Spring onset timing
Spring onset timing has advanced in recent decades in both observations and model simulations and will shift earlier under the SSP585 scenario (Figs. 1–3). Over 1950–2014 in the Northern Hemisphere (NH), CMIP6 ensemble mean of 26 models indicates 117.0 ± 6.0 days to first leaf (hereafter CMIP6-leaf, see Methods), later than SI-x leaf based on Berkeley Earth (hereafter Obs-Berkeley-leaf, 107.3 ± 2.2 days; Fig. S1a). These numbers advance by approximately one day during 1981–2014. Over the historical period, the spread of spring onset timing in CMIP6 encompasses the interannual variability of observation-based SI-x (Fig. 1a). Under the SSP585 scenario in 2015–2099, CMIP6-leaf advances to 104.5 ± 8.6 days. The largest disagreement of mean onset dates between CMIP6 models is present in Western NA, Northern Russia, and over the Tibetan Plateau during the historical period, and this spatial pattern persists into the SSP585 period (Figs. 2hj and S2).
Models experience considerable differences in simulated plant phenology (Fig. S3), mean LAI values (Fig. S4), and mean LAI threshold-based spring onset timing (Fig. S1cd, S3, S5a). Compared to GLASS LAI (see Methods), ensemble mean overestimates LAI values in Eastern Asia and western Canada and underestimates LAI in Southern Russia (Fig. S4). Large disagreements exist among the models in Canada, Northern Europe, and East Asia. These regions continue experiencing large differences among model LAIs in SSP585, along with Eastern US.
The start of spring as indicated by LAI 25% threshold DOYs are much later in the models than in GLASS LAI or those indicated by SI-x leaf (Figs. 2, S1, S5). Across the NH, mean LAI 25% DOY based on simulated LAIs (hereafter CMIP6-LAI25%, see Methods) is 137.7 ± 19.7 during 1981–2014, 20 days later than in GLASS LAI (hereafter Obs-GLASS-LAI25%, 117.9 ± 2.6). CMIP6-LAI25% advances to 134.9 ± 24.4 in 2015–2099 (Fig. S1cd). In every period, CMIP6-LAI25% is later than both CMIP6-leaf and Obs-GLASS-LAI25% (Figs. 2cf and S1) and exhibits larger inter-model variability than CMIP6-leaf (Figs. 1c, 2g-j). The timing also varies considerably across models (Figs. 1b, S1cd, S5a). The largest disagreement among models is present south of 40oN while large differences between CMIP6-LAI25% and Obs-GLASS-LAI25% are also present in the Tibetan Plateau, Russia, and Northern Europe. Overall, models estimate later spring onset over high-latitude (north of 55oN) and high-altitude regions (Fig. S6).
Differences in mean spring onset timing vary across models and regions (Figs. 2 and S5) but increase through time. Difference between CMIP6-leaf and CMIP6-LAI25% is 22 days during 1981–2014 and increases to 30 days under the SSP585 scenario (Fig. S1). Across the seven models that provide daily temperature and LAI output, LAI25% is later than SI-x leaf at lower latitudes, though EC-Earth3-CC displays uniformly earlier spring in LAI25% and MIROC-ES2L has earlier LAI indicated spring onset at lower latitudes. Obs-GLASS-LAI25% indicates earlier spring than Obs-Berkeley-leaf in higher latitude and higher elevation regions while model simulations show a reduced advancement in spring onset (Fig. S5c). This pattern persists into the 2015–2099 period with a larger difference between the two sets of indicators. Across the NH, CMIP6-LAI25% exhibits a later spring onset than CMIP6-leaf, with a smaller delay at higher latitudes and higher elevations and a consistent spatial pattern during different simulation periods (Fig. 2cf).
Trends in the start of spring
Long-term trends in the start of spring, as measured by SI-x leaf trends over 1950–2014, exhibit overall agreement between observations and the ensemble mean but vary considerably across models (Figs. S7ab, S8). During 1950–2014 in the NH, the ensemble mean trend of CMIP6-leaf is -0.72 ± 0.21 day/decade, whereas mean Obs-Berkeley-leaf trend is -0.85 day/decade (Fig. S7ab). The magnitude of the trends increases during the more recent period (1981–2014), with CMIP6-leaf mean trends of -1.44 ± 0.4 days/decade, slightly greater than Obs-Berkeley-leaf (-1.28 days/decade) and CPC (Obs-CPC-leaf, -1.21 days/decade). The magnitude of trends increases by ~ 0.9 day/decade under SSP585. During 2015–2099, CMIP6-leaf exhibits an ensemble mean trend of -2.39 ± 0.71 days/decade. However, the trends vary considerably across models and even among different members of the same model (Fig. S7ab).
The spatial pattern of spring advancement exhibits large variations among models and between the ensemble mean and observations. Both Obs-Berkeley-leaf and Obs-CPC-leaf exhibit relatively large advancement in the Tibetan Plateau, North Russia, and Southern Europe during 1981–2014 (Fig. S8), and trends in central and north Asia are statistically significant in Obs-CPC-leaf. This pattern is not present in CMIP6-leaf or any individual model-based SI-x (Fig. S8). Moreover, models exhibit significant trends in very few regions (e.g., Eastern NA and Europe in EC-Earth-3-CC and mid-to-high latitude regions in Asia in GFDL-ESM4). The greatest warming trend in the ensemble mean is in Southern Europe, Eastern US, and the eastern and southern parts of the Rocky Mountains (Fig. 3be), but these regions also exhibit large disagreements across models (Figs. 3hj, S8, S9b). When compared to observations, CMIP6-leaf tends to overestimate warming in NA, particularly Western Canada and Eastern US, and underestimate warming in central and north Asia (Fig. S10). During 2015–2099, most models exhibit significant earlier spring across the NH (Figs. S11b, S12f) and the greatest earlier trend is present in regions along the Pacific coast of NA, in central NA, Europe, and mid-latitude regions of Asia (Fig. 3e). Though models mostly agree that spring is starting earlier under the historical and future scenarios (Figs. S9b, S11b, S12bdf), large disagreements among models are also present, with ACCESS-CM2, CMCC-ESM2, CNRM-CM6-1, CNRM-CM6-1-HR, EC-Earth3-CC, IPSL-CM6A-LR, TaiESM1, and UKESM1-0-LL experiencing a much larger earlier trend while FGOALS-g3, KIOST-ESM, MRI-ESM2-0, and NESM3 exhibiting a much smaller trend (Figs. S8, S10). Compared to the historical period, both the trends and variability of CMIP6-leaf increase in 2015–2099 (Fig. 3behj).
Trends of CMIP6-LAI25% vary considerably geographically and across models, but also display an overall earlier spring onset and increasing variability over the NH (Figs. 3adgi, S7cd, 9a, 11a, 12ace). During 1950–2014, the mean NH trends of CMIP6-LAI25% is -0.41 ± 0.31 day/decade, ~ 0.3 day/decade smaller than those indicated by CMIP6-leaf (Fig. S7). During 1981–2014, trends of both CMIP6-LAI25% (-0.05 ± 0.29) and Obs-GLASS-LAI25% (-0.01) are close to 0, likely due to the delayed start of spring in areas south of 40oN and Southern Europe (Figs. S9a, S11a). The trends change to -0.97 ± 1.06 days/decade during 2015–2099 under the SSP585 scenario, ~ 1.4 days/decade fewer than trends estimated from CMIP6-leaf (Fig. S7). During 1950–2014, the largest trends are present along 60oN in Eurasia and Eastern NA, though trends vary across models in Southern Europe and south of 40oN (Fig. 3g). Though CMIP6-LAI25% mean trends are negative (i.e., becomes earlier) in the US, northern Canada, and southern Russia and positive in southern Europe, Obs-GLASS-LAI25% trends in these regions are of smaller amplitude, and sometimes in the opposite direction (Fig. S9a). Over the 2015–2099 period, the trend is much stronger in mid-to-high latitude (north of 45oN) and high-altitude regions (Figs. 3d, S11a, S12e). In southern Europe and the US, the models diverge in LAI25% responses with large positive trends (delay in spring onset) in CMCC-ESM2, negative trends in EC-Earth3-CC, IPSL-CM6A-LR, and MIROC-ES2L, and contrasting trends in CNRM-ESM2-1.
Although the timing indicated by LAI25% and SI-x leaf vary across models (Fig. S5c), variabilities and trends derived from the two sets of indicators exhibit relatively good agreement, especially during longer temporal periods and in mid-to-high latitude regions (Figs. 3cf, S9cd, S11cd). During 1981–2014, Obs-GLASS-LAI25% exhibits weaker trends in spring onset timing than Obs-Berkeley-leaf at higher latitudes (north of 50oN) and stronger trends at low latitudes (south of 30oN). This latitudinal pattern is also present in most of the models (except KIOST-ESM) with some variations (Fig. S9c). Similar agreement of spring onset trends at mid-to-high latitudes are also present in the ensemble means based on all available models (Figs. 3–4). Although trends of CMIP6- leaf increase faster than CMIP6-LAI25%, their variabilities are coherent and the difference in their trends is still relatively small and spatially uniform at mid-to-high latitudes (Figs. 3, 4, S7, S9, S11, S12). Across the NH, the spread among different indicators is smaller at higher latitudes with the smallest spread present between 55.5oN-84.5oN. However, differences between LAI- and thermal-based indicators increase in the SSP585 scenario across different latitudes, and CMIP6-LAI25% can even be delayed at lower latitudes. At mid-to-high latitudes (north of 40oN except southern Europe), both observations and models exhibit positive and significant correlations between SI-x leaf and LAI25%, except for ACCESS-ESM1-5 in the future scenario and CMCC-ESM2 and KIOST-ESM in some moisture-limited regions (Figs. S9d, S11d). Overall, models exhibit an NH mean correlation of 0.47 ± 0.22 during the historical period and 0.43 ± 0.23 under SSP585 between the two sets of indicators.