Detecting temperature induced spurious precipitation in a weighing rain gauge

Abstract

We present a quality control algorithm to detect spurious precipitation events, which occur at weighing rain gauges in the automated precipitation monitoring network of MeteoSwiss. Although small in intensity, the spurious precipitation events must be removed during routine quality control, because they have a negative impact on climatological and meteorological applications of precipitation data. Automated monitoring, expert inspection and systematic analysis lead to the hypothesis that spurious precipitation is induced by rapid temperature changes at the load cell of the weighing rain gauge. Constrained by the black box nature of the signal processing performed by the firmware, we trained a statistical classifier on features extracted from measurements provided by the weighing gauge, and expert labels obtained from the routine quality control. The high sensitivity and specificity of the trained Support Vector Machine provide strong evidence in favor of the hypothesis. Furthermore, the classifier is suitable for operational deployment as an automated real-time quality control test, flagging individual 10‑minute precipitation intensities as either spurious or non-spurious. Our results also suggest that a modification of the instrument should be possible, such that it will not generate spurious measurements in the first place. We are therefore collaborating with the manufacturer on possible software and hardware improvements to the instrument.

Keywords

• automated quality control
• spurious precipitation
• weighing rain gauge
• temperature sensitivity
• statistical classification
• support vector machine