Answers are grounded in this paper and cite their sources.
High-frequency forest soil temperature records remain scarce. We tested whether hourly ERA5-Land, a 9 km land reanalysis, can reproduce soil temperature at 10 cm beneath closed subtropical canopies. At four forest stations in Fujian, China, we compared two consecutive years of 10 min observations per station with seven estimators that used no soil temperatures from these stations. These were elevation-adjusted ERA5-Land soil temperature, two land-surface models (NoahPy and ecLand) driven by ERA5-Land, and four machine-learning models trained on soil temperature measured at nine other forest stations. NoahPy was the most accurate estimator in all eight site-years, with root-mean-square errors (RMSEs) of 0.46–0.75 °C and reproduced the observed mean daily range of 0.5–0.9 °C and the seasonal amplitude. The machine-learning models (RMSEs 0.71–1.71 °C) had smaller mean biases than interpolated ERA5-Land but overestimated the daily range 1.5–2.5-fold and peaked 1.0–3.5 h early. Interpolated ERA5-Land and ecLand (RMSEs 1.35–1.95 °C) overestimated the daily range about twofold. Removing NoahPy’s vegetation attenuation of topsoil thermal conductivity raised its daily range ratio from 0.93 to 3.24 and its RMSE to 1.99 °C. At these four stations, ERA5-Land reproduced forest-floor soil temperature when it drove a land-surface model that strongly damps topsoil heat conduction.
No takes yet. Share an insight, caveat, or question.
Qing et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: