Determining the intensity of rainfall using the analysis of sound frequencies resulting from the impact of raindrops

Knowing the intensity and duration of rainfall can be useful in many environmental analyses, including the estimation of rain erosivity and soil erosion. There are various devices to record the intensity and duration of rainfall, but purchasing and maintaining them are costly and often requires an operator to take care of them. The present research deals with the feasibility of using the analysis of sound signals caused by the collision of droplets with surfaces and objects in nature to determine the intensity and duration of rainfall. For this purpose, in the laboratory of the Department of Soil Science, Faculty of Agriculture, University of Tabriz, in 2022, rain simulators were designed to produce rains of different intensities, then, the sound signals caused by the impact of raindrops with the metal tray that was placed under the rain were recorded and transferred to the computer for processing. Then, the frequency size of audio files was extracted in MATLAB software. The results showed that with the increase in rainfall intensity, the audio amplitude and frequency size of the audio signals increased. Then, the frequency measurements were automatically placed in two clusters in SPSS software using the two-stage clustering method. Then the mean and standard deviation of each cluster were calculated and according to the correlation of each with each other and with the intensity of rainfall, and in order to avoid the multi-collinearity phenomenon, only the average of the second cluster was used as the input of gene expression programming and linear regression models. In order to test the accuracy and correctness of the results obtained from the models, the coefficient of determination (R2), root mean square error (RMSE), geometric mean of error ratio (GMER), geometric standard deviation of error ratio (GSDER) statistics were used. The values of R2, RMSE (mm/h), GMER(mm/h) and GSDER (mm/h) for the gene expression programming model in the training series data were 0.97, 1.85, 1.11 and 1.09 respectively and for the validation series data were 0.96, 2.05, 1.14 and 1.12 respectively. While the values of the above criteria in the regression model were 0.94, 2.74, 1.25 and 1.34 respectively for the training series data and 0.92, 2.91, 1.28 and 1.37 respectively for the validation series data. The results of the above statistics indicate that the gene expression programming model is relatively more accurate than the regression and overestimation model, and the estimated data of the regression model is relatively more spread than the gene expression programming model.