Revolutionizing understanding of hydrologic processes: harnessing computer vision for advanced flood forecasting in urban environmentsproject period: 3 1 2024 through 2 27 2029MOST in-situ surface…

The U.S. Geological Survey awarded a $499,999 Cooperative Agreement under the Assistance to State Water Resources Research Institutes program (CFDA 15.805) to Cornell University to develop a distributed sensor network that integrates edge-AI and photogrammetric techniques to automate real-time urban flood monitoring and mapping. Key deliverables include:

1. A minimum viable product (MVP) sensor network to map urban pluvial flood extents and depths in real-time, including storm sewer stage sensors, LoRaWAN gateways, and flood mapping sensor nodes.

2. A dataset containing 3D photogrammetry models, sensor data, and flood event information to support hydraulic and hydrologic analysis for USGS flood monitoring projects in Madison, WI and Long Island, NY.

The project aims to provide urban officials with rapid flood alerts and automate the process of urban pluvial flood mapping to enhance risk characterization and mitigation. Syracuse University is a sub-awardee supporting runtime management protocol development for the sensor network.

Revolutionizing understanding of hydrologic processes: harnessing computer vision for advanced flood forecasting in urban environmentsproject period: 3 1 2024 through 2 27 2029MOST in-situ surface water sensors measure depth, which is combined with elevation data to infer inundation extents across the landscape. This project pilots a low-power sensor network that autonomously observes inundation extents directly using cameras, edge-AI, and photogrammetry. The sensor network provides real-time alerts of the location and nature of urban flood events, and automatically generates spatially continuous high-water mark maps. The data generated by this sensor network will be used to validate urban hydrologic models of flood risk, and train algorithms to detect flooding in satellite imagery. Activities to be performed include:1. Prototype adaptive flood detection and mapping techniques that integrate edge-AI andphotogrammetric principles into a distributed urban hydrologic sensing network to automate highwater mark mapping.2. Utilize lorawan and cellular communications to synchronize different sensor nodes to create anurban flood monitoring platform that utilizes minimal power and incurs low communication costsunder non-emergency conditions, and automatically collects and transmits response-critical dataduring active flood events.3. Install urban flood monitoring networks overlapping with USGS ngwos infrastructure in madison, WI and USGS flood modelling efforts in long island, NY with remote power and robust telemetryinfrastructure.4. Demonstrate how incremental learning can reduce errors over time, and create camera-based sensorplatforms that collect information not imagery, protecting citizen privacy from unnecessarysurveillance.5. Utilize sensor data to validate accuracy of pluvial flood maps from urban hydrologic models andsatellite data, providing insight into flood dynamics that extends beyond the sensor footprint.expected outcomes of this research include design specifications of a distributed sensor networkthat synchronizes storm sewer stage monitoring with a camera-based flood mapping platform to (i) alertofficials of flooding for rapid response and (II) automate the process of urban pluvial flood mapping forplanning and mitigation. This sensor network will produce data on urban flooding occurrence and extentthat will support analysis on pluvial flood risk dynamics that will generalize beyond the study area. Specific deliverables associated with the sensor network, its data, and analysis include:1. Creation of a minimum viable product (MVP) for a camera-based sensor network to map urban pluvialflood extents and depths in real time (floodviz sensor node). 2. Creation of a minimum viable product (MVP) for the urban flood network, including sensor installation and calibration protocols, lorawan network backend architecture and protocols. 3. A dataset supporting hydraulic and hydrologic analysis for two ongoing USGS flood monitoring projects, including 3D photogrammetry models of the landscape, sewer system, and pipe network and georeferenced infrastructure data layers to support urban hydrologic sensor prototyping and testing. 4. Manufacture and installation of six urban flood network sensor node neighborhoods, each including a storm sewer stage sensor, lorawan gateway, six urban flood mapping sensor nodes, and power supply. 5. Flood event datasets, consisting of spatially continuous high-water mark maps. 6. Publications in the peer-reviewed literature and open-access technical journals. 7. Annual presentations of project results professional meetings. The intended beneficiaries of this project include urban residents, emergency managers, and the urbanhydrologic research community. Subrecipient carnegie mellon university supports student-leddevelopment of runtime management protocols for the urban flood network.