Detecting discontinuities in time series of upper-air data: development and demonstration of an adaptive filter technique

Illustrates the use of an adaptive moving average filter in detecting systematic biases and to compare its performance with the Schwarz criterion, a parametric method. The advantage of the adaptive filter over traditional parametric methods is that it is less affected by seasonal patterns and trends. The filter has been applied to upper-air relative humidity and temperature data. The accuracy of locating the time at which a bias is introduced ranges from about 600 days for changes of 0.1 standard deviations to about 20 days for changes of 0.5 standard deviations.