Equifinality and overparameterisation undermine confidence in predictions by soil organic matter models

The simulation of soil organic matter (SOM) dynamics, including SOM persistence, is a vital component of broader models representing vegetation dynamics or the impact of environmental change on the biosphere and climate. One of the biggest challenges in the application of SOM models is that their complexity is often not supported by sufficient data for parameter optimization. Inevitably, this leads to the calibration of more parameters than can be reliably optimised with available data, resulting in equifinality. This is the phenomenon that multiple parameter sets generate behavioural models: similarly well-performing models that cannot be ruled out.

This study assessed how equifinality affects the variability of predictions made by behavioural SOM models. We used a mechanistic, microbially-driven soil organic carbon and nitrogen model and evaluated it against an artificial data set. After the models were successfully calibrated and run into steady state, carbon inputs were doubled to evaluate how models with different mathematical formulations and different amounts of data used to optimize parameters reacted to this external forcing.

The key results are summarised as follows. (1) The accurate simulation of total SOM in steady state is an insufficient criterion to evaluate model performance. (2) The amount of calibration data determines how many model parameters can be jointly optimised without their values compensating for each other (i.e., identifiable parameters). And (3) the type of calibration data is equally important, as it dictates which pools can have their size and turnover rate constrained. For example, the size of particulate organic matter (POM) and mineral-associated organic matter (MAOM) can only be accurately simulated when data on these pool sizes are available. Similarly, the turnover rate of MAOM can only be reliably simulated if Δ¹⁴C data for MAOM are present. Our results emphasise the necessity of optimising only identifiable model parameters to avoid hidden uncertainty in model predictions. Adopting this approach consistently represents an important step forward to increase confidence in predictions made by SOM models.