Daily Maximum Temperatures Induce Lagged Effects on Leaf Unfolding in Temperate Woody Species Across Large Elevational Gradients

Abstract

The timing of leaf unfolding in temperate woody species is predominantly controlled by the seasonal course of temperature in late winter and early spring. However, quantifying lagged temperature effects on spring phenology is still challenging. Here, we aimed at investigating lagged and potentially non-linear effects of daily maximum temperatures on the probability of leaf unfolding in temperate woody species growing across large elevational gradients. We analyzed 5280 observations of leaf-out time of four tree species (European beech, horse chestnut, European larch, Norway spruce) and one shrub species (common hazel) that were recorded by volunteers over 40 years at 42 locations in Switzerland. We used a case-crossover sampling design to match leaf-out dates with control dates (i.e., dates before or after leaf-out), and analyzed these data with conditional logistic regression accounting for lagged temperature effects over 60 days. Multivariate meta-analyses were used to synthesize lagged temperature and elevational effects on leaf unfolding across multiple phenological stations. Temperature effects on the probability of leaf unfolding were largest at relatively short lags (i.e., within ca. 10 days) and decreased with increasing lags. Short- to mid-term effects (i.e., within ca. 10 to 20 days) were larger for late-leafing species known to be photoperiod-sensitive (beech, Norway spruce). Temperature effects increased for the broadleaved species (horse chestnut, hazel, beech) with decreasing elevation, particularly within ca. 10 to 40 days, i.e., leaf unfolding occurs more rapidly at low elevations for a given daily maximum temperature. Our novel findings provide evidence of cumulative and long-term temperature effects on leaf unfolding, whereby the efficiency of relatively high temperatures to trigger leaf-out becomes higher shortly before bud burst. These lagged associations between temperature and leaf unfolding improve our understanding of phenological responses across temperate woody species with differing ecological requirements that occur along elevational gradients.

Keywords: broadleaved species; conifers; distributed lag models; elevation; lag effects; maximum temperature; multivariate meta-analysis; phenology.

FIGURE 1

Visualization of the time-stratified case-crossover design. The example shows the observed leaf-out time of beech at the phenological station Liestal (350 m a.s.l.; Supplementary Table S1) in Switzerland and T_max (maximum temperature) from July 2006 until end of June 2007. The red dot indicates the case (date of leaf unfolding: 17 April 2007), the black dots indicate the controls (dates before or after leaf unfolding: 3 April 2007, 10 April 2007, and 24 April 2007). The arrows indicate the 60-day lags that are considered in the distributed lag models. April 2007 is delimited by a white shaded box.

FIGURE 2

Visualization of a DLNM (distributed lag non-linear model). The example shows the odds ratio of beech at the phenological station Liestal (350 m a.s.l.; Supplementary Table S1) in Switzerland along T_max (maximum temperature) and lag dimension. A conditional logistic regression model was fitted to time-stratified case-crossover data (Figure 1) from 1972 to 2011. The odds ratios are interpreted as the ratio between the odds of a specific T_max at a specific lag compared to the odds of the reference T_max (0°C) at the same lag (eq. 3). The red line indicates the odds ratio along T_max at a lag of 0 days, the orange lines indicate the odds ratios along the lag dimension at T_max of 0°C (odds ratio = 1), 5°C, 10°C, 15°C, 20°C, and 25°C.

FIGURE 3

Change of T_max (maximum temperature) during the day of leaf unfolding with increasing elevation for larch, horse chestnut, hazel, beech, and Norway spruce. The species are ordered from early-leafing (top) to late-leafing (bottom). The blue lines indicate the regression lines based on the linear models.

FIGURE 4

Summaries of DLNMs (distributed lag non-linear models) based on conditional logistic regression for larch, horse chestnut, hazel, beech, and Norway spruce. The odds ratios along the lag dimension are shown for T_max of 5°C, 10°C, 15°C, and 20°C. The species are ordered from early- to late-leafing species (left to right). The station-specific odds ratios are represented by the dark-blue lines (low-elevation stations) to light-blue lines (high-elevation stations). For sake of clarity, no confidence intervals are shown for the station-specific odds ratios. The red lines are the pooled estimates of the odds ratios (including 95% confidence intervals) from the multivariate meta-analysis. An inverse hyperbolic sine transformation has been applied to the y-axis to increase the visibility of smaller values.

FIGURE 5

Summaries of DLNMs (distributed lag non-linear models) based on conditional logistic regression for larch, horse chestnut, hazel, beech, and Norway spruce. The odds ratios along the lag dimension are shown for T_max of 5°C, 10°C, 15°C, and 20°C. The species are ordered from early- to late-leafing species (left to right). The odds ratios for three elevations (300 m, 700 m, 1100 m a.s.l.) and 95% confidence intervals were estimated from the multivariate meta-regression. An inverse hyperbolic sine transformation has been applied to the y-axis to increase the visibility of smaller values.