Enhanced understanding of the terrestrial carbon cycle through multiple constraints in model-data-integration approaches

Abstract:

The representation of exchanges of carbon, water and energy between the land surface and the atmosphere still reveals significant model limitations in explaining temporal and spatial variability. Despite agreement between models for contemporaneous periods, prognostic simulations reveal a strong between-model divergence regarding the role of the land surface in the global carbon cycle. The integration of multiple data-streams in inverse modelling approaches for parameterization and model evaluation, ultimately leads to model improvement. Here we explore multiple-constraint approaches ranging from in situ to regional and global spatial scales. Constraints include stocks and fluxes of water and carbon. We show that integrating multiple datasets contributes to a better representation of ecosystem dynamics in different models, from forest and dynamic vegetation models to land surface schemes. At site scale, model-data comparisons reveal substantial differences in the modelled temporal dynamics of carbon stocks and turnover times and their relationships with climate, especially at annual scales. Inter-annual variability remains a problem for all models, even after parameter optimization. At regional and global scales, the integration of multiple data-streams to constrain albedo, phenology and primary productivity patterns yields a significant improvement in regional simulations of vegetation dynamics, from seasons to longer-term trends. The role of environmental controls and vegetation dynamics in explaining recent trends in the amplitude of the seasonal cycle of atmospheric CO₂ is evaluated using an improved dynamic vegetation model. We conclude by identifying major challenges in model-data-integration: to explore the information content in longer time series; avoid confounding effects of missing processes on parameter estimation; set up cost functions for multivariate-data integration; quantification of uncertainties arising from data bias, model structure, and initial conditions; and test modelling concepts (e.g. PFTs) to guide development of new principles. These challenges emphasize the importance and value of integrating multivariate/multi-temporal information in Earth System models for an enhanced understanding and description of the terrestrial carbon cycle.