To what extent do uncertainty and sensitivity analyses help unravel the influence of microscale physical and biological drivers in soil carbon dynamics models?

摘要

Soil respiration causes the second largest C flux between ecosystems and the atmosphere. Emerging soil carbon dynamics models consider the complex interplay of microscale interactions between the physical and biological drivers of soil organic matter decomposition occurring in the 3D soil architecture. They are expected to provide a way to upscale results to the macroscopic level and as such appear as an alternative modelling approach to the traditional "black-box" macroscopic models. However, these models still need to be tested under a broader range of their parameters values and structures than has been the case to date. We thus conducted uncertainty and global sensitivity analyses to test the robustness of previous predictions on dissolved organic carbon biodegradation obtained by one of these microscopic carbon dynamics models, LBioS. Six parameters of the carbon dynamics module of LBioS, associated with bacterial metabolism and three microscopic 3D descriptors of soil architecture were considered as uncertain inputs. We built two complete factorial designs in which the minimum and maximum of uncertainty intervals are considered. Each factorial design is assigned to a particular structure of the model, one including dormancy of bacteria and the other considering optimal bacterial activity. The scenarios took place in 3D computed tomography images of an undisturbed cultivated soil. The sensitivity indices at different simulations dates were computed with an ANOVA procedure taking into account main effects and interactions among factors. The uncertainty analysis shows that only in the limiting case of low accessibility of resources to bacteria the different microbial metabolisms tested can modify to a small extent the system responses, and uncertainty linked to parameters describing soil architecture becomes preponderant. In the case of optimal accessibility output variability is due predominantly to uncertainty of the microbial metabolism parameters. The sensiti