Total precipitable water (TPW) is an important parameter in many atmospheric studies ranging from climate studies to short-term weather forecasting. TPW is highly variable in both space and time and it is therefore desirable to obtain spatially continuous predictions of atmospheric water vapor for short time intervals. Multiple satellite-based systems are routinely available to measure TPW. Over the North America region, the most relevant are the GOES Sounder, surface-based Global Positioning System (GPS) sites, and the new Microwave Integrated Retrieval System (MIRS) water vapor profile retrievals over land and ocean from polar orbiting satellites. An operational NOAA blended TPW product distributed to forecasters currently uses these three types of measurements to create maps of TPW on an hourly basis at 16 km resolution. Ideally the information contained in all these different measurements is used jointly to get maps of atmospheric water vapor. Each data type contains unique strengths and limitations which make this a challenging problem.

We present a geostatistical methodology to derive spatially continuous hourly maps by blending GOES Sounder, GPS, and MIRS and measurements over North America. One of the challenges in blending different measurements is to develop a methodology that is robust for vastly different amounts of data availability and spatial coverage. Due to satellite paths and retrieval limitations, some of the hourly periods contain no measurements from one of the sensors, while other periods feature large overlap in their spatial coverage. The methodology further accounts for the different measurement errors associated with the different measurement systems. We apply the proposed geostatistical methodology to create approximately 40 hourly maps each for two time periods in January and June 2011. Their predictive performance is evaluated using cross-validation, which shows generally high correlations.