Hurricane Flood Models

chevron-icon
Back
project-presentation-img
Presentation
Expert Review🌟
photrek
Project Owner

Hurricane Flood Models

Funding Requested

$150,000 USD

Expert Review
Star Filled Image Star Filled Image Star Filled Image Star Filled Image Star Filled Image 4
Community
Star Filled Image Star Filled Image Star Filled Image Star Filled Image Star Filled Image 4.3 (8)

Overview

Photrek with Kenn Mayfield and John Lawson will develop Hurricane Flood Models leveraging risk-aware machine learning, 3D immersive interactive environments, and computational meteorology. There is a critical, immediate need for better flood models because of the heightened risk of severe hurricanes and cyclones. The transition from El Niño to La Niña in 2024 has created record ocean temperatures and triggered anomalous westerly trade winds. We will build from the Bangladesh datasets Kenn Mayfield has modeled and extend the methods to the Mississippi delta and other large river systems.

Proposal Description

How Our Project Will Contribute To The Growth Of The Decentralized AI Platform

Photrek and Kenn Mayfield’s teams have successfully launched risk-aware ML and 3D interactive applications on the AI Marketplace. Together these teams will push towards a commercially viable capability that addresses an urgent societal need. We will launch an application that models flooding with user accessible interaction, rigorous scientific analysis and robust engineering methods.

Our Team

This project brings together three exceptional efforts related to first responder toolsets.  Nelson and Attieh from Photrek have demonstrated the ability to deliver innovation for SingularityNET. Mayfield will join Photrek as a Working Partner to incorporate generated imagery into his immersive environments. Lawson is a computation meteorologist with expertise in information-theoretic evaluation of forecasting. Photrek will employ a strong team of developers to build and host the application.

View Team

AI services (New or Existing)

Coastal Flood Model

Type

New AI service

Purpose

Photrek will host a Coastal Flood Model service that produces flood assessment reports following severe coastal storms. The service will consist of 3D immersion utilizing flood map data generated imagery and meteorological reporting. As possible the service will utilize prior AI Marketplace apps developed by Photrek and Mayfield.

AI inputs

Photrek will seek to interface the Coastal Flood Model app with the following existing apps: - Coupled VAE Data Generator - generates images - Risk Assessment - assesses the Robustness Accuracy and Deciveness of forecasts - Flood Risk Visualization In Complex Virtual Worlds

AI outputs

N/A

Company Name (if applicable)

Photrek

The core problem we are aiming to solve

The great river deltas such as the Ganges in Bangladesh and the Mississippi in the US are experiencing anomalously high risks of flooding due to the impacts of global warming. The transition in 2024 from an El Niño to a La Niña pattern has compounded this risk due to record ocean temperatures and anomalous trade winds.   As reported in Live Science (Pappas, 2024), “During La Niña, the upper atmosphere winds calm, reducing wind shear. This allows the convection of warm, moist air from the ocean surface to form big storms.”

While humanity has mitigated some of the worst risks associated with hurricane and cyclone wind damage through a combination of improved forecasting and better coastal construction, the risks associated with flooding remain poorly modeled and difficult to mitigate. Both public and commercial first responders need actionable information about damage assessment following a severe flood storm. Photrek and its partners will provide an interactive application with graphical and analytical tools regarding flood impacts.

Our specific solution to this problem

Photrek and its partners will provide flood damage impact reports that incorporate 3D interactive graphics, risk-aware generated imagery, and meteorological analysis. The 3D immersion will include AI Avatars’ representing citizen stakeholders, public first responders, and commercial resources.  Mapping data will be selected and imported then converted into first-person explorable topology of the water coverage, roads, buildings and vegetation.

Actual imagery will be used to train generative models based on Photrek’s risk-aware data generation.  The generated imagery will be used to augment the 3D visualization. Photrek is currently enhancing the resolution of its generative algorithms under the Simulating Risky Worlds project in order to support generation of satellite and related imagery.

The imagery and mapping will be part of comprehensive meteorological reports.  Information-theoretic assessments of forecasts using Photrek’s Risk Assessment will be use to select high-quality sources for the reports.

Photrek will employ a comprehensive group of developers, systems administrators, program managers, and community engagement staff to ensure that we are able to deliver a valuable capability and host the application on the SingularityNET AI Marketplace.

Project details

Photrek proposes a vital new toolset capable of generating interactive flood maps to assist first responders in the event of an unusually energetic hurricane's landfall. The capability will provide interactive reports and visualizations that assist in forecasting and assessing flood damage.

Historic Crisis Looming

Figure 1 (McGrath et al., 2024) shows that 2023 saw record-breaking ocean temperatures well above the trends of preceding years from 1979 to 2022. 2024’s ocean temperatures are expected to peak above 21 degrees Celsius and remain high throughout the season. Adding urgency to the situation is the transition to a La Niña climate pattern which will direct hurricanes toward the Central and North American coasts. These factors likely influence Penn State University’s prognosis of ~33 named storms, which would be a record-breaking season (Mann et al., 2024).

Existing flood prediction algorithms work upon principles of averaging that tend to disregard or minimize the importance of outliers, however these new energetic storms themselves represent “outlier outliers” - 200 year storms rather than 1% or 1 in 100-year storms are now expected more frequently due to climate change. In this new context, existing models are not accurate enough to usefully predict flooding effects and damage to infrastructure.