Land models are a necessary tool for understanding the earth system, yet these models are often unconstrained. Process representations vary between models and parameter values are unknown, contributing to uncertainty in modeling land processes. As a result, when land models are intercompared, without tuning, there is large divergence in simulated values.
We demonstrate how the observed surface temperature (Ts), can inform model evaluation and development. Typical evaluation in land modeling relies on variables that lack information on the surface energy balance. Ts is the diagnostic variable of the modeled surface energy balance and is an ideal supplement to traditional model evaluations. This study focuses on modeling seasonal snow packs, but many of the lessons learned are transferable to land modeling in general.
We use the Structure for Unifying Multiple Modeling Alternatives Model, which allows systematic evaluation of process representations. We test the impact of modeling decisions that impact the surface energy balance of snow, specifically the representation of vertical snow layer structure, the thermal conductivity, and the stability corrections. Additionally, we examine how data uncertainty affect simulated Ts through prescribed error structures.
Internal modeling decisions had a relatively small impact on Ts, even when we deliberately biased the physics in the tested model components. Using Ts, we fingerprint errors in the forcing data for future use in model development and evaluation. While some of the meteorological variables had distinct patterns in Ts, the model response often overlapped, limiting our ability to identify errors in the surface energy balance through Ts alone. For instance, the Qli and wind both strongly governed the minimum nighttime Ts. Finally, many of the modeled Ts responses are governed by feedbacks between radiative and turbulent fluxes.