Infrared visibility prediction by statistical methods

Abstract

This study attempts to find statistical methods of predicting infrared visibility (IRV), as calculated from hourly meteorological observations from a North Atlantic weather ship. Simple and multiple regressions expressing IRV as a function of its component weather variables, and exponential data transformations, for time lags of 1 to 24 hours, gaveR ² values from 0.68 (1-hour lag) to 0.09 (24-hour lag). These have limited predictive power for lags up to 6 hours, almost none for longer lags. Two-category discriminant analysis, using class breaks at 2 km or 10 km is of little use, due to uneven data distribution.

Possibly more promising would be an application of Machine Output Statistics (‘MOS’), used routinely for temperature forecasts, to this problem.