A Nonlinear Filter to Remove Impulse Noise from Meteorological Data

Abstract

Nonlinear median filters were modified to use threshold logic and used to remove impulse noise (spikes) from a set of meteorological data. The impulse noise in the dataset, which originated in the communications section of the Portable Automated Mesonet, could be characterized as random bit noise. Most of the pulses had a duration of one time interval, which in this case was one minute. The filters were effective irrespective of the frequency of occurrence and of the amplitude of the noise spikes. Pulses were removed even when the frequency of occurrence rose to every other data point as was observed in several short intervals. The amplitude of pulses removed ranged over three orders of magnitude.