In this study, four types of high resolution satellite-derived precipitation estimates, i.e.: TRMM Multi-satellite Precipitation Analysis (TMPA) 3B42, CPC Morphing technique (CMORPH), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN), and Naval Research Laboratory (NRL)-Blended data, were examined using rain gage data collected from the dense rain gage network over the Korean peninsula [Automated Weather Station (AWS)] operated by Korean Meteorological Administration (KMA), as a part of the Program to Evaluate High Resolution Precipitation Products (PEHRPP). Three-hourly satellite products for three summer months (June, July, and August) of the period from 2003 to 2006, in the 0.25° grid format, are intercompared with one-minute AWS data from about 520 stations of South Korea.

In order to investigate error characteristics in space and time, all rainfall data were averaged over the varying space and time. The time domain spans from 3 hour to one month while space domain extends from 0.25° to 2°. It is noted that RMSE, mean bias, and correlation widely vary with chosen time and space domain. However, the TMPA and CMORPH seem to behave similarly although their mean biases are quite different. On the other hand, the PERSIANN and NRL-Blended show quite different correlation patterns from TMPA and CMORPH.

EOF analysis was conducted to examine the spatial coherence between the satellite estimates and rain gage measurements. Results indicates that all satellite estimates used in this study agree well with AWS gage measurements in the first mode, explaining about 70% of variance, but lesser degree of agreement is found in the higher modes. Amongst satellite products, the CMORPH shows the best statistics in the regression analysis while the TMPA 3B42 shows most similar spatial pattern to that of the AWS data.