Study site
The study was conducted on the northwestern slope of the lateral moraine of the SchwarzenbergerSeespitze glacier (Stubaier Alpen, Tyrol, Austria; 11 °06′E, 47 °03′N) at an altitude between 2630 and 2680 m a. s. l. The area is characterized by a cold temperate, continental climate with about 0 and −5·5 °C average annual temperature and annual precipitation of 800–900 and 1600–1700 mm in 2000 and 3000 m a. s. l. respectively (Klimadaten von Österreich 1971–2000). Siliceous bedrock predominates in the surrounding mountains, hence the substrate of the moraine consists of unsorted, siliceous debris with a high proportion of smallsized gravel and no organic soil layer. Extending over a horizontal distance of less than 100 m, the site covers a continuous gradient from a basinshaped depression with very high snow accumulation to an upper slope with an average snow cover.
Data collection
Fourteen data loggers (Stowaway Tidbit TBI3220 + 50, Onset Corporation, Bourne, MA, USA, range −20 °C to +50 °C) were placed along a snowmelt gradient on the moraine slope. The loggers were buried 5 cm below the soil surface, the layer of the most compact rooting. This approach prevents direct sun insolation and disturbance from tourists or avalanches. From October 2001 to October 2003 measurements were taken at 60minute intervals (i.e. 24 measurements per day and logger).
Ten abundant species representative of high alpine habitats of the central Alps were selected for phenological observations (see Tab 1). Nomenclature and habitat affiliations follow Adler, Oswald & Fischer (1994). For each species, 90 adult individuals (86 for Saxifraga bryoides) in close proximity to the dataloggers (max. 2·5 m horizontal and 0·5 m altitudinal distance) were tagged for monitoring. Observations were made at c. 10day intervals during the 2002 and 2003 growing seasons. This resulted in 9 sets of observations for 2002 and 10 sets for 2003. The growing season in 2002 was characterized by average temperatures (ZAMG 2002) while the summer in 2003 was exceptionally warm (ZAMG 2003). During each observation newly developed generative shoots were marked with coloured wool threads. Phenological phases of each shoot were recorded, except for Sedum alpestre for which we registered the phenological status of each flower. Phenological development of generative shoots within a year was classified in five to eight phases depending on the species. Easily distinguishable morphological traits were used to define these phases (Tab 2). For the 30 (i.e. 3·3%) individuals which died in the winter of 2002, we measured its nearest neighbour of the same species in the growing season of 2003.
Table 1. Ecological requirements of study species. TS represents predictions of timetoevent models representing the thermal demands (mean ± standard deviation) of individuals – given as accumulated temperature sums above 1 °C – to reach flowering. Habitat requirements are extracted from Adler, Oswald & Fischer (1994)Species  Habitat  Pollination  TS 


Agrostis rupestris (Poaceae)  Swards  Anemophilous  11678 ± 437 
Cardamine resedifolia (Brassicaceae)  Snowbeds, moist rocks, open soil  Entomophilous  2328 ± 791 
Gnaphalium supinum (Asteraceae)  Snowbeds, moist swards  Entomophilous†  6894 ± 289 
Leucanthemopsis alpina (Asteraceae)  Snowbeds, swards, scree  Entomophilous  6905 ± 649 
Oxyria digyna (Polygonaceae)  Swards, scree  Anemophilous  4432 ± 2344 
Poa alpina (Poaceae)*  Scree, nutrient rich swards and pastures  Anemophilous  10427 ± 632 
Poa laxa (Poaceae)  Windexposed rocks, scree  Anemophilous  9111 ± 407 
Sedum alpestre (Crassulaceae)  Snowbeds, rocks, scree  Entomophilous  4679 ± 627 
Saxifraga bryoides (Saxifragaceae)  Windexposed rocks, scree  Entomophilous  11260 ± 2281 
Veronica alpina (Scrophulariaceae)  Snowbeds, moist swards, richly manured pastures  Entomophilous  7009 ± 63 
Table 2. Description of distinctive marks used to morphologically differentiate phenophases of ten high alpine plant species. Phenophases given in bold mark the anthesis of the species (compare Table 4) Species  Phenophases 

Agrostis rupestris  Panicle visible – panicle elongated –flowers– anthers dry – caryopsis propagating 
Cardamine resedifolia  Buds –flowers– fruit with corolla – fruit, corolla dropped off – pod dehiscent 
Gnaphalium supinum  Buds covered with leaves – buds visible –flowers– corolla shrivelled – involucre turned yellow – fruits dehiscent 
Leucanthemopsis alpina  Buds sessil – buds pedicellate – ligulate flowers visible –1. tubular flower open– all tubular flowers open – tubular flowers dark coloured – fruits immature – fruits dehiscent 
Oxyria digyna  Buds –flowers– stigmas emerged – fruits winged – tubercles brownish – fruits dry 
Poa alpina_{(flowering)}  Panicle visible – panicle elongated –flowers– postanthesis – caryopsis propagating – dry panicle 
Poa alpina_{(pseudoviviparous)}  Panicle visible – panicle elongated – pseudobulbs kneeled – spikelet dry with bulbs – spikelet dry without bulbs 
Poa laxa  Panicle visible – panicle elongated –flowers– anthers dry – glumes dry – caryopsis propagating 
Sedum alpestre  Buds –corolla bright yellow– corolla darkened – fruits immature – fruits dehiscent 
Saxifraga bryoides  Buds – corolla closed –corolla opened– carpels red – corolla dropped off – dispersal 
Veronica alpina  Blue coloured bracts – corolla visible –corolla opened– fruit < calxy – fruit => calyx – fruit dehiscent 
The study area was enclosed with an electric fence to deter large herbivores from grazing.
Data preprocessing
We checked for deviations among the 14 data loggers by storing them together at room temperature for several days before and after deployment in the field. Measurements taken in the field were corrected according to differences among the loggers during that time.
Three environmental variables were used to analyse flowering phenology of study species. Two of them were derived from the raw temperature data stored by the data loggers:
 (i)
Temperature sums (TS) were calculated from the first of April until each of the nine or ten observation dates for each logger in 2002 and 2003, respectively. As alpine plants may be photosynthetically active during periods of low temperature, we used a threshold value of 1 °C. The measurement values exceeding this threshold were summed. Calculations using other threshold temperatures (from 0 °C to 10 °C) revealed very similar results but did not fit as well.
 (ii)
The time since snowmelt (SM) is given as the number of days from the start of the growing season to each observation where the ground above the logger was snow free. Ground was determined to be snow covered if data loggers indicated temperatures ranging from −0·5 to 0·5 °C and temperature amplitudes (variations) were low. The daily temperature amplitudes were analysed using a moving window of 3 days. Measurements from the first of April – the first measurement without snow cover was taken in June – to each observation without snow cover were counted for each logger and divided by 24 (i.e. the number of measurements each day).
 (iii)
The third variable photoperiod (PH) was defined as minutes of daylight without snow cover at the day of observation.
Each plant was assigned individually to the nearest datalogger and included with its TS, SM and PH values in the following analyses.
Data analysis
Plants without generative shoots and obviously diseased or damaged shoots were excluded from analyses.
To examine the synchronizing effect of the environmental factors TS, SM and PH on plant phenology, parametric timetoevent models for interval censored data (Hosmer & Lemeshow 1999; Klein & Möschberger 2003) were used. For each phenophase of each species in each of the two observation years models with TS, SM and PH as response variables were fitted as null models, i.e. we modelled the probability that an individual would not have reached a specific phenophase above a certain TS, SM or PH.
where f() is a link function depending on the assumed standard distribution of responses, T is either TS, SM or PH, β_{0} is the intercept, σ is the scale parameter and ε is the error. To account for random effects of individuals a penalized variable with an assumed gamma distribution (frailty term) was included in the models. We explored models with ten standard distributions: minimum extreme value, Weibull, normal, lognormal, logistic, loglogistic, exponential, logexponential, Rayleigh and logRayleigh. The model with the smallest Akaike information criterion (AIC; Sakamoto, Ishiguro & Kitagawa 1986) – calculated as −2 × loglikelihood + 2 × n, where n is the number of parameters in the model – was regarded as the model with the best fit.
Applying the timetoevent models, the median ‘time’ [expressed in accumulated °C for temperature sums (TS), number of days since snowmelt for SM and daylight minutes for PH] to reach the phenological phases was predicted for each individual in each year. These individual predictions were used to assess how strongly the respective climatic factor synchronizes the phenological development of species. We assumed that an environmental factor acting as a strong trigger will lead to very similar predicted values for all individuals of a species in both years, whereas factors with only weak impacts on phenology will result in greatly varying individual median ‘times’ between years and along the snowmelt gradient. To quantify the variance of individual predictions the standard deviation (SD) was calculated for each phase and year. High SDvalues indicate a weak or missing impact of the respective factor, whereas a SD of zero represents perfect synchronization. The magnitude of SD depends on the absolute values of individual predictions. Hence, to allow for comparisons among the three environmental factors, which have different units, as well as among phenophases within one factor, where late phenophases have higher values than early ones (i.e. in both cases large differences in absolute values can be expected), predictions of each timetoevent model were standardized to cover a range from zero to one by using the formula (x−x_{min})/(x_{max}−x_{min}), where x is a value of the environmental variable and x_{min} and x_{max} are the minimum and maximum of these variable occurring in the data for the respective model.
Linear mixedeffects models (Laird & Ware 1982) were developed with the following parameters (unless explicitly specified): we assumed a gaussian error distribution, a maximum likelihood algorithm was chosen to approximate the loglikelihood criterion of parameter estimation, withingroup errors were allowed to have unequal variances, the potential nonindependence of SDs within species as well as between years was accounted for by using both variables as group levels in the calculation of random effects, we allowed for random intercepts as well as random effects of these groups for each fixed effect (i.e. the fixed effect of a model was also used as random effect).
To compare the synchronizing effect of environmental factors on flowering phenology SDs as response, and TS, SM and PH were used as levels of the fixedeffects variable. To investigate the reliability of SDs derived from TS, SM and PH, comparisons between the exceptionally warm year 2003 and the average year 2002 were carried out by applying a linear mixedeffects model for each environmental factor using the year of observation as a binomial predictor and allowing for random intercept for species, which were used as the grouping variable.
Another model examined the intraseasonal trends of the effect of temperature by using second order polynomials of mean temperature sums of the phenophases as fixed effects. The significance of the first order polynomial indicates a linear increasing or decreasing impact of temperature as a trigger of plant development during the growing season. A significant second order term indicates a maximum or minimum of impact of temperature in midseason.
Climatic conditions of the previous summer may influence the phenological development due to a varying degree of maturation of the preformed buds. Hence, the accumulated temperature of the total growing season of 2002 was used to uncover such carryover effects from 2002 to 2003. The respective TS of each individual for 2002 was added to that of each prefloral phenophase in 2003 before applying the timetoevent analysis.
PH can be expected to have a stronger influence on species in habitats with early melting snow cover compared to snowbed species, which should be triggered by energy input. Accordingly, the five snowbed species were compared to the five species inhabiting other habitats like swards, scree or rocks (compare Table 1).
For each mixedeffects model the total number of observations (n_{obs}), the number of groups (n_{gr}), the denominator degrees of freedom, the ttest statistic and the associated Pvalue are given for the respective fixed effect.