OPT/NET’s AI for Flood Detection
Implementing OPT/NET’s EOD analysis product — TSAR AI for Flood Detection
This article is based on an extended paper that evaluates TSAR AI’s efficacy as a tool for flood detection & emergency response, and draws a comparison to the current standard operating procedures.
Please contact us to request the full paper.
Introduction
In recent years, we have increasingly experienced the adverse effects of climate change and extreme weather phenomena. Globally and at the European level, floods remain the most common type of natural disaster, causing extensive physical damage (i.e. to housing, infrastructure, vehicles, housing, crops, etc.), casualties (deaths and/or injuries) and leading to significant disruptions in biodiversity and the overall ecological balance (contamination and damages to the ecological system). Some countries are more prone to disasters and the related fallout, because they are more exposed to physical, social, economic, and environmental vulnerabilities.
“In the future, climate change will induce even more severe hurricanes. Not only should these be better understood, but there is also a necessity to improve the assessment of their impacts. Flooding is one of the most common powerful impacts of these storms. Analyzing the impacts of floods is essential in order to delineate damaged areas and study the economic cost of hurricane-related floods.”
—
ESA, Flood Monitoring
Therefore, public authorities have been under pressure to develop reliable and accurate flood risk maps — which are the essential tools for land use planning in flood-prone areas, and the building blocks for plans regarding sustainable flood risk management protocols emphasising prevention, protection and preparedness. In that sense, Artificial Intelligence (AI) has gained popularity due to its established capability for obtaining accurate prediction models that contribute to water resource management strategies, policy suggestions and analysis, and further evacuation modelling. In fact, AI-driven tools based on timely and accurate Earth Observation Data (EOD) have been successful in crisis management for years.
OPT/NET’s approach
At OPT/NET, we deliver the next generation of AI systems: a hybrid platform which combines the processing & automation capabilities of AI with the natural problem-solving abilities of humans. Our patented AI approach autonomously detects patterns of significance in any time-series data without necessitating the onerous task of labelling massive training datasets.
OPT/NET’s EOD analysis platform, TSAR AI , integrates AI Knowledge Packs (AI KP), which are available for specific use cases such as flood detection. You can think of an AI KP as a “book” containing domain-specific expertise, ready to be plugged in when and where needed. The flood detection Knowledge Pack is capable of high quality flood forecasting that considerably outperforms conventional approaches such as Object-Based Image Analysis or the manual labels generated by trained human resources. Dedicated AI KPs are gleaned from integrated Analysis Ready Data (multi temporal Synthetic Aperture Radar (SAR) — Sentinel-1 or TerraSAR-X provided by Copernicus and Airbus, respectively) allowing for spatial flood prediction. The platform empowers managers and decision makers to interpret massive datasets even if they do not possess advanced data science skills, and to act decisively during emergency situations in near real-time.
The experience acquired by OPT/NET in flood detection projects leads us to objectively conclude that SAR imagery should be used for weather-related disasters, which simply cannot be tackled with optical imagery alone. We believe change detection based on multi-temporal SAR images and AI technologies are to be progressively utilized for disaster monitoring, especially for flood events.
Use Case — TSAR AI for flood detection in Houston area after Hurricane Harvey
We present a use case centred around Hurricane Harvey, which swept through Houston in 2017. We made use of Sentinel-1 SAR multi-temporal data, and employed our change detection approach using the TSAR AI platform to generate a flood map which was then compared to a reference map for evaluation purposes.
As Harvey swept through Texas, an estimated 13 million people were affected, nearly 135,000 homes damaged or destroyed in the historic flooding, and up to a million cars were wrecked.
According to the United States Geological Survey (USGS), Hurricane Harvey was the most significant rainfall event in U.S. history since records began in the 1880s. Hurricane Harvey made landfall near Rockport, Texas, on August 25, 2017, as a Category 4 hurricane with wind gusts exceeding 150 miles per hour.
As Harvey moved inland, the forward motion of the storm slowed down and produced tremendous rainfall amounts over southeastern Texas, with 8-day rainfall amounts exceeding 60 inches in some locations, which is about 15 inches more than average annual amounts of rainfall for eastern Texas and the Texas coast.
TSAR AI Evaluation
In this use case, we validated that the TSAR AI solution is reliable and effective at generating a rapid mapping product, exhibiting promising results which are comparable to trained human analysts. Crucially, it only takes TSAR AI around 15 minutes vs. days (for a team of analysts) to classify the terrain and detect which areas are flooded. Unlike typical AI approaches, ours eliminates the need for time consuming labelling tasks and model building. This is a significant advantage as the generation of manually labeled ground-truth maps is typically very time consuming, tedious and error prone. For the Hurricane Harvey case, it took a team of analysts nearly 1 full-time week to produce a reference ground-truth map covering an area of ~1km2 (which still contains a large degree of uncertainty due to the occlusions by various objects such as buildings and trees). Producing a truly ‘100% true’ ground truth is impossible in this case without close up in-situ inspections.
Considering the reference map defined for a subset located at Brazoria County produced through visual interpretation of National Oceanic and Atmospheric Administration (NOAA) aerial imagery, we conclude that TSAR AI achieves up to a 99% total classification accuracy with ‘zero’ false positives and 100% precision, and identifies 60 km2 of flooded and 8 km2 of non-flooded areas (excluding normal water, rivers and pounds).
As our quest to acquire new insights (in this case, generated by our AI KP’s) builds, so does our vision for accurate EOD to be utilized across a wide variety of practical applications to help protect and secure the environment from extreme events. The TSAR AI platform is a highly reliable product that enables constant and effective monitoring of very large and flood-prone regions in near real time, that also accounts for variability of Land Use and Land Cover. In this particular case, for example, it is possible to distinguish flooded areas along the borders of the Brazos river (evident in several cross sections of the river, in particular those located to the Northwest and in the center of the subset), flooded croplands, and detect water between Palustrine Forested Wetland, Palustrine Emergent Wetland and Pasture/Hay. As such, another advantage of TSAR AI is evident insofar as its capability for detecting water in landscape environments that also contain forest and/or shrub patches. This minimizes the ‘double bouncing’ between forest stems and the water layer on the forest floor itself.
The Hurricane Harvey case demonstrates the effectiveness of TSAR AI for the flood detection task in Texas, United States. However, it is not limited in efficacy to that region, and can easily be transposed to any territory in the world and begin quickly and efficiently processing large time-series of pre- and post-event images (EOD) to evaluate the impacts and resilience of the affected area(s).
Conclusion
TSAR AI demonstrates a high maturity both as a technical project and as a business structure. Applying our AI approach to EOD provides drastically improved speed and efficiency of emergency response protocols. OPT/NET believes that it is in an unmatched position to provide clients with the insight they need to make informed flood-risk and climate adaptation decisions today and decades into the future.
About OPT/NET
OPT/NET B.V. is a team that builds comprehensive AI products based on decades of hardcore critical industry experience. Having served and protected the networks of clients all over the globe, the OPT/NET team was specially suited to develop a series of advanced AI products capable of dealing with an ever-increasing volume of data and complexity. Initially serving as a tool for our own telecom consulting practice, the OPT/NET AI engine has grown into a series of stand-alone platforms with unlimited potential across a variety of critical and data-intensive industries.
We believe in making humans superhuman, not replacing them. The OPT/NET AI Engine provides domain experts in both structured and unstructured data environments the ability to rapidly develop advanced real-time AI solutions that help them do their jobs more effectively, without requiring an advanced degree in data-science. Importantly, our solutions are human-driven, AI-assisted.
For more information visit www.opt-net.eu
