Forecasting Short-Term Movement and Intensification of Tropical Cyclones Using Pattern-Recognition Techniques
Abstract
Weather forecasters recognize patterns within meteorological data fields and associate these patterns with a response of observed weather events, such as rain or hail. Two pattern recognition techniques, a linear statistical technique (correlation analysis) and a nonlinear neural network technique (back propagation) were tested to recognize patterns within the surrounding wind and height fields of tropical cyclones and to relate these patterns to short-term (24 hours into the future) movement and intensification of the cyclones. Two independent databases were obtained for tropical cyclones occurring within the western North Pacific region. The developmental database represented 292 cases of tropical cyclones and associated fields from 1978 to 1987; the test database contained 54 cases from 1988 to 1989. Gridded field data (5x5 with a 300 nm grid spacing) consisted of five upper-air levels of geopotential height and u and v components of the wind centered on the tropical cyclone. The forecast ability of both pattern recognition techniques was compared to that of persistence, the forecasters at the Joint Typhoon Warning Center (JTWC), and three objective forecast techniques employed at JTWC. The neural network technique, back propagation, predicted short-term intensification more accurately than the forecasters at JTWC and persistence; the correlation analysis method was not as accurate. Both pattern-recognition techniques forecast short-term movement better than persistence and comparably to the forecasters at JTWC; however, both techniques were biased toward forecasting persistent movement.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 08, 1991
- Accession Number
- ADA256705
Entities
People
- John Pickle
Organizations
- Phillips Laboratory