Impact of Sampling Bias on the Trend of Total Precipitable Water Derived from Latest 10-Year Data of COSMIC,SSMIS and HIRS Observations

Satellites provide global observations for the atmospheric water vapor, the transportation and trends of which are of great importance. However, due to large spatial and temporal variations of different satellite observations, the trend analysis of water vapor using satellite datasets might be highly affected by their sampling bias. To verify it, the variability and trend of the de-seasonalized monthly mean TPW (total precipitable water) derived from SSMIS (Special Sensor Microwave Imager Sounder), COSMIC (Constellation System for Meteorology, Ionosphere, and Climate) and HIRS (High-resolution Infrared Radiation Sounder) datasets are compared with the corresponding collocated TPW from ERA-Interim reanalysis of the European Centre for Medium-Range Weather Forecasts (ECMWF) for the period of January 2006 – December 2015. The four sets of data are firstly evaluated using collocated ground-based microwave radiometer (MWR) TPW measurements from the Atmospheric Radiation Measurement (ARM) sites. It indicates that all of them show good consistencies with the MWR, except that the TPW from HIRS data shows a little drier bias due to its limitation in only clear sky conditions. An increasing trend in the global TPW can be found in all three satellite datasets over the chosen periods. However, the comparisons with ERA-Interim reanalysis data show that the sampling bias in different satellite datasets has significant impact on the TPW trend and variability. For instance, the SSMIS data shows the largest TPW value with an increasing trend of 0.54mm/decade. Meanwhile, it shows the best agreement with collocated ERA-Interim data, indicating that the sampling bias of SSMIS data has the smallest influence on the accuracy of the TPW trend among three satellite datasets. The TPW from COSMIC data shows a very large increasing trend of 0.82 mm/decade. However, the big difference between the COSMIC collocated and original ERA-Interim data indicates that the COSMIC data has a high sampling bias, and thus its trend could not be well representative of the global averaged one. For the HIRS data, the clear sky TPW variability shows the worst agreement with the collocated ERA-Interim data. But there is a big drop of TPW trend because of the strong signal of La Niña event in the year of 2010 and 2011, which does not show in other datasets due to different sampling patterns. Additionally, the further recalibration of the HIRS radiance data from different satellites may help to mitigate the inconsistency problem.