Bayesian Fusion for High-resolution Rainfall Sensing Using Pervasive Sensors

Recently, “smart” urban infrastructure has been suggested as a strategy to mitigate social and economic risks posed by extreme weather events and to improve the sustainability of urban centers. Smart infrastructure is enabled by real-time sensing, modeling, optimization, and actuation, which permit the infrastructure to reason about the world around it and to autonomously reconfigure in order to meet changing needs. For example, researchers are exploring the use of smart sewer systems that combine real-time measurements of urban rainfall patterns with runoff models and remote actuated control structures to optimize the conveyance capacity of the sewer network during wet weather. However, a significant barrier to the design and deployment of smart infrastructure is the inability of current sensing technology to characterize the environment at sufficient space and time resolutions to support the forecasts necessary to enable autonomous near-real-time decision making via predictive control algorithms. In the case of urban flooding, extrapolation from a recent study indicates that the minimum resolution for rainfall observations is on the order of 0.5 km and 1 minute. Unfortunately, this observational resolution is not feasible with traditional sensing technologies. Currently, information from multiple remote and embedded sensors, including satellites, weather radar stations, and rain gauges, is combined to achieve precipitation observations over broad areas. However, the resolution achievable by these methods is still too coarse to support many physics-based models that relate the rainfall process to actionable forecasts (e.g., flood depths).