Spatiotemporal Decomposition of Greenhouse Gas Flux Partitioning via Bayesian Eddy Covariance Inversion Coupled with High-Resolution Land Surface Parameterization

Abstract

Accurate quantification of terrestrial greenhouse gas (GHG) fluxes remains a critical challenge in understanding carbon cycle feedbacks under accelerating climate perturbation. This study presents a novel Bayesian eddy covariance inversion framework integrated with high-resolution land surface parameterization (BECI-LSP) to decompose net ecosystem exchange (NEE) into gross primary productivity (GPP) and ecosystem respiration (Reco) across heterogeneous biomes. Flux tower observations from 14 FLUXNET sites spanning boreal, temperate, and tropical ecosystems were assimilated using Markov Chain Monte Carlo sampling with adaptive hyperprior regularization. The BECI-LSP framework reduced flux partitioning uncertainty by 34.7% relative to conventional nighttime regression methods, demonstrating superior performance in capturing diurnal and seasonal NEE variability. Results indicate a statistically significant underestimation of tropical GPP in legacy models, with implications for IPCC AR7 carbon budget reconciliation.