13th Conference on Applied Climatology and the 10th Conference on Aviation, Range, and Aerospace Meteorology

Tuesday, 14 May 2002: 8:00 AM
Exploration of global teleconnection indices for long-range temperature forecasts
Robb M Randall, Air Force Institute of Technology, WPAFB, OH; and R. P. Lowther
The Air Force Combat Climatology Center (AFCCC) is continuously tasked to provide temperature and other forecasts for locations at which Department of Defense (DoD) personnel are performing long-range exercises and real-world mission planning support. DoD needs long-range forecasts to estimate how much fuel is necessary to keep energy production and purchases at the proper levels to accommodate all energy needs on their installations and within their worldwide areas of operation. Currently, the best long-range temperature forecasts the weather community has for worldwide locations are the climatological standard normals. This study creates a stepping-stone into the solution of long-range forecasting by finding a process to predict temperatures better than the climatological standard normals or simple frequency distributions of occurrences. This same solution is also highly sought after by non-DoD users.

Northern Hemispheric teleconnection indices, created by rotated principle component analysis (RPCA), and the Standardized Southern Oscillation Index are statistically compared to the Heating Degree Days (HDDs) and Cooling Degree Days (CDDs) of 14 U.S. locations. HDDs and CDDs were summed over three month periods for a seasonal summation. Teleconnection indices found to be leading modes, using RPCA, in a particular month were compared to the HDD/CDD summation of the following three months in order to create a predictive model.

First, a comparison of simple linear regression was accomplished and results showed numerous valid models, but the amount of data resolved by the models was rarely over 30%. The HDDs and CDDs where then categorized and analyzed with a classification tree, data mining program, however, the results proved difficult to extract any predictive quantitative information.

Finally a CART analysis was performed on the uncategorized HDDs/CDDs and regression tree analysis showed an excellent conditional predictive outcome. At each conditional outcome a range is produced using the predicted standard deviations about the mean. When teleconnection indices were put into the conditional model, 95% of the time the resulting HDDs/CDDs fell into the calculated range. This conditional model was verified using an independent data set withheld from the original analysis.

Supplementary URL: