Designing Adaptive Sensing Policies for Meteorological Phenomena via Spectral Analysis of Radar Images

In this work we consider the problem of determining efficient scanning strategies for adaptive networks of radars used to monitor meteorological events. While most meteorological radars traditionally scan 360 degrees, repeatedly, more accurate readings can be obtained if adaptive sector scanning is performed. We propose a scanning strategy based on analyzing eigenimages computed given a set of observed radar images. These eigenimages allow us to identify spatial regions with high radar measurement variance, and thus provide us with the necessary information to reconfigure radar resources towards regions where the storms’ states seem to be changing the fastest. These are exactly the regions where newer and more accurate readings are usually necessary. We show that performing adaptive reading in this manner provides gains over the traditional sit-and-spin approach, and also over a baseline random scanning strategy. We also show that there is a difficult trade-off between the minimum size of storm that one might want to track, and the capability of saving scanning resources by not focusing on irrelevant sectors.