Benchmarking as utilisation of information by land surface models

There is understandable confusion in the land surface modelling community regarding the distinction between model benchmarking and evaluation. Here we reinforce the notion that benchmarking requires the prescription of model performance expectations before models are evaluated. That is, for any given metric, how well should we expect a model to perform? A weak example of benchmarking might require improvement over a previous model version, or better, define thresholds of performance required for a model to be fit for a particular application. A much stronger example defines benchmark performance as a function of a model's complexity and the amount of information provided to it in its input variables and parameters regarding the fluxes it is required to simulate. This attempt to quantify ‘utilisation of information' by a model and offers us an objective definition of what makes a “good” model and so provides obvious candidates for benchmark performance thresholds. Practical examples of an implementation of this approach will be offered from the Protocol for the Analysis of Land Surface models (PALS) web application and PLUMBER model comparison experiment.