retrospective atmospheric boundary conditions (9 variables: Rainfall, Snowfall, 2m Air Temperature, 2m Specific Humidity, Surface Pressure, Downward Shortwave Radiation, Downward Longwave Radiation, 10m Wind Speed, and Cloud Cover Fraction) for 1901-2010 in 3-hourly resolution are generated. 20th Century Reanalysis (20CR) [compo2011] [Compo el al., 2011] on global 2° resolution is dynamically downscaled into T248 (~0.5°) grid using a spectral nudging technique [Yoshimura and Kanamitsu 2008] in a Global Spectral Model (GSM) [Figure 2]. This successfully keeps the low frequency signal of original reanalysis, providing additional high frequency signals, which are lacking in previous products [e.g., Weedon et al., 2011]. It is essential to investigate phenomena at higher spatiotemporal scales such as extreme events. In order to relieve known artifacts (e.g., ripple patterns and persistent overcast in high latitude region), additional techniques, such as single ensemble correction [Yoshimura and Kanamitsu, 2013] and vertically weighted damping [Hong and Chang, 2012], are applied. Model biases in the downscaled 20CR are corrected using observational data (e.g., GPCC for precipitation, SRB for short/long wave radiation, and CRU for air temperature and daily temperature range). In addition to previously introduced bias correction algorithms [e.g., Weedon et al., 2011], variability in higher temporal (<month) resolution is carefully corrected [Kim et al., in preparation]. Also, wind-induced precipitation undercatch correction is applied considering different types of gauges and their global distribution [Hirabayashi et al., 2008]. Through the above mentioned data generation strategy, GSWP3 has further reliability and consistency over the century long target timespan with higher spatiotemporal resolutions.