Abstract
A statistical method has been developed using satellite, climatological, and persistence data to predict tropical cyclone position 12, 24, 48 and 72 h after initial observation. The satellite measurements were infrared window channel (11.0 μm) equivalent blackbody temperatures (TBB), which gave representations (through the cloud and surface temperature fields) of the structure of the cyclones and the circulation features surrounding them. There were 197 individual measurements of TBB for each cyclone observation. Algorithms have been prepared using digital data from a single satellite image, 14 climatological and persistence type variables, and a combination of these data sources. The algorithms were developed using a unique statistical procedure based on an eigenvector preprocessing and the use of independent tests for screening decisions.
Independent testing of these algorithms showed that the average error made by the algorithms developed from the single satellite observation were comparable to the 48 h Joint Typhoon Warning Center (JTWC) forecast and were approximately 10% better for 72 h forecasts. Forecasts using only the climatological and persistence variables were about 20% worse than JTWC for 24 h forecasts and 10% worse for 48 and 72 h forecasts. When both satellite and nonsatellite variables were included, the performance was comparable to JTWC's for the 24 and 48 h forecasts and approximately 25% better than JTWC's for the 72 h forecasts.
The performance of the objective algorithms for various partitions was analyzed. It is shown that both the satellite and nonsatellite variables make significant and unique contributions.
