Leaf onset in the northern hemisphere triggered by daytime temperature

Abstract

Recent warming significantly advanced leaf onset in the northern hemisphere. This signal cannot be accurately reproduced by current models parameterized by daily mean temperature (T(mean)). Here using in situ observations of leaf unfolding dates (LUDs) in Europe and the United States, we show that the interannual anomalies of LUD during 1982-2011 are triggered by daytime (Tmax) more than by nighttime temperature (T(min)). Furthermore, an increase of 1 °C in Tmax would advance LUD by 4.7 days in Europe and 4.3 days in the United States, more than the conventional temperature sensitivity estimated from T(mean). The triggering role of Tmax, rather than the T(min) or T(mean) variable, is also supported by analysis of the large-scale patterns of satellite-derived vegetation green-up in spring in the northern hemisphere (>30 °N). Our results suggest a new conceptual framework of leaf onset using daytime temperature to improve the performance of phenology modules in current Earth system models.

Figure 1. Responses of in situ-observed LUDs to T_max and T_min in Europe and the United States during 1982–2011.

The frequency distributions of the length (in months) of T_max preseason in (a) Europe and (d) the United States are shown. The T_max preseason is defined as the period with the highest negative partial correlation between LUD and averaged T_max for the months preceding LUD. Frequency distributions of the highest partial-correlation coefficients between LUDs and preseason T_max in (b) Europe and (e) the United States after controlling for corresponding T_min, cloudiness and precipitation. Frequency distributions of partial-correlation coefficients between LUD and T_min in (c) Europe and (f) the United States during the same preseason as in a after controlling for corresponding T_max, cloudiness and precipitation. Note that LUDs of multiple species in Europe and only lilacs (Syringa L.) in the United States were analysed. The mean values of partial-correlation coefficients across all phenological stations, the percentages of significantly negative partial correlations and the percentages of significantly positive partial correlations (in parentheses) are provided in b,c,e and f.

Figure 2. The relationship of the satellite-derived onset dates of vegetation green-up with T_max and T_min in the northern hemisphere during 1982–2011.

(a) The spatial pattern of the length (in months) of the preseason defined as the period with the highest negative partial correlation between VGD and averaged T_max for the months preceding VGD. (b) The frequency distribution of the length of the preseason shown in a. (c) Partial-correlation coefficients (R) between preseason T_max and VGD after controlling for corresponding T_min, cloudiness and precipitation. (d) Partial-correlation coefficients (R) between T_min and VGD during the T_max-derived preseason after controlling for corresponding T_max, cloudiness and precipitation. The 1% and 5% significance levels of the partial correlations correspond to ±0.49 and ±0.38, respectively.

Figure 3. The sensitivity of LUD and VGD to T_max and T_mean during 1982–2011.

The frequency distributions of the temperature sensitivity of LUD to (a) T_max (SVT_max) and (b) T_mean (SVT_mean) and (c) the ratio between SVT_max and SVT_mean (SVT_max/SVT_mean) in Europe and the United States. In the right panel of figure, the spatial distributions of the sensitivity of VGD are shown (d) T_max (SVT_max) and (e) T_mean (SVT_mean) and (f) the ratio between SVT_max and SVT_mean (SVT_max/SVT_mean) in the northern hemisphere during 1982–2011.

Figure 4. The ratios of future VGD changes predicted by a T_max-based GDD concept model to that predicted by a T_mean-based GDD concept model.

Both T_mean-based GDD approaches and T_max-based GDD models were applied to predict the VGD changes (▵VGD) between 1991–2010 and 2081–2100, using 24 climate models and different climate change scenarios (RCP2.6, RCP4.5 and RCP8.5). For each RCP, the T_max-based predictions and the T_mean-based predictions were averaged across all models and the distributions of their ratio (T_max-based predictions/T_mean-based predictions) are shown in (a), (b) and (c). The ratio <1 (blue bar) represents that the future VGD changes predicted by T_mean-based approaches are larger than those predicted by T_max-based approaches and vice versa (red bar). The percentage of ratios <1 and the percentage of ratios >1 are both provided in a,b and c.