Relative Uncertanities in Global SOC Projections in Multiple GCM and RCP

Abstract:

In the previous model inter-comparison studies (e.g., CMIP5), the projected soil organic carbon in ESMs and terrestrial ecosystem models showed the lack of coherence in the end of simulation periods. These uncertainties arise from the different structure of models and different parameters in soil processes and vegetation dynamics. On the other hand, we have different levels of uncertainty sources in the future projection, which are CO₂ emission due to the different social development and judgments to climate change and the different climate patterns derived from different global climate models (GCMs). "Which uncertainty source is more important?" still remained unclear in climate change impacts assessments for terrestrial C cycling. This information could be a one of the criteria for how large uncertainty among terrestrial ecosystem models in projected SOC dynamics is acceptable, because we cannot entirely eliminate uncertainties in the modeling. ISI-MIP project, which was conducted by Potsdam Institute for Climate to assesse climate change impacts on human society and ecosystems, have orthogonal experiment design on RCPs, GCMs, Global terrestrial vegetation models (GVMs) [Warszawski et al., 2014 in PNAS]. In this study, we examined SOC and vegetation C dynamics in six GVMs obtained from ISI-MIP. This experiment enabled us to decompose the relative contributions to total uncertainty of the projection factors using ANOVA as follow;

σ_overall² = σ_RCP² + σ_GCM² + σ_GVM² (+ σ_interactions²)

 In this study, we applied ANOVA to the changes in SOC stock, Veg C stocks, and NPP in the various spatial scale and in the various time periods. At the end of the simulation period (2100), global SOC stock changes from 2000 ranged from $-$195 to 471 Pg-C in the entire simulation set. On the other hand, global Veg C ranged from $-$27 to 543 Pg-C. The results of ANOVA showed that, in global SOC stock changes, the GVM uncertainty dominated throughout the projection period (up to 92% of the total variance). For global Veg C changes, GVMs and RCPs contributed 40% respectively of the total variance by 2100. Thus, the range of uncertainties are similar regarding global C stock changes, however, the causality of uncertainty were different among variables in GVM. We will discuss about the reginal patterns and their possible mechanisms.