A new algorithm to estimate precipitable water over land using the split-window data from the NOAA satellite is proposed. The estimation is based on the ratio of the split window channel brightness temperature difference and the ratio of the split window channel brightness temperature variance. The algorithm involves two main steps. The first is to extract high-confidence pixels which do not contain a large influence due to unresolved clouds, and to calculate the first-guess of precipitable water using the split-window variance ratio technique proposed by Jedlovec (1990). The second step corrects the first-guess of precipitable water using the difference between the mean-air temperature of the model atmosphere and the mean-air temperature calculated from the satellite data.
By using this algorithm, the precipitable water over an area of about 33km×33km can be estimated, while the influences due to unresolved clouds and unknown temperature profiles decrease. Compared with four radiosonde observations during August of 1992, the satellite-derived precipitable water data are in good agreement with the radiosonde data.