Biological and Environmental Research - Earth and Environmental System Sciences
Earth and Environmental System Modeling
09 June 2020

A Demonstration That Linear Two-Pool Models are Insufficient to Infer Temperature Sensitivity of Soil Carbon Decomposition From Incubation Respiration Time Series Data

Linear two-pool models suffer from high parametric equifinality.

Science

Linear models are regularly used to infer temperature sensitivity of soil carbon decomposition from incubation time series data. The authors comprehensively analyzed uncertainty of the parameter inference process using recently published observations (Tang and Riley, 2020). Because of very high equifinality, they conclude that the inferred Q10 values are not useful for predictive soil carbon models.

Impact

Our results suggest that (1) contrary to common practice, linear two-pool models are not able to infer soil carbon decomposition parameters that can directly inform predictive soil carbon models; (2) apparent active and slow (and likely passive) soil carbon pools can emerge from the interaction of a single substrate with microbes and mineral processes, supporting the idea that more mechanistic model treatments are needed to interpret soil incubation experiments; and (3) empirical experiments should collect more measurements that can directly inform mechanistic understanding and development of process-rich models.

Summary

Predicting climate and biogeochemistry feedbacks requires a robust estimation of the temperature sensitivity of soil carbon decomposition. This task is often addressed by deriving from empirical experiments the respiratory temperature sensitivity parameter, e.g., Q10 that measures the increase of respiration per 10 °C warming in temperature. The authors showed that the popular linear two-pool models, whether or not they consider the interactions between active and slow pools, are not able to provide a robust estimation of Q10 and other related parameters. In particular, when the posterior parameters are applied in predictive models, the input carbon flux will amplify the parametric uncertainty to produce very divergent predictions of soil carbon stocks. We also showed that such high parametric uncertainty is likely not rectifiable given the limited accuracy and information commonly measured in incubation experiments. Next, by analyzing the respiration time series with a microbe-explicit soil carbon model, we demonstrated that multiple pool model structures can be inferred when only one substrate is actually being decomposed. Therefore, we contend that empirical measurements should not try to infer temperature sensitivity of non-measurable soil carbon pools. Rather, experimentalists should collect more information that can inform the processes that lead to the emergent dynamics of soil carbon decomposition, and mechanistic models that explicitly represent the different processes should be more extensively explored.

Contact
William J. Riley
Lawrence Berkeley National Laboratory