Abstract
Floods are the most common type of natural disaster, often leading to loss of lives and properties in the thousands yearly. Among these events, urban flash floods are particularly deadly because of the short timescales on which they occur, and because of the population density of cities. Since most flood casualties are caused by a lack of information on the impending flood (type, location, and severity), sensing these events is critical to generate accurate and detailed warnings and short term forecasts. However, no dedicated flash flood sensing systems, that could monitor the propagation of flash floods, in real time, currently exist in cities. In this paper, first, a new sensing device that can simultaneously monitor urban flash floods and traffic congestion has been presented. This sensing device is based on the combination of ultrasonic range finding with remote temperature sensing, and can sense both phenomena with a high degree of accuracy, using a combination of L1-regularized reconstruction and artificial neural networks to process measurement data. Second, corresponding algorithms have been implemented on a low-power wireless sensor platform, and their performance in water level estimation in a six months test involving four different sensors is illustrated. The results demonstrate that urban water levels can be reliably estimated with error less than 2 cm, and that the preprocessing and machine learning schemes can run in real time on currently available wireless sensor platforms.