Modeling Carbon Emissions Driven by Land-Use Change: Uncertainty and Sensitivity Analysis of Dynamic Global Vegetation Models

Abstract

As a key component of the climate system, terrestrial carbon dynamics are commonly simulated using Dynamic Global Vegetation Models (DGVMs), which provide essential information for estimating the global carbon budget. However, substantial uncertainties persist in DGVM-based estimates of land-use change carbon emissions, and their process-level origins remain poorly constrained. Here, we harmonize DGVM-based net land-use emissions (E_NLUC) from 23 TRENDY2025 DGVMs using a transient, time-varying PFT-level land-cover correction that explicitly removes the loss of additional sink capacity (LASC), and apply a hierarchical Bayesian mixed-effects variance decomposition framework to characterize between-DGVM heterogeneity in fitted coefficients associated with land-use predictors. LASC correction reveals that uncorrected DGVMs artificially converge toward bookkeeping estimates, masking persistent structural deficiencies—most notably the limited representation of wood harvest, forest management, and other peat-related activities incorporated in bookkeeping models. Between-DGVM heterogeneity in fitted coefficients is greater for persistent regrowth area than for current deforestation, particularly at decadal-to-centennial timescales and in the tropics, where the fitted association with persistent regrowth area varies substantially among models. This uncertainty can propagate into residual land sink estimates, complicating the diagnosis of land–atmosphere carbon exchange. Our results highlight that improving the representation of secondary forest recovery, including sub-grid land-use processes and forest management, is critical for reducing uncertainties in land-use and land-use change emissions in next-generation DGVMs.