Development and Calibration of a Low-Cost, Piezoelectric Rainfall Sensor through Machine Learning.

In situ measurements of precipitation are typically obtained by tipping bucket or weighing rain gauges or by disdrometers using different measurement principles. One of the most critical aspects of their operational use is the calibration, which requires the characterization of instrument responses both in laboratory and in real conditions. Another important issue with in situ measurements is the coverage. Dense networks are desirable, but the installation and maintenance costs can be unaffordable with most of the commercial conventional devices. This work presents the development of a prototype of an impact rain gauge based on a very low-cost piezoelectric sensor. The sensor was developed by assembling off-the-shelf and reused components following an easy prototyping approach; the calibration of the relationship between the different properties of the voltage signal, as sampled by the rain drop impact, and rainfall intensity was established using machine-learning methods. The comparison with 1-minute rainfall obtained by a co-located commercial disdrometer highlights the fairly good performance of the low-cost sensor in monitoring and characterizing rainfall events.