A new approach to predict soil temperature under vegetated surfaces.

In this article, the setup and the application of an empirical model, based on Newton's law of cooling, capable to predict daily mean soil temperature () under vegetated surfaces, is described. The only input variable, necessary to run the model, is a time series of daily mean air temperature. The simulator employs 9 empirical parameters, which were estimated by inverse modeling. The model, which primarily addresses forested sites, incorporates the effect of snow cover and soil freezing on soil temperature. The model was applied to several temperate forest sites, managing the split between Central Europe (Austria) and the United States (Harvard Forest, Massachusetts; Hubbard Brook, New Hampshire), aiming to cover a broad range of site characteristics. Investigated stands differ fundamentally in stand composition, elevation, exposition, annual mean temperature, precipitation regime, as well as in the duration of winter snow cover. At last, to explore the limits of the formulation, the simulator was applied to non-forest sites (Illinois), where soil temperature was recorded under short cut grass. The model was parameterized, specifically to site and measurement depth. After calibration of the model, an evaluation was performed, using ~50 % of the available data. In each case, the simulator was capable to deliver a feasible prediction of soil temperature in the validation time interval. To evaluate the practical suitability of the simulator, the minimum amount of soil temperature point measurements, necessary to yield expedient model performance was determined. In the investigated case 13-20 point observations, uniformly distributed within an 11-year timeframe, have been proven sufficient to yield sound model performance (root mean square error <0.9 °C, Nash-Sutcliffe efficiency >0.97). This makes the model suitable for the application on sites, where the information on soil temperature is discontinuous or scarce.