Validating Flood Model Simulations Using Camera Information and Crowd Source Information

In this study, we validate the Jupiter Intelligence FloodScore Planning (FSP) flood model’s simulation of three past flooding events with two new datastreams: New York City 311 street flooding complaints and Helios observations in the New York metropolitan area. New York City’s 311 database includes details about millions of 311 complaints since 2010 and updates daily. Additionally, Helios weather analytics from L3 Harris provide real-time weather information from tens of thousands of existing traffic cameras world-wide. We first use a hit ratio to determine the percentage of reported 311 complaints that are located in an area where the Jupiter model estimates maximum flood depth to exceed 0.50 ft. We then used similar hit ratios to determine the agreements made between the Helios observations and the model. The “wet road” hit ratio depicts the percentage of reported “wet road” conditions reported by the Helios dataset that are in an area where the model estimates maximum flood depth to exceed 0.50 ft. The “dry road” hit ratio depicts the percentage of correct rejections made by the Helios dataset and the model (the percentage of reported “dry road” conditions reported by the Helios dataset that are in an area where the model predicts the maximum flood depth to be less than 0.50).

The 311 hit ratios range from 0.6-0.8, where a 1.0 hit ratio represents perfect agreement, demonstrating the model’s ability to capture flooding events in highly urbanized areas. Additionally, both the “wet road” hit ratios and “dry road” hit ratios calculated using the Helios data range from 0.4-0.7. There was one instance where a 311 call was in close proximity to a wet road observation made by Helios, demonstrating close agreement, with Helios providing the temporal information on flooding which can only be inferred from multiple 311 call reports. The use of such datasets have many applications to high-resolution urban flood model validation, each with its own value and characteristic information.