3.2
Detection and characterization of thin cirrus using combined MISR and MODIS observations

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Monday, 30 January 2006: 4:30 PM
Detection and characterization of thin cirrus using combined MISR and MODIS observations
A305 (Georgia World Congress Center)
Michael J. Garay, Univ. of California, Los Angeles, CA; and D. M. Mazzoni and R. Davies

Clouds are one of the most important, but poorly understood, aspects of the global climate system. Optically thin cirrus clouds, in particular, have attracted special attention because they provide a positive net cloud radiative forcing and are often inadequately represented in global climate models. Observationally, thin cirrus traditionally present a challenge to satellite retrieval methods because, by their nature, they are difficult to detect from nadir-pointing instruments and, while they have been actively studied by limb sounding instruments, such as the High Resolution Infrared Radiation Sounder (HIRS), these instruments do not provide much information about the horizontal extent of such clouds.

The newest generation of satellite instruments on the EOS Terra satellite platform offers the potential to overcome these difficulties. Two instruments, the Multi-angle Imaging Spectroradiometer (MISR) and the Moderate Resolution Imaging Spectroradiometer (MODIS), each provide information that may lead to better detection and characterization of thin cirrus. Based on comparisons with ground-based lidar, we have found that the 1.38 and 11 micron channels on MODIS have some sensitivity to the presence of thin cirrus clouds. In these same comparisons, MISR is able to sense thin cirrus directly because it has cameras with oblique viewing angles that provide factors of two and three increases in optical pathlength. We will present results from these case studies that demonstrate the sensitivity of MISR and MODIS to thin cirrus at horizontal scales ranging from 275 m for MISR to 1 km for MODIS.

Additionally, we have been exploring machine learning approaches that are able to exploit the massive amounts of geophysical information available from satellite instruments to yield new retrieval methods. We have already developed a thin cirrus cloud detection algorithm for MISR using Support Vector Machines (SVMs), which are a form of supervised learning algorithm similar to neural networks. Using MISR and MODIS data together we will demonstrate the utility of the SVM approach for “fusing” observations from the two instruments to provide superior thin cirrus detection and characterization at a global scale.