P4.50
Detection of daylight arctic clouds using MISR and MODIS data
Tao Shi, The Ohio State University, Columbus, OH; and E. E. Clothiaux, B. Yu, A. J. Braverman, and D. N. Groff
Preliminary studies of MISR and MODIS daylight radiances and standard cloud data products have indicated that the information content in MISR and MODIS radiances is sufficient for accurate detection of Arctic clouds during daytime. In the study the accuracy of cloud detections is evaluated relative to more than 2.685 million 1.1-km resolution expert labels applied to 3.946 million pixels with valid radiances from 32 scenes that contain both clear and cloudy pixels and 25 scenes with only clouds or clear skies. Using Fisher's quadratic discriminate analysis (QDA) classifiers with expert training labels and MISR radiances, MISR radiance-based features, MODIS radiances, and MODIS radiance-based features as input vectors to the classifiers led to classification accuracies of 87.51%, 88.45%, 96.43%, and 95.61%, respectively. The accuracies increase to 96.98% (96.71%) when QDA with expert labels is applied to combined radiances (features) from both MISR and MODIS. One of the more interesting findings of the study is the classification accuracies (i.e., 96.53% and 99.05% for mixed cloud/clear and pure scenes, respectively) and scene coverages (i.e., 74.91% and 78.44% for mixed and pure scenes, respectivley) of those pixels for which the automated MODIS operational cloud mask algorithm and a second automated algorithm that uses MISR angular radiance-based features agree. Having spectral and angular radiance-based cloud classifications in agreement is almost an error free indicator of the class type (i.e., clear or cloudy) for a pixel.
Poster Session 4, Radiation Poster Session IV: Remote Sensing
Wednesday, 12 July 2006, 5:00 PM-7:00 PM, Grand Terrace
Previous paper Next paper