Measuring heavy rain where it falls: the 25square sensor
One capacitive rain sensor per quarter square kilometre. We built hardware, firmware and the data path for the mFUND project 25square.
Weather models are only as good as their ground truth. For street-level heavy-rain nowcasting that ground truth did not exist in 2020: radar sees clouds from above, and one or two municipal gauges miss a rain cell that hits six blocks entirely. The research project 25square (mFUND, German federal ministry of transport) set out to change that with one sensor per quarter square kilometre, roughly 580 units for Bochum.
Our part in the consortium with Okeanos and the Bochumer Institut für Technologie was the sensor itself: system architecture, hardware, bare-metal firmware and the path from sensor to backend. Instead of a tipping bucket or a piezo disc we measure capacitively. A freely oscillating impact plate above a fixed base plate forms a capacitor, every drop detunes an oscillator, and an STM32 tracks the frequency shift with nothing more than its timers. Frequencies are robust to measure. Microvolts are not.
The sensors run on PoE or LoRa, speak MQTT and deliver events rather than integrated sums. The measurement principle is patent pending. What we took away from it is in every DAQ project we build today: digitise as close to the source as possible, and design the electronics so that 580 units survive without a maintenance schedule.
The full write-up with circuit details, firmware stack (lwIP, MQTT) and field results is on Stephan Bökelmann's blog.