Predicting monthly near-surface soil temperature from air temperature and the leaf area index

Abstract

Soil temperature at the land surface (Tsoil) is a key variable for modeling processes that occur belowground. Although Tsoil is correlated to the near-surface air temperature (Tair), these values are often not equal. Evidence confirms that Tsoil – Tair (ΔT) varies by ecosystem in both positive and negative directions. The goal of this study was to calibrate a semi-empirical model to predict Tsoil from Tair and its interaction with the leaf area index (LAI) at a monthly timescale. The latter variable is introduced to quantify the reduction in solar radiation reaching the land surface using Beer-Lambert's Law with an extinction coefficient k = 0.55. Bootstrap distributions for the three unknown parameters in the model were optimized using monthly Tair and Tsoil measurements from 169 eddy covariance towers in the FLUXNET2015 dataset accompanied by LAI observations at 0.5-km2 resolution from the MODIS satellite. The optimized model had a mean bias error (MBE) of -0.08 °C and a root mean square error (RMSE) of 2.26 °C. When Tsoil was assumed equal to Tair (ΔT = 0), these metrics worsened to -1.51 °C and 4.09 °C, respectively. Data from the US National Ecological Network (MBE/RMSE = -0.51°/2.81 °C) and published global predictions of ΔT obtained with a machine learning algorithm (MBE/RMSE = 0.73°/1.76 °C) validated the optimized model at the global scale. The data confirmed that ΔT tends to be positive when Tair 〈 0 °C, and transitions to negative values when Tair 〉 0 °C and LAI is greater than 1.89. In the proposed model, LAI values close to 1.89 indicate a 37 % reduction in solar radiation (e − 1 in Beer-Lambert's Law). Overall, this work confirms that Tsoil ≠ Tair in most ecosystems and that ΔT is predominantly explained by LAI in non-freezing conditions. Re-calibration is recommended before applying the model in barren ecosystems, or at spatiotemporal scales different from those in this research.