A novel approach to rainfall measuring: methodology, field test and business opportunity

Being able to measure rainfall is crucial in everyday life. The more rainfall measures are accurate, spatially distributed and detailed in time, the more forecast models - be they meteorological or hydrological - can be accurate. Safety on travel networks could be increased by informing users about the nearby roads’ conditions in real time. In the agricultural sector, being able to gain a detailed knowledge of rainfalls would allow for an optimal management of irrigation, nutrients and phytosanitary treatments. In the sport sector, a better measurement of rainfalls for outdoor events (e.g., motor, motorcycle or bike races) would increase athletes’ safety. Rain gauges are the most common and widely used tools for rainfall measurement. However, the existent monitoring networks still fail in providing accurate spatial representations of localized precipitation events due to the sparseness. This effect is magnified by the intrinsic nature of intense precipitation events, as they are naturally characterized by a great spatial and temporal variability. Potentially, coupling at-ground measures (i.e., coming from pluviometric and disdrometric networks) with remote measurement (e.g., radars or meteorological satellites) could allow to describe the rainfall phenomena in a more continuous and spatially detailed way. However, this kind of approach requires that at-ground measurements are used to calibrate the remote sensors relationships, which leads us back to the dearth of ground networks diffusion. Hence the need to increase the presence of ground measures, in order to gain a better description of the events, and to make a more productive use of the remote sensing technologies. The ambitious aim of the methodology developed in this thesis is to repurpose other sensors already available at ground (e.g., surveillance cameras, webcams, smartphones, cars, etc.) into new source of rain rate measures widely distributed over space and time. The technology, developed to function in daylight conditions, requires that the pictures collected during rainfall events are analyzed to identify and characterize each raindrop. The process leads to an instant measurement of the rain rate associated with the captured image. To improve the robustness of the measurement, we propose to elaborate a higher number of images within a predefined time span (i.e., 12 or more pictures per minute) and to provide an averaged measure over the observed time interval. A schematic summary of how the method works for each acquired image is represented hereinafter : 1. background removal; 2. identification of the rain drops; 3. positioning of each drop in the control volume, by using the blur effect; 4. estimation of drops’ diameters, under the hypothesis that each drop falls at its terminal velocity; 5. rain rate estimation, as the sum of the contributions of each drop. Different techniques for background recognition, drops detection and selection and noise reduction were investigated. Each solution has been applied to the same images sample, in order to identify the combination producing accuracy in the rainfall estimate. The best performing procedure was then validated, by applying it to a wider sample of images. Such a sample was acquired by an experimental station installed on the roof of the Laboratory of Hydraulics of the Politecnico di Torino. The sample includes rainfall events which took place between May 15th, 2016 and February 15th, 2017. Seasonal variability allowed to record events characterized by different intensity in varied light conditions. Moreover, the technology developed during this program of research was patented (2015) and represents the heart of WaterView, spinoff of the Politecnico di Torino founded in February 2015, which is currently in charge of the further development of this technology, its dissemination, and its commercial exploitation.