[A-CG35] Global Carbon Cycle Observation and Analysis

[ACG35-P04] Assessment of a process-based terrestrial carbon cycle model, VISIT, using multiple data-driven estimations

*Ruci Wang¹, Kazuhito Ichii¹, Prabir Patra², Naveen Chandra² (1.Center for Environmrntal Remote Sensing, Chiba University, 2.Japan Agency for Marine-Earth Science and Technology)

Keywords:Terrestrial carbon cycle, Remote sensing, Top-down estimation, Bottom-up estimation, VISIT model

Terrestrial carbon cycle is one of the most uncertain components in global carbon cycle. Therefore, evaluation and refinement of terrestrial carbon cycle models using multiple observation data are necessary. In this study, we employed a terrestrial carbon cycle model, VISIT, which simulates the carbon, water, and nutrient cycles of terrestrial ecosystems, including vegetation and soil, and assessed the fluxes of greenhouse gases between the atmosphere and terrestrial ecosystems. The data-driven estimation of terrestrial carbon cycles include machine-learning based gross primary production (GPP) and net ecosystem production (NEP), atmospheric inversion based atmosphere-land carbon exchanges, satellite data based leaf area index (LAI) and biomass, and inventory-based soil carbon. We evaluated model outputs with these data-driven estimates for both global scale and sub-continental scale (e.g. 11 zones). The findings demonstrate that there is a strong consistency in the monthly changes between the VISIT model and machine-learning based GPP from 2001 to 2011, particularly in Russia, Europe, and South Asia. We will show evaluation results of VISIT model outputs with more data-driven products. This study provide a baseline of the current model evaluation, and we will discuss the way of model-data fusion and assimilation.

However, when comparing VISIT and SVRGLOBAL for long-term outcomes, only South Asia and Europe exhibit comparable GPP outcomes. The most probable reason is that the model structure and hypotheses are uncertain, as well as the parameter certainty brought on by parameter interactions and model over-parameterization.