College of Engineering | Apply to CSU | Disclaimer | Equal Opportunity Statement | Privacy | Search CSU


You must be on the CSU network—either physically or using VPN—to watch this or any of the videos on this site.

May 30, 2017
David Duncan
Hosted by Christian Kummerow (advisor), Russ Schumacher, Steven Rutledge, Christopher O’Dell (CIRA), Steven Reising (Electrical and Computer Engineering), Sid-Ahmed Boukabara (NOAA/STAR)


Passive microwave observations from satellite platforms constitute one of the most important data records of the global observing system. Operational since the late 1970s, passive microwave data underpin climate records of precipitation, sea ice extent, water vapor, and more, and contribute significantly to numerical weather prediction via data assimilation. Detailed understanding of the observation errors in these data is key to maximizing their utility for research and operational applications alike. However, the treatment of observation errors in this data record has been lacking and somewhat divergent when considering the retrieval and data assimilation communities. In this study, some limits of passive microwave imager data are considered in light of more holistic treatment of observation errors.

A variational retrieval, named the CSU 1DVAR, was developed for microwave imagers and applied to the GMI and AMSR2 sensors for ocean scenes. Via an innovative method to determine forward model error, this retrieval accounts for error covariances across all channels used in the iteration. This improves validation in more complex scenes such as high wind speed and persistently cloudy regimes. In addition, it validates on par with a benchmark dataset without any tuning to in-situ observations. The algorithm yields full posterior error diagnostics and its physical forward model is applicable to other sensors, pending intercalibration.

This retrieval is used to explore the viability of retrieving parameters at the limits of the available information content from a typical microwave imager. Retrieval of warm rain, marginal sea ice, and falling snow are explored with the variational retrieval. Warm rain retrieval shows some promise, with greater sensitivity than operational GPM algorithms due to leveraging CloudSat data and accounting for drop size distribution variability. Marginal sea ice is also detected with greater sensitivity than a standard operational retrieval. These studies ultimately show that while a variational algorithm maximizes the effective signal to noise ratio of these observations, hard limitations exist due to the finite information content afforded by a typical microwave imager.