Big-data-big-model Fusion to Improve Prediction of Global Soil Carbon Dynamics with Earth System Model

Thursday, December 13, 2018 - 08:00
Add to Calendar

Soil carbon is a critical component in terrestrial ecosystems, yet our understanding of the vertical soil carbon distribution is limited and enormous uncertainties still exist in the projection of different earth system models. In this study, by applying Bayesian inference in the matrix form of ORCHIDEE-MICT model, the vertical soil carbon distribution was assimilated by a global dataset containing more than 50,000 soil profiles. The assimilated model projection can explain more than 80% variance of the observation, and the explanatory power keep consistency at different depth of the soil profile. An estimated global carbon storage of 2048 Pg C was obtained after data-model fusion, with 1373 Pg C in the top 100cm of the soil. Surprisingly, the results suggest an anti-conventionally more responsive decomposition of soil carbon to the temperature in tropics (Q10 ~ 2) than that in boreal regions (Q10 ~ 1.3). The peculiar Q10 pattern reflects defects in model structure and refreshes our previous knowledge on the exponential temperature-respiration relationship. To summarise, the big-data-big-model fusion achieves significantly reduced uncertainty in earth system model projection and reveals critical defects in the model structure. The results facilitate a better understanding of the terrestrial carbon cycle and an ever-precise vertical soil carbon projection of the world by earth system models.

Link for More Information: