Pervasive Sensing for Real-Time Rainfall Quantification

The proliferation of wireless sensors in everyday consumer products presents new opportunities to monitor the environment at unprecedented space and time scales. This study explores the utility of pervasive sensors for improving the resolution of areal precipitation estimates through fusion with weather radar observations. While these sensors are not specifically designed to measure rainfall intensities, the data they collect can be repurposed to provide quantitative measurements of environmental variables at the location of the sensor. Due to different measurement accuracies (which may be time dependent), types of spatial and/or temporal measurement support, and measurement frequencies of the component sensors, it is unclear how best to combine measurements from pervasive sensors with those from traditional sensors. The method developed in this study employs Markov random field models to compute the likelihood of rainfall at sub-grid pixels. These likelihoods are used to "unmix" the block-averaged rainfall rate measured by the radar. The statistical nature of the model permits the data evidence to drive the fusion of the sensors' measurements. The performance of these methods will be illustrated using case studies exploring synthetic and real-world data.