Community Flood Response & Watershed Resilience through Sensors and Data

We propose a technological approach to storm response and watershed resilience based upon massive sensor networks, and present lessons learned with regards to long-term operations, real-time alerts, and data-driven visualizations. The presence of a single real-time streamgage can help response crews determine which roads and facilities most are at immediate risk (Figure 1). However, when scaling up, network maintenance can be prohibitive to ensuring the network stays fully operational while minimizing downtime. Starting in 2020, we began operating a network of 150+ wireless water level monitors in collaboration with communities throughout four states across the Great Lakes. Upon installation, high precision GPS was used to survey each site (e.g. sensor and streambed elevations), which was essential for correlating stream levels with the elevations of assets at risk of flooding. The custom enclosures were designed to be fully detached from a fixed bracket, so that in the event a monitor needed to be replaced, it could be done quickly without the need for resurveying. When compared with nearby USGS gages, the sites maintained Pearson correlation coefficients exceeding R2 > 0.95 before and after swaps. Based on over one hundred swaps across our network of 150+ monitors over the past few years, minimizing time-consuming field tasks such as resurveying has reduced the expected time at a given site from an hour (e.g. for reinstallation) to approximately ten minutes. We also explore example use cases where continuous stream monitoring has benefited local communities. In two instances, we were able to locate where ice jams built up and alert grounds crews, ensuring nearby property was protected. The monitors subsequently captured the subsequent waves as the jams broke and also provided datapoints for estimating the stream velocity of the channel. In addition, in response to an intense 30-minute 10-year storm event, we observed that a project upgrade met its design goals while crews were still on site. (Figure 2). The data has also proven useful in instances where monitors upstream and downstream have helped identify non-point sources contributing to unexpected runoff. Ultimately, when coupled with asset information and digital elevation models, there is promise for real-time visualizations such as flood maps to help users identify the setpoints at which they would like to be notified. As more data is collected, we have begun exploring the use of learning rainfall-runoff models to forecast water levels. We close by exploring these opportunities and present our preliminary results.