Test of a New Method for Removing the Growth Trend from Dendrochronological Data

摘要: Tests of the compound increment function, introduced by Warren (1980) as a means for removing growth trend from dendrochronological data, are herein reported. In particular, interand intra -site correlations residuals generated new method compared with those standard exponential fits. It is also shown that, in presence non -climatically induced responses, such might arise thinning, fits can lead to spuriously high correlations. Accordingly, and because method's virtual elimination negative very low positive correlations, it appears be more satisfactory portraying trend. Es wird uber der von eingefuhrten zusammengesetzten Zuwachsfunktion als Mittel zur Eliminierung des Wachstumstrendes aus dendrochronologischen Daten berichtet. Insbesondere werden die zwischen -und innerstandortlichen Korrelationen Restvarianzen, wie sie mit dem neuen Verfahren und herkommlichen Exponentialausgleich entstehen, miteinander verglichen. Ferner gezeigt, das bei nicht -klimatisch verursachten Reaktionen Baume, z. B. Folge Durchforstungen, zu hohen unechten fuhren kann. Aus diesem Grunde sowie wegen seiner Fahigkeit Beseitigung negativer sehr schwach positiver erscheint neue geeigneter fur Wiedergabe Wachstumstrends. Des tests de la fonction dite "compound function" introduite par WARREN comme un moyen pour eliminer tendance croissance presente dans les donnees dendrochronologiques sont decrits ciapres. En particulier, interet intrasite residus generes nouvelle methode comparees avec celles fournies lissages exponentiels classiques. Il est egalement montre qu'en reactions induites le climat, cela peut etre cas eclaircie, peuvent conduire hautes qui erronees. consequence, et parce que cette elimine negatives ainsi positions tres faibles, elle apparait plus satisfaisante representer croissance. INTRODUCTION A approach ring width data has recently been described senior author (Warren 1980). brief, assumes that jth year record represented by: ( 0, x' ti yj = Sijai(xj ti)bi exp-ci(xj-ti)], Sii J i=0 1, xi > (1) compounding supposed functions. The denote times, relative start record, "pulse" added trend; <_ 0 thus refers age tree record. 56 AND MACWILLIAM estimating number parameters model their values. time base, N, selected basic function transformed to: fn(y) + bPn (x t) -c(x fitted first N years fit extrapolated next years. If points do not fall significantly above extrapolation, these previous set over enlarged range. judged curve, same form, original, indicated (1). possibly curve then process repeated. Certain difficulties have overcome; full details given aforementioned paper. stated therein fittingsystem parameters, opposed believed most critical. Application total 10 cores trees three sites belonging single watershed led selection 30 suitable compromise, at least this specific material. We now report results applying method, 30, all 73 obtained sites. was previously remarked 1980) heuristically, stronger cross correlation residuals, or indices, between cores, likely associated common environmental changes, particular climatic fluctuations. define residual observed minus calculated width, ie. yobs -ycal index value taken /ycal latter seems preferable since its variance will tend stable. what follows, values used unless otherwise specified. Thus, correlation, one successful course our investigation, we found cause question conjecture determined there circumstances where necessarily true, arise. One section is, therefore, devoted aspect. DATA designated PR, CC VB. Of available 25 each PR remaining 23 estimated option computer program developed Fritts al. 1969. stemming fitting were matrices computed site. terminated 1976 ranged length 107 257 (with but 187 years), 227 67 167 Accordingly based on For CC, core effectively zero other under either method. consequently incorrectly dated was, excluded further calculations. There remain 72 which 24 CC. 24x25/2 300 coefficients 276 253 Removing Growth Trend 57 RESULTS must remembered any site, mutually independent. inferences can, however, drawn. averages deviations are: VB Exp. 0.519 (0.170) 0.481 (0.171) 0.410 (0.195) New 0.480 (0.133) 0.444 (0.112) 0.450 (0.153) (standard parentheses) two methods quite close, differing less than 10% derived are, noticeably consistently less, about %, histograms statistics typified Figure 1 Those somewhat longer tail left