Abstract

Archiving of radiosonde measurements dates back far into our recent history. These data are raw in nature and can be incongruous due to changes in equipment, observation location, measurement collection and errors caused by system failure, data transmission and processing. For these data simple G... Show moreArchiving of radiosonde measurements dates back far into our recent history. These data are raw in nature and can be incongruous due to changes in equipment, observation location, measurement collection and errors caused by system failure, data transmission and processing. For these data simple Gaussian estimation of the mean and standard deviation is insufficient as they can be affected by large outliers in a time series. We compare two robust estimators of the mean and standard deviation for detection of outliers in atmospheric temperature readings. The first method, developed by Huber, provides a robust estimate of the mean and standard deviation through winsorising (a transformation that reduces the effect of outliers). This method gives a two-sided estimate of the standard deviation. The second method, developed by Lanzante, uses a weighted estimator that gives higher weight to values closer to the center of mass of the distribution, but only computes a single standard deviation. Because the Huber method calculates standard deviations on both sides of the mean, it should be more accurate at identifying outliers in skewed time series. We apply these methods to numerous long term radiosonde stations to understand the effectiveness of both methods. Show less