A quality control system for historical in situ precipitation data

Abstract

<p>In situ precipitation data are recorded at observing stations typically using either manual or automated gauges (some countries have ruler measurements of snowfall, which were then converted to their water equivalent using some version ratio). Unfortunately, there are random erroneous values, which could be unusually large values or false 0s. The latter usually arose from mis-recorded missing values (i.e., missing values were mis-recorded as 0 precipitation in the climate Archive).</p><p>In doing quality control (QC) of Canadian in situ precipitation data records, we have found that it is necessary to apply a pair of QC procedures to identify these two types of random errors: one procedure is applied to the untransformed monthly precipitation series, which is good at finding outliers of unusually large values; another is applied to the log-transformed monthly precipitation series, log(P+0.1) (in mm), which is good at identifying outliers of zero or near-zero monthly total precipitation. The four nearest stations’ data for the same month are used to determine if the suspect outlier is a real extreme value or an erroneous value. All the monthly values identified to be erroneous are set to missing, and so are the corresponding daily values. </p>