Poster Session P1.34 Satellite and numerical model data—driven cloud ceiling and visibility estimation

Monday, 1 August 2005
Regency Ballroom (Omni Shoreham Hotel Washington D.C.)
Richard Bankert, NRL, Monterey, CA; and M. Hadjimichael, P. H. Herzegh, G. Wiener, J. Cowie, and J. M. Brown

Handout (255.5 kB)

For aviation and other purposes, ground observations (e.g, METAR) of cloud ceiling and visibility are not always available where needed. Through proper analysis, satellite (GOES-12) and numerical model (RUC) data can be very useful in filling the observation gaps and, potentially, providing forecasts of ceiling and visibility. Data mining techniques, used in a Knowledge Discovery in Database (KDD) process, are being applied to GOES-12 and RUC data with METAR ceiling and visibility observations serving as the ground truth. A 1-2 year period of hourly data records for 178 METAR locations will be analyzed over four geographic regions east of the Rocky Mountains in the continental U.S. Various data relationships that will be used to estimate ceiling and visibility are being derived within three studies: GOES-only data, RUC-only data, and a combination of the two data sources. A preliminary performance evaluation of a RUC-only cloud ceiling algorithm has been completed and demonstrates the potential of the KDD process. Dividing the hourly data for 51 METAR stations in Iowa (one of the geographic areas) into two clustered areas of training and testing sets, a 3-step approach is used to assess cloud ceiling. The first step is to determine whether a ceiling exists. Testing set results indicate a Probability of Detection (POD) of .66 and a True Skill Score (TSS) of .58. The second step is a classification of the ceiling cases into high or low ceiling (1000 m threshold). The testing results are POD: .88 and TSS: .64. The final step is to estimate the ceiling height for low ceiling cases. The testing set produced an average absolute error of 191.7 m with a correlation coefficient of .70. An algorithm developed from a combination of RUC and GOES-12 data is expected to produce higher skill. Further evaluation will include visibility algorithms and other geographic regions.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner