Transformer-Driven Simulation of Global Land-Atmosphere Interactions and Its Climate Feedback Mechanisms

Abstract:

Land-atmosphere interactions (LAI) are critical processes in the Earth system, regulating energy, water, and carbon cycles and exerting significant feedback effects on regional and global climate. Traditional Earth system models (ESMs) rely on simplified parameterization schemes for LAI processes, leading to notable uncertainties in simulating surface energy balance and climate feedback. This study proposes a Transformer-based global LAI simulation framework (Trans-LAI) that integrates self-attention mechanisms with physical constraints to capture the spatiotemporal dependencies and nonlinear interactions in LAI processes. The framework assimilates multi-source observation data, including satellite-derived land surface temperature (LST), latent heat flux (LE), sensible heat flux (H), and in-situ flux tower data, to optimize the representation of key LAI processes such as vegetation-atmosphere water vapor exchange and surface energy partitioning. Validation results based on 200 global flux tower sites show that the Trans-LAI framework improves the simulation accuracy of LE and H by 18% and 21% respectively compared to the traditional Community Land Model (CLM5.0). Climate feedback simulation under CMIP6 SSP3-7.0 and SSP5-8.5 scenarios indicates that the enhanced LAI simulation reduces the uncertainty of global mean surface temperature projection by 12-15% by the end of the 21st century. Specifically, in tropical rainforest regions, the Trans-LAI framework captures the negative climate feedback effect of enhanced evapotranspiration under warming conditions, while in arid and semi-arid regions, it accurately simulates the positive feedback effect of reduced vegetation coverage on drought intensification. This study provides a new paradigm for improving the simulation accuracy of LAI processes in ESMs and enhances the reliability of climate change projection, offering important scientific support for formulating climate adaptation strategies.