The operational global numerical weather prediction model at the Japan Meteorological Agency (JMA) currently adopts a Simple Biosphere (SiB) model. JMA has been developing a new SiB model (New-SiB) as an upgrade to the current SiB model (Op-SiB). New-SiB has improved treatment of soil and snow processes compared with Op-SiB. The results of integration experiments performed over a period of 3 years indicate that New-SiB predicts snow-covered areas more accurately than Op-SiB. Moreover, the monthly mean surface air temperature during the polar night is better simulated using New-SiB. To validate the diurnal cycle of near-surface meteorological elements in New-SiB, short-range forecasts obtained using the two SiB models are verified using CEOP (Coordinated Enhanced Observing Period) in situ observation datasets for the Enhanced Observing Period 3 (EOP-3). New-SiB generally simulates diurnal variations in near-surface temperature more accurately than Op-SiB, especially over snow-covered areas; however, the short-range forecast experiments also reveal a number of shortcomings in both models. The CEOP datasets are highly valuable in evaluating numerical weather-prediction models.