Towards a More Careful Construction of Land Models

Most land models have a convoluted heritage. Model enhancements are often incorporated in a rather ad-hoc fashion with limited attention given to process coupling, numerical error control, systematic tests, and model extensibility and flexibility. As a result, it can be very difficult to robustly incorporate new modeling approaches and evaluate competing modeling alternatives.

This presentation summarizes experiments with the Structure for Unifying Multiple Modeling Alternatives (SUMMA) to evaluate alternative methods for constructing land models. Specifically we introduce a fully-coupled solution for the hydrologic and thermodynamic state equations (with numerical error control and adaptive substepping), and compare the fully coupled solution with the uncontrolled and ad-hoc operator splitting and asynchronous coupling methods that are typically employed in the current generation of land models. We also introduce and compare several strategies for model speedup. We demonstrate that the current generation of land models is contaminated by troublesome numerical artifacts that complicate model analysis, especially parameter estimation and hypothesis testing. Further, we demonstrate that clear separation of the physical process parameterizations from the numerical solution methods, and adopting robust global solvers, substantially increase model flexibility and extensibility. We will end the presentation with a discussion on ways to improve the construction of land models, in order to simplify sharing of code and concepts across model development groups and accelerate land model development.