Leaf area index in Earth system models: how the key variable of vegetation seasonality works in climate projections

Abstract

Earth system models (ESMs) are widely used in scientific research to understand the responses of various components of Earth systems to natural and anthropogenic forcings. ESMs embody terrestrial ecosystems on the basis of the leaf area index (LAI) to formulate various interactions between the land surface and atmosphere. Here, we evaluated the LAI seasonality of deciduous forests simulated by 14 ESMs participating in the Coupled Model Intercomparison Project Phase 5 (CMIP5) and CMIP6 to understand the efficacy of recent ESMs in describing leaf dynamics in the northern extratropics from 1982 to 2014. We examined three indicators of LAI seasonality (annual mean, amplitude, and phase) and three phenological dates (start (SOS), end (EOS), and length of growing season (LOS)) of the models in comparison to the third-generation LAI of Global Inventory Modeling and Mapping Studies (GIMMS LAI3g) and the Climate Research Unit gridded time series dataset. CMIP6 models tend to simulate larger annual means (1.7 m2 m−2), weaker amplitudes (0.9 m2 m−2), and delayed phases (226 DOY) compared to the GIMMS LAI3g (1.2 m2 m−2, 1.2 m2 m−2, and 212 DOY, respectively), yet are similar to the CMIP5 models (2.2 m2 m−2, 1.0 m2 m−2, and 225 DOY). The later phase is attributed to a systematic positive bias in EOS of the CMIP5 and CMIP6 models (later by 22 and 18 d, respectively) compared to the GIMMS LAI3g (261 DOY). Further tests on phenological responses to seasonal temperature revealed that the majority of CMIP5 and CMIP6 ESMs inaccurately describe the sensitivities of SOS and EOS to seasonal temperature and the recent changes in mean SOS and EOS distributions (2005–2014 minus 1982–1991). This study suggests that phenology schemes of deciduous forests, especially for autumn leaf senescence, should be revisited to achieve an accurate representation of terrestrial ecosystems and their interactions.