The Pluviophone: Measuring Rainfall by Its Sound

PP electronic synoptic (a), with the adjustable wave guide schema on left-hand side, and a schematic view of the experiment (b). It is simultaneously raining both on the meteorological station and the canopy tree. Both the meteorological station and the PP are transmitting data to the computer.

Diagram showing the overall principle of our analysis for a pass-band analysis. Starting from a fully recorded spectrogram (a), we extract a pass-band time series (here at the resonance frequency of the PP guide) and smooth it here with an 8 min moving window (b); we compare it to the rainfall reference measurements similarly smoothed (c), and get a regression line to which a determination coefficient is assigned for the corresponding 8 min smoothing (d); and the R² coefficient variations along to the smoothing parameter (e) is quantitatively illustrating the final assessment of the PP.

Spectrogram (a), and 8 min smoothed pass-band time series (b) of the field experiment. They correspond to curves a, b, and c of Fig. 2, respectively. Rainfall rates (in black) are superimposed on the time domain (plain gray), 1818–2018 Hz (dashed gray) and 1718–2118 Hz (dash-dotted gray) pass-bands (in mV²). Time series are sometimes multiplied by constants to be superimposed.

Final error variations of the PP during the field experiment (a), and example of prediction of the rainfall rate (b), with the corresponding correlation (R² = 0.84*** for a 15 min smoothing). They correspond to curves (d) and (e) of Fig. 2, respectively. The RMSE (a) between rainfall rates and PP records are displayed as a function of the moving window size (in minutes). The time domain (plain gray), 1818–2018 Hz (dashed gray) and 1718–2118 Hz (dotted gray) pass-bands are shown with their associated uncertainties. Observed and predicted rainfall time series (b) are shown in plain and dashed lines, respectively.