What Hurricane Dorian Taught Us About Flood Risk vs. Wind Risk
Empowering Organizations to Make Better Operational Decisions and Predict Future Weather-related Risks
“Floods have always been one of the more challenging weather-related events for insurers to assess. CrowdAI’s insights after Hurricane Dorian can do a great deal to increase confidence in underwriting. “ — Nick Lamparelli, Co-founder and CUO of reThought Insurance
Hurricane Dorian formed on August 24, 2019, wreaking havoc on hundreds of thousands of people — from the Caribbean all the way to Canada. The Bahamas experienced catastrophic damage when Dorian made landfall as a CAT 5. By the time Dorian reached landfall in Cape Hatteras, NC it was downgraded to a CAT 1. However, even at a lower strength, significant damage was still done, especially to a 12-mile stretch of Ocracoke, NC. Storm surge reached seven feet, flooding roads and buildings. The residents of Ocracoke were in disbelief at the speed at which the water flooded the island — even in areas that historically had never flooded before.
Despite the potential for this amount of devastation, more than a third of homeowners in the path of Dorian dropped their flood insurance policies within the past ten years. Without this insurance on a family’s most valuable asset, they can be left with nothing, having to pay out of pocket and possibly start their savings from scratch.
So why are homeowners abandoning flood insurance?
One possible explanation is that people systematically undervalue the impact of low-probability, high-risk events. In 2012 the Biggert-Waters Act was passed by Congress as a remedy to the tremendous financial problems plaguing the National Flood Insurance Program (NFIP), following years of unprecedented flood events in the U.S. Provisions called for premium rate increases of up to 25% per year, making flood insurance unaffordable for many consumers.
There has been a steady decline of NFIP’s policies-in-force since its high point of 5.7 million in 2009 and total annual premium remains under $4 billion.
Hurricane Dorian is not an isolated incident. The losses from storms continue to rise as a result of their increased frequency and increased severity, compounded by the increase in population in the path of storms.
There is more data than ever before about the impact storms like Dorian have on homes and infrastructure, and it presents a unique opportunity for artificial intelligence to help make sense of it all.
Alan Demers, President of InsurTech Consulting and former VP of Claims Innovation and Technology at Nationwide shares his perspective, “It is crucial for insurers to accurately separate flood from wind damage, since rising water damage is excluded from most homeowner policies. There are significant legal exposures insurers face following such events which can be avoided and mitigated by applying data, information and documenting the presence of flood waters and flood damages from other damages. There have been numerous lessons learned from events such as hurricane Katrina that led to a battlefield of wind vs. flood litigation.”
CrowdAI uses deep learning technology to empower organizations to make better operational decisions before and after a catastrophic event.
Before a storm: seamlessly identify major contributors to exposure, create an expected claims value for varying degrees of damage, and establish an action plan to streamline post-claim response.
After a storm: quickly triage on-the-ground resources and make more educated financial decisions.
Within 24 hours after Hurricane Dorian passed, NOAA (in coordination with FEMA and other state and federal partners) collected thousands of aerial images of the Atlantic coast, covering over 3,000 miles.
CrowdAI is imagery agnostic, which means that regardless of the source (satellite, aerial, drone, video, handheld cameras, etc), the technology is able to extract a wide range of insights about the physical environment, including roof attributes, infrastructure, vegetation, and more.
As soon as imagery from NOAA became available over the Outer Banks of North Carolina, CrowdAI overlayed deep learning data layers to identify flooding and its impact on roads and buildings then classify roads as completely flooded, partially flooded, or not flooded at all. This is a distinction from other machine learning technologies who are only able to provide a binary conclusion (flooded or not flooded; damaged or not damaged).
Flooded roads can be a big problem after a storm. Being able to move people and resources from one area to another is critical for first responders, insurers, and homeowners alike.
In this area of Ocracoke, some roads remain flooded, while others are passable. While it’s not possible to measure the depth of the water from this type of imagery, CrowdAI’s models were trained to identify only the roads in the image and determine if water is present, which could create a hazard.
Red roads are completely flooded, while yellow roads are partially flooded. Blue roads seem to be completely dry, and therefore navigable without issue. Because CrowdAI uses an image segmentation approach, the technology can categorize individual segments of each street in this manner, so it’s possible to tell which section of Maple Street is flooded, for example.
Flood waters can do a great deal of damage to a building, applying pressure to its sides and saturating the ground beneath and around the foundation. If not identified and addressed, flood damage has the value to undermine the structural integrity of a building, an effect that might not be felt for years.
To help understand where the most building damage might have occurred, CrowdAI identified water on land surface across the NOAA imagery, ruling out areas of “known” water, such as lakes and wetlands. By overlaying these “flood water” polygons with building locations, CrowdAI detected which buildings were adjacent to water on one or more sides — this indicates a strong likelihood of flooding.
CrowdAI delivers this data into existing workflows and whatever format is most relevant to the end-user.
For first responders: Triage which areas require urgent assistance and which skills sets are needed based on the severity of the damage.
For event assessment: Insurers and others need to quickly gauge overall exposure for purposes of reserving and other financial needs. Identifying flood impact areas from others with wind only is extremely helpful in early-stage assessment.
For claims: Immediately after a storm, it can sometimes take several days before the insured can file a claim for a myriad of reasons outside their control. Carriers use this data to improve their catastrophe claim response, prioritizing and discerning which properties to inspect or perhaps resolve over the phone through use of damage images. It is crucial for insurers to clearly separate wind (and wind-related rainwater leakage from holes and openings in a structure which are covered by homeowner policies). These insights also improves the accuracy of the payout and reduce time to close a claim, so those in need can quickly rebuild and have one less thing to worry about in an already difficult time.
For underwriters: Store the flood footprints into a database to improve future risk assessment by scoring the addresses against a historical reference. Example: If a home wasn’t flooded in a 100-year flood, high confidence it is not at risk of a future flood even if the property is not in a flood zone.
This is very similar to the work CrowdAI did for one of the largest telecommunications companies in North America after Hurricane Michael and for a multinational insurance company after floods in Western Europe in 2018.
In the coming weeks, CrowdAI will work with carriers whose insured properties were impacted by Dorian to integrate CrowdAI’s conclusions with historical customer data. By marrying these two data sets, insurers will be able to equate the extent of damage to an estimated loss amount. The outcome will produce predictive insights that will help forecast future claims of high wind and flood damage and also improve the underwriting of properties in potentially high-risk areas.
At CrowdAI, we are committed as a team to build technology that is used to make the world a better place. In keeping with that commitment, CrowdAI shares samples of this data set free of charge to nonprofits administering aid. For-profit organizations can send CrowdAI individual addresses of their assets and CrowdAI can send back relevant insights.
Learn more by emailing email@example.com and visiting crowdai.com.
CrowdAI’s mission is to empower organizations to make better operational decisions by forecasting changes in the physical world.
Extract a wide range of assets and produce predictive insights whenever imagery is available using proprietary computer vision deep learning technology.
Built by engineers from Google Maps and IBM Watson, backed by Y Combinator, and trusted by brands encompassing insurance, telecommunications, utility, oil & gas, ride-sharing, government agencies, location intelligence, and more.