The 30-year-long bias-corrected monthly mean surface air temperature data from historical automatic weather stations (AWSs) at Mizuho, Relay Station, and Dome Fuji, is available online at https://doi.org/10.17592/002.2025010413. Nonetheless, these data are not continuous records and some months have no observations. To fill in the missing observations, the monthly mean temperature datasets were reconstructed from ERA5 2-metre temperature (ERA5 T2m). ERA5 T2m is less influenced by the assimilation of AWS data and is capable of reasonably predicting the near surface temperature at the AWS sites. However, ERA5 T2m has a slight warm bias compared to quality controlled AWS. ERA5 T2m has an overestimation of lower temperatures (< -40 degree C), but the biases show a linear increase with decreasing temperature. Therefore, a regression model with AWS observations can be used to estimate the inherent bias in ERA5 T2m. Here the regression model was calibrated using data from 1993 to 2009 (calibration period), and then applied to the ERA5 T2m data from 1993 to 2022. The reliability of the bias-corrected ERA5 T2m was tested from 2010 to 2022 (testing period). During this period, the root-mean-square error (RMSE) at three AWS stations was less than 1.4 degree C. This is nearly equivalent to or less than, the measurement error of AWS observation (1.5 degree C). In addition, the RMSE for the testing period (2010-2022) is substantially smaller than that for the calibration period (1993-2009), except for Mizuho. This implies that the quality of bias-corrected ERA5 T2m is relatively homogeneous from 1993 to 2022.
A complete 30-year long time series of monthly means is reconstructed from AWS records filled with bias-corrected ERA5 T2m at Mizuho, Relay Station, and Dome Fuji. The uncertainty of the reconstructed dataset is less than 1.5 degree C. This value is much smaller than the standard deviations of monthly temperatures.
In this repository, there are three types of monthly mean temperature data in CSV format from 1993 to 2022 at three stations: 1) AWS only data, 2) bias-corrected ERA5 T2m, and 3) AWS filled with bias-corrected ERA5 T2m.