prior) Winter-Spring Spring Summer Winter or winter-spring precipitation was the first variable entered in five of the six analyses using SJS-1 or SJS-2 data. Spring precipitation came first in the sixth case. These results are in agree- ment with results of the preliminary analysis as well as with the findings of Fritts in Arizona and Colorado. A similar growth response has been noted by Landis and Mogren (this vol. p.39l). Inter- relationships of climatic variables have also been noted by Evans and Reid (this vol. p.403). The SJS-3 data did not show a similar regression relationship and apparently is not well suited for climatic inference. Table 5 presents the stepwise multiple regression equations developed for the three study sites using the Durango climatological data. Table 5. Stepwise multiple regression coefficients used in prediction equations Site a Var b Var c Var d Var e Var C Stand. Error summer summer spring prior prior winter winter SJS-1 +.0392 precip -.0161 precip -.0327 temp +.0209 precip +.0192 temp 2.57 0.32 summer winter spring prior spring SJS-2 +.0537 precip +.0508 precip -.0524 temp +.0295 temp X X 2.44 0.49 summer summer prior winter winter SJS-3 -.4638 temp -.2313 temp -.0891 temp +.0726 precip X X 45.99 1.58 Equation for above coefficients: Annual tree-ring width z a (xa) + b (xb) + .. . + e (xe) + C (1) Where: a, b, .. ., e are stepwise regression coefficients x a' xb' ... , x are the climatic variables for the growth year e C is a constant 75