Climate models embody our understanding of how the terrestrial biosphere works. Processes within these models represent our current hypotheses and assumptions about how the physical system works. But how do we evaluate these complex models?

The standard approach is to compare a model, or a series of models against independent datasets. However, when these models diverge from each other and/or observations, how do we isolate the actual cause of the problem? How do we ensure it is process A and not in fact the unfortunate interaction between two other (perhaps unrelated) processes? Current models are too complex and lack the modularity to simply test individual processes.

In this paper we proposed a model agnostic, modular framework – the multi-assumption architecture and testbed (MAAT). The framework can be used to systematically run multiple model simulations to explore how the different underlying model assumptions, hypotheses and parameters lead to the predicted model behaviour.

To demonstrate the value in this approach we demonstrate two potential applications. Firstly, we ran a sensitivity analysis of a simple groundwater model. Secondly, we implemented a widely used leaf-scale photosynthesis model to examine the process variability due to different assumptions.

The real value in this tool will come from future work. MAAT is open-source and we welcome engagement from the wider community. We hope that other users will embrace this tool to facilitate improvements in their model systems moving forward.

Paper: Walker, A. P., Ye, M., Lu, D., De Kauwe, M. G., Gu, L., Medlyn, B. E., Rogers, A., and Serbin, S. P.: The multi-assumption architecture and testbed (MAAT v1.0): R code for generating ensembles with dynamic model structure and analysis of epistemic uncertainty from multiple sources, Geosci. Model Dev., 11, 3159-3185,, 2018.