Non-symmetric responses of leaf onset date to natural warming and cooling in northern ecosystems

Abstract

The northern hemisphere has experienced regional cooling, especially during the global warming hiatus (1998-2012) due to ocean energy redistribution. However, the lack of studies about the natural cooling effects hampers our understanding of vegetation responses to climate change. Using 15,125 ground phenological time series at 3,620 sites since the 1950s and 31-year satellite greenness observations (1982-2012) covering the warming hiatus period, we show a stronger response of leaf onset date (LOD) to natural cooling than to warming, i.e. the delay of LOD caused by 1°C cooling is larger than the advance of LOD with 1°C warming. This might be because cooling leads to larger chilling accumulation and heating requirements for leaf onset, but this non-symmetric LOD response is partially offset by warming-related drying. Moreover, spring greening magnitude, in terms of satellite-based greenness and productivity, is more sensitive to LOD changes in the warming area than in the cooling. These results highlight the importance of considering non-symmetric responses of spring greening to warming and cooling when predicting vegetation-climate feedbacks.

Fig. 1.

The distribution of PEP725 sites and comparisons of LOD responses to warming and cooling at the species scale. A) The locations of PEP725 sites for long-term in situ LOD observations of seven European dominant tree species since the 1950s. B) LOD responses (log-transfer, see Methods) to warming and cooling at the species scale from PEP725 ground observation data. The warming and cooling samples were obtained by using P < 0.05 for temperature changes and partial correlation analysis. The bar represents the standard error. Student's t-test was used to test the significance of the difference between the warming and cooling conditions. Significance code for differences: ***P < 0.001, **P < 0.01, and *P < 0.05.

Fig. 2.

Warming and cooling grid cells used in this study and comparisons of LOD responses to warming and cooling at the biome scale. A) Locations of warming and cooling grid cells during the warming hiatus (1998–2012). Grid cells with P < 0.05 for temperature changes and partial correlation analysis were retained (see Methods). B) The biome types of grid cells. C) Trends in LOD in warming and cooling areas for biomes from 1998 to 2012. D) LOD responses (log-transfer) to warming and cooling at the biome scale from satellite-based LOD data. The bar represents the standard error. Student's t-test was used to test the significance of the difference between the warming and cooling conditions. Significance code for differences: ***P < 0.001, **P < 0.01, and *P < 0.05.

Fig. 3.

Comparisons of changes for chilling accumulation (CA), heat requirement (HR), and water availability indicator (i.e. SPEI) in warming and cooling areas obtained from satellite-based analysis during the warming hiatus. A) Relationship between CA and HR. The green line indicates an exponential decay regression fitted using CA and HR. B, C, and D) present the trends of CA, HR, and SPEI in the warming and cooling areas during 1998–2012, respectively. The bar represents the standard error. Student's t-test was used to test the significance of the difference in the absolute values of trends for CA, HR, and SPEI in the warming and cooling areas. Significance code for differences: ***P < 0.001 and **P < 0.01.

Fig. 4.

Connections among temperature change, LOD, and spring greening magnitude. The comparison for regressions between spring GPP/NDVI and LOD in the warming and cooling areas using the satellite-based NIRv GPP A), the LRF GPP B), TL-LUE GPP C), and GIMMS NDVI D), respectively. The spring-accumulated GPP and mean NDVI were calculated during the period from March to May. The red and blue kernel density plots represent the density distribution of warming and cooling grids in GPP/NDVI-LOD space, respectively. The four GPP/NDVI datasets all showed that the differences in the slopes between warming and cooling conditions were significant (P < 0.001) by using covariance analysis. Significance code for differences: ***P < 0.001.