Investigating Differences in Soil Temperature Ranges of Land Surface Models Using Site Simulations from ESM-SnowMIP

Abstract:

While relatively complex land-surface models have existed for about three decades, there exists a wide range in the diurnal and seasonal soil temperature patterns of current models. The Canadian Land Surface Scheme (CLASS) has been shown to overestimate the diurnal and seasonal range in soil temperature under vegetation canopies. However, CLASS was not an outlier in the recent site level simulations from the Earth System Model - Snow Model Intercomparison Project (ESM-SnowMIP), which showed that issues that have existed in land surface and snow models for years have continued to contribute to differences in model performance. Results from three forest sites show that while the bias in snowpack temperature is warm in some models and cold in others, most models display a winter cold soil bias. Models exhibiting the strongest cold bias tend to be associated with one or many of the following issues: a large or fixed snow thermal conductivity, a large or fixed snowpack density, no organic surface soil layers, a composite snowpack, no liquid water in the snowpack, and an inability to freeze soil water in one model. Previous studies at open sites have shown that the suppression of turbulent exchanges under stable conditions results in a positive longwave radiative feedback, leading to a cold snowpack that melts late. However, many of these models with a winter cold soil bias show an underestimation of snow water equivalent, shorter snow seasons, and greater variability (diurnal and seasonal ranges) in soil temperature and less variability in turbulent heat exchanges. We examine surface and soil temperature patterns in relation to turbulent heat exchanges over diurnal and seasonal periods using data from a subset of participating models from the ESM-SnowMIP site simulations, as well as differences in model parameterizations to elucidate the factors controlling the skill of the surface temperature simulations.