Abstract
Estimates of the amount of atmospheric water vapour derived from algorithms for a ground-based single-channel (21.0 GHz) microwave radiometer have been investigated. Ten datasets covering 44 days were used to derive the methods and two other sets (in total 32 days) were used to assess the quality of these. It is shown how the rms estimation error can be reduced by recognizing the rapid variations in sky brightness temperatures during periods when cloud liquid is present. Data was either discarded, guided by the variability, or an adaptive Kalman filter was applied with different parameter values for different degrees of variability. The resulting estimates were compared to the estimates obtained from a dual-channel algorithm (21.0 and 31.4 GHz), which were used as reference. The amount of water vapour was represented as the ‘wet delay’, the excess radio path length due to the atmospheric water vapour. Applying the Kalman filter to the single-frequency estimates reduced the wet delay rms error from 20 mm to 9 and 14 mm for the two datasets. Further reduction of the rms error was achieved by the removal of data in periods with high variability; discarding about 40% of the data led to rms errors of 5 and 7 mm for the two datasets.