Techniques for Reconstructing Data on Flowering

摘要: One of the objectives applied phenological inves-tigations is to reconstruct, for various species, historicalseries data on same biological phenomenon ofinterest, such as budding, raceme formation, floral buddevelopment, beginning flowering, full flower-ing, end fruit setting, and develop-ment fruits (Efron, 1994; Liu, 1995; Heitjan, 1997).Phenology offers a good tool studying plants and, inparticular, interactions between climate variation andplant behavior (Chuine, 1999). However, in order useconventional statistical techniques, complete series ofphenological collected over many years are neces-sary.The purpose this work was test different meth-ods reconstructing missing comparetheir reliability. In past, were usu-ally substituted by average value recordeddata. Alternatively, species with sameperiods flowering used (Bricchi et al. , 1995),and results obtained method good, eventhough more variable (Bassi 1996; Cenci ,1997). We compared three techniques reconstruct-ing data—smoothing, similar split-plot design,and correlation technique—using 20-yearphenological observations 50 plant species.MATERIAL AND METHODSThe database study contained phenolog-ical 153 growing at Gui-donia (Rome, 42 ° N) surrounding area, whichwere continuously performed Montelucci, an Italianresearcher, from 1960 1982. some valueswere historical manyspecies.The set analysis concerned forwhich onset recorded each year.This date expressed number days begin-ning first day year. For species,three points corresponding 20-year wereassumed be reconstructed. These werethree dates flowering: earliest, latest,and medium (closest corre-sponding time series). To estimate possible influ-ence temporal type ofreconstruction, all arrangedinto groups comprising early onsets (theminimum values series), late (the max-imum values), closest toaverage series.The methods smoothingtechnique, univariate miss-ing one time. It differs theother two methods, which, or less directly, use aseries related (Simonoff, 1996).The smoothing technique flexible: it allows thearbitrary choice mathematical functions that adaptwell data. Both linear exponential functionscan used. particular, inthis simulation reconstruction depends only theresults terms deviation truevalue.The basic formulas calculating univariatesmoothing six mathemati-cal functions, four exponential. Thechoice function require-ment “weighting” calculations andon their number. A greater “weight” can attributed tothe year either farthest it. considered periods 3and 5 before/after date.The second split-plotdesign. This least-squares theentire reference (153 species), minimizing thedeviation andother species.The third based therelationship (correlation) species. The spe-cies regarded correlating other if theyflowered approximately ( ± 30 days).Statistically significant correlations cal-culating regressions, arithmetic means