The Japanese Road Traffic Law dictates that drivers must prevent making 'water splashes', to avoid pedestrians' clothing and belongings getting wet. However, drivers sometimes mistakenly make water splashes at night or in rain. It is effective to share information about water splashes between drivers to help drivers avoid water splashes easily. To collect the watersplash information, it is insufficient just to detect puddles or their causes, such as cracks, potholes, or ruts, because water splashes occurs not only on puddles but also on flat road surfaces where water flows. In this paper, we present a method to detect actual water splashes caused by vehicles. We use an acoustic sensing method that classifies sound recorded inside the vehicle into two classes, which are 'splashing' and 'non-splashing', with supervised learning. We achieved an F measure of approx. 90% and confirmed the effectiveness. Additionally, we confirmed that acoustic features focused on a low-frequency range are effective.