Accurate and early detection of Localized Heavy Rain by integrating multivendor sensors in various installation environments

Abstract

In this study, we focus on the accurate and early prediction of Localized Heavy Rain (LHR) using multiple sensors. Traditional sensors, such as rain gauges and radar, cannot detect LHR until cumulonimbus clouds cover the sensors. In contrast, Surface Meteorological Monitoring Networks (SMMNs) can accurately measure rainfall in the vicinity of the sensors, thereby detecting LHR earlier than traditional sensors. By evenly placing the sensors around a large city, a SMMN should be useful in predicting LHR. However, since most sensors are placed in a different installation environment, their raw sensor data may significantly differ depending on their surrounding environment (i.e., altitude and sky view factor). Therefore, we propose a calibration scheme for a SMMN that utilizes many sensors in various installation environments and implement a novel LHR prediction system that produces accurate and early LHR predictions. Our system proved to accurately predict LHR 30 minutes earlier than traditional schemes.