Extended Abstract (304K)P7.3
Cloud ceiling height estimation using GOES-10 and COAMPS data
Richard L. Bankert, NRL, Monterey, CA; and M. Hadjimichael
Data mining techniques, used in a Knowledge Discovery from Databases (KDD) methodology, are applied to a satellite and numerical model database for cloud ceiling height diagnosis. The database is composed of 2.5 years of location-specific and time-coincident GOES-10 and COAMPS data along with METAR observations of cloud ceiling height (ground truth) for the California coast and specific inland locations. Patterns and relationships in the GOES and COAMPS data that best estimate cloud ceiling conditions are determined. An algorithm is developed from these relationships that consists of a 3-step approach to be applied to a given pixel (or grid point): (1) determine if a ceiling exists, (2) determine if the ceiling is high or low (1000 m threshold) and (3) compute ceiling height for low ceiling cases. Separating the data into training and testing sets, the KDD algorithms (GOES-only, COAMPS-only, GOES/COAMPS-combined) were shown to have significantly more skill than the current operational translation algorithm applied to COAMPS output. The GOES/COAMPS-combined KDD algorithm had the best overall performance with a Probability of Detection (POD) of .80 and a True Skill Score (TSS) of .77 for ceiling identification (step 1); POD is .92 and TSS is .63 for low ceiling identification (step 2); Root Mean Square Error (RMSE) is 168 m for low ceiling height estimation (step 3).
Poster Session 7, Retrievals and Cloud Products: Part 1
Thursday, 23 September 2004, 9:30 AM-11:00 AM
Previous paper Next paper