Author(s): Mohammed I. I. Alkhatib; Amin Talei; Tak Kwin Chang; Andreas A. Hermawan; Valentijn Pauwels
Linked Author(s): Amin Talei
Keywords: No Keywords
Abstract: This study proposes a novel acoustic rainfall sensing approach based on the machine learning technique, Gaussian Process Regression (GPR). Rainfall audio data is collected for 15 rainfall events from five different spots in an urban environment in Subang Jaya, Selangor, Malaysia. These spots are close to a weather station recording the rainfall data with 1-min resolution. In total, 40 acoustic features were extracted from the rainfall audio data, of which 7 were selected through cross-correlation analysis and visualization methods. Using the 7 selected acoustic features as inputs, the calibrated GPR model estimated rainfall with R2=0.784, RMSE=0.270 (mm/min), and MAE=0.191 (mm/min). It was concluded that the proposed model could potentially be utilized for rainfall crowdsourcing through citizens' science.
Year: 2022