Non-parametric error distribution analysis from the laboratory calibration of various rainfall intensity gauges

The analysis of counting and catching errors of both catching and non-catching types of rain intensity gauges was recently possible over a wide variety of measuring principles and instrument design solutions, based on the work performed during the recent Field Intercomparison of Rainfall Intensity Gauges promoted by World Meteorological Organization (WMO). The analysis reported here concerns the assessment of accuracy and precision of various types of instruments based on extensive calibration tests performed in the laboratory during the first phase of this WMO Intercomparison. The non-parametric analysis of relative errors allowed us to conclude that the accuracy of the investigated RI gauges is generally high, after assuming that it should be at least contained within the limits set forth by WMO in this respect. The measuring principle exploited by the instrument is generally not very decisive in obtaining such good results in the laboratory. Rather, the attention paid by the manufacturer to suitably accounting and correcting for systematic errors and time-constant related effects was demonstrated to be influential. The analysis of precision showed that the observed frequency distribution of relative errors around their mean value is not indicative of an underlying Gaussian population, being much more peaked in most cases than can be expected from samples extracted from a Gaussian distribution. The analysis of variance (one-way ANOVA), assuming the instrument model as the only potentially affecting factor, does not confirm the hypothesis of a single common underlying distribution for all instruments. Pair-wise multiple comparison analysis revealed cases in which significant differences could be observed.