Text S1 Additional information on the CMIP6 models

We obtained daily maximum and minimum surface air temperature (tasmax and tasmin) and leaf area index (LAI) from participating models in the CMIP6 ensemble (https://esgf-node.llnl.gov/search/cmip6/) from both historical simulations from the core Diagnostic, Evaluation and Characterization of Klima (DECK) experiments and SSP5-8.5 scenario simulations from the Scenario Model Intercomparison Project. The historical simulations run from 1850-2014 and are forced with estimates of natural (e.g., volcanic eruptions, solar variability) and anthropogenic (e.g., greenhouse gas emissions, land-use change) climate forcing. The Shared Socioeconomic Pathways (SSPs) present alternative scenarios of future emissions and land-use changes estimated using integrated assessment models and run from 2015–2100 except for long-term extensions. Table S1 lists model information for all 26 CMIP6 models used, and Table S2 describes the phenology schemes of the models with daily LAI and prognostic phenology. Although we calculated SI-x for up to 30 members for models with multiple ensemble members, we only included the 1st member (listed in Table S1) in the multi-model analysis (Hagedorn et al., 2005; Lehner et al., 2020). Although members of the same model could experience disagreements on the spring onset trends even at a hemispheric scale due to internal climate variability (Figure S7), they agree on mean spring onset timing (Figure S2) and experience a smaller spread of trend variability at a longer timescale (i.e., compared to the 1979-2014 means in Figure S1, the spread of mean trends of the members from the same model is smaller over the 1950-2014 period).
When phenology is simulated prognostically through biogeochemical processes, LSMs calculate LAI based on carbon allocation to the leaf carbon pool (e.g. Krinner et al., 2005; Lawrence et al., 2019). The start and end of the growing season can be simulated explicitly and determined by environmental factors such as surface air temperature, soil moisture, and daylength (Krinner et al., 2005; Lawrence et al., 2019; Sitch et al., 2003). For instance, in the Community Land Model version 4.5 (CLM4.5, the land surface model used in CMCC-ESM2) and the Organising Carbon and Hydrology In Dynamic Ecosystems v2.0 (ORCHIDEE v2.0, the land surface model of IPSL-CM6A-LR), for unmanaged deciduous ecosystems, onset is explicitly regulated by a growing degree day (GDD) threshold for seasonal deciduous plants (plus chilling in ORCHIDEE) whereas combined thermal, chilling and water stress criteria are required for moisture-limited plants (Botta et al., 2000; Krinner et al., 2005; Lawrence et al., 2019; White et al., 1997). Alternatively, phenology can be modeled implicitly, where onset and offset of the growing season are not explicitly regulated and LAI follows variations in leaf biomass (e.g., as in Surfex 8.0c, the LSM of CNRM-ESM2-1; Hamdi et al., 2014). In addition, some models prescribe phenology to simulate the carbon cycle (e.g., as in CABLE2.4, the LSM of ACCESS-ESM1-5; Law et al., 2017; see Tables S1-S2 for more details).