Economics of Flood Damage Claims - Large Events Dominate Overall Losses - Should Effective Mitigation Strategies Focus On "The Big Ones"

What type of meteorologic events dominate flood damage claims? Is it from many frequent small storms or a few infrequent black-swan events? Understanding what size of events cause the most losses can help us focus on the most effective flood loss mitigation measures - this is essential for achieving high returns on investment in flood mitigation strategies. By reducing flood losses in an economically efficient manner, high benefit-cost ratios can be achieved.

This post summarizes the distribution of FEMA's flood damage claims and explores what type of storm events - big or small - govern extreme weather losses.

FEMA summarizes the number of payout claims and the total value of payouts for significant flood events, i.e., those with 1500 or more payouts. The total value of payouts, adjusted for inflation to 2018 dollars, is plotted below against the total number of claims in the events between 1978 and 2017:

FEMA Average Claim Amount for Various Event Sizes (Number of Claims) - Small Significant Flood Events with a Minimum of 1500 Claims Per Event

The median number of payouts is 4115 with a median payout amount of $29,700, adjusted to 2018 dollars. This is over a total of 116 flood events from 1978 to 2017.

Larger storm result in more extensive flood damages and numbers of payout claims as shown on the following chart that labels some of the largest tropical storm / hurricane events:

FEMA Average Claim Amount for Various Event Sizes (Number of Claims) - Small and Large Significant Flood Events with a Minimum of 1500 Claims Per Event

How do the larger events affect the flood damage and payout values? The average flood claim payout of $36,200 is above the median value reflecting the skew in catastrophic event distribution - the right tail of rare black-swan events in the probability distribution of events pulls the average above the median.

Five of the 116 event have claim counts that are over ten to forty times the median number of claims. That is, Hurricane Ike and Irene had over 40,000 claims compared to the median count of just over 4000 claims. And Hurricane Harvey had over 160,000 claims. The losses are greater for these larger events with the best-fit line showing average claim values of over $50,000 to over $120,000 for these largest significant events. What effect do these claim counts have on weighted claim amount - they increase the claim-count-weighted average loss to $60,600 - more than double the median claim amount per event that is not weighted by the number of claims in each event.

So when looking at the economic losses associated with a significant flood event, we need to consider the size of the event. And when we develop strategies and best practices for flood resilient communities and flood risk mitigation, striving for significant damage reduction and return on investment in averted flood damage losses, we must also consider what events cause the most damages. Canada's Disaster Mitigation and Adaptation Fund (DMAF), for example, requires return on investment (ROI) evaluations for eligible risk reduction projects. It would appear that to achieve meaningful flood damage reduction ROI we must target solutions toward events leading to the most damages

Looking at FEMA's significant flood events, data show that 3 of 116 events account for 54% of the total inflation-adjusted damages. Those 3 events are Hurricane Harvey, Superstorm Sandy and Hurricane Katrina. And the top 20 events, each with total event claims of over $500M, account for 81% of the total claims. So it is clear that to reduce the bulk of flood damages we have to consider how to increase resiliency in existing communities during the largest storm events. If we target flood risk reduction for the small catastrophic events, the smaller 97 events, we will be addressing only 20% of the total claim value. So the 80/20 rule, the Pareto principle, does apply to flood damage reduction.

FEMA Inflation Adjusted Significant Flood Event Payout Distribution - Pareto Distribution and the 80/20 Rule

The impact on a few recent large events on damages helps show sample-bias in catastrophic event losses as explored in a previous post. That is, up to 2004 prior to Hurricane Katrina, the distribution of losses based on the 1978-2004 sample of events did not consider the true 'population' distribution of flood events that includes very extreme, right-tail events. As Fleming demonstrated in "Yep, We're Skewed", short samples with high skew underestimate losses of the true population of events.

***

For statistics geeks:

Could it be that the common chorus of explaining recent floods losses as being due to climate change may in fact be explained simply by statistics and larger sample sizes overcoming short sample biases (underestimation)?

Could it be that growth in high risk areas is driving flood damages higher? AON Benfield's review of Hurricane Harvey suggests that growth in at-risk areas explains some flood impacts:

 "Given the volume of water, local infrastructure across southeast Texas was simply unable to handle such an enormous amount of rainfall in a short amount of time. This led to major water run-off that quickly accumulated across a very large area. With so much residential and commercial growth throughout this part of the state – combined with abundant concrete and poor absorbing clay soil –this only worsened the flood impact."

Solutions to flood risk mitigation therefore cannot only be local infrastructure solutions to convey enormous amounts of water but rather land use planning policies to direct development and redevelopment away from high flood risk areas. As AON Benfield notes " Hurricane Harvey’s rainfall reached the 1,000-year rainfall return period based on many time intervals during the course of a number of hours and days.", and it is not cost effective, or technically feasible, to have local infrastructure convey the runoff from events of this magnitude.

***

Canada Connection (for those appreciate tree-sauce, skatey-punchy, and noble antler cows):

CatIQ claim datasets have been used to evaluate flood damages across Canada according to the size of the flood event (i.e., related to the number of claims). A similar pattern of increasing damages with increasing event size and distribution is apparent in the CatIQ datasets. In contrast to the FEMA claims noted above for many hurricane events, the CatIQ data reflect basement flooding claims primarily, as overland flooding has not been insured in the past and is not widely held. What is the magnitude of these Canadian claims? Aviva Canada provided this summary of claim trends and magnitude:

"In 2014, water damage claims accounted for 44% of dollars paid out on all Aviva Canada property damage claims, compared with 39% in 2004. The average cost per residential water damage claim has increased significantly – going from $11,709 in 2004 to $16,070 in 2014, a 37% increase."

So basement flooding damages are significantly less than FEMA's large scale catastrophe claims. CatIQ data shows that for larger events (those with higher claim counts) the average claim amount does increase above the Aviva Canada values noted above. Comprehensive benefit-cost analysis used to develop ROI rankings for flood mitigation projects would apply the lower range of typical damages to frequent to moderate events and the higher damages/claim amounts to the frequent events, factored by their probabilities. The most frequent storms, typically 5 to 10 year return period events as in a recent study by Atkins for the US EPA do not generate damages.

Book Review: The Rightful Place of Science: Disasters & Climate Change - Ontario Cities Flooding Perspective

Roger Pielke, Jr.'s 2014 book  The Rightful Place of Science: Disasters & Climate Change is a must-read for anyone interested in understanding and mitigating flood damages in Ontario cities. It reinforces several local themes presented in this blog as well, e.g., extreme rainfall is not increasing (due to climate change or anything else), flood damages are influenced by other factors and are not increasing as a result of more frequent of extreme rainfall. (see 2017 update at bottom of post, and 1970-2019 hurricane trend).

Pielke has been recently highlighted in the Wall Street Journal where he shares his "Unhappy Life as a Climate Heretic".

Hurricanes

Pielke demonstrates that the frequency and severity or strength of hurricanes (tropical cyclones) has not increased over the past century. This is good news for Ontario where many river flood hazards are defined by Hurricane Hazel as the regional storm regulatory event in many Conservation Authority jurisdictions.

He also delves into damages resulting from these events, normalized to reflect the increase in the number of people and the amount of property in vulnerable areas. While losses have increased it is due to the increase in assets at risk - the normalized damages are in fact 'flat'.

In Ontario, the percentage of properties lie in river flood plains where hazards are governed by hurricane events is in the very low single digits. Nonetheless, decreasing hurricane frequency is a good thing.

Extreme Precipitation

Citing IPCC, Pielke notes that there have been statistically significant decreases and increases in 'heavy precipitation' events, and that there are strong regional and subregional variations in the trends. It is important to note the definition of heavy precipitation is rainfall above the 95% percentile of daily rainfall .. so not really short-duration, high-volume, extreme rainfall that causes widespread urban flooding in Ontario.

As reported on this blog, Ontario has regional trends in annual maximum observed rainfall volumes. Charts at right present Environment and Climate Change Canada's Engineering Climate Datasets Version 2.3 trend data. For short durations, less than 6 hour duration, there are four times more statistically significant decreases in maximum rainfall than increases. So southern Ontario regional trends show decreasing rainfall severity. You could "tease out" a teeny, tiny Ontario-wide increase if you average the whole province together, but that would underestimate the increases up north and misstate the decreases down south.

Normalized Catastrophic Losses in Canada

Intact has reported on damage trends in their 'insurance is evolving' webpage. They indicate that "Payouts from extreme weather have more than doubled every five to 10 years since the 1980s." and provide the chart at the right.

So while losses are increasing, how is GDP increasing in Canada? That is, are normalized losses increasing in Canada? Or are those trends like those identified by Peilke? The website tradingeconomics.com provides this Canadian GDP growth chart to the right. Since the early 1980's, GDP has increased about 500% (approximately $300B USD to over $1500B USD). That could explain a portion of the absolute increase in losses since the early 1980's identified by Intact.

* NEW * using Statistics Canada data for expenditure-based Gross Domestic Product and catastrophic loss data presented by Intact Centre for Climate Adaptation have been used to assess normalized losses as a fraction of GDP. While losses have increased significantly, but the normalized Canadian catastrophic losses are up and down. There is a low r-squared upward trend in normalized losses (i.e., not a strong trend). So using Pielke's 'detection vs. attribution' distinction, one could argue there is a detectable trend upward, however the attribution - what caused it - is unclear. As noted below, we have significant quantifiable upward trends in urbanization in Ontario cities over past decades, suggesting increased runoff stresses under a stationary climate, or extreme weather trends.

 Pielke cites a 2014 IPCC report that notes the following:

"Economic growth, including greater concentrations of people and wealth in periled areas and rising insurance penetration, is the most important driver of increasing losses."

Ironic - some more losses are because of rising insurance penetration. Makes sense though. Here are Canadian cat losses normalized based on personal property net written premiums (according to Facts of the Property and Casualty Insurance Industry in Canada 2015 is published by Insurance Bureau of
Canada (IBC), 2015):

Wow! Not much of a trend there to suggest that extreme weather is getting a lot worse - normalized losses are up and down with the maximum relative losses back in 1997 (note the 2014 and 2015 premiums are assumed to be 2% greater than previous years to extend this from 2013 to 2015 where cat loss information was available; similarly 1987-1989 premiums are assumed to increase at 2% to arrive at the reported 1990 value). The r-squared is very low too (0.012) meaning not a strong trend here over time. This muted trend is in stark contrast to the non-normalized losses typically reported by the insurance industry.

Intersection of Vulnerability and Extreme Events

A refreshing observation in Pielke's book is that it is the intersection of intrinsic vulnerabilities and exposure to extreme weather events that causes damages, like flood losses. More properties and belongings in the wrong place explain increased losses. So I made a fancy graphic to explain it:

I just found this infrastructure climate vulnerability Venn diagram in the PIEVC Engineering Protocol
For Infrastructure Vulnerability Assessment and Adaptation to a Changing Climate PRINCIPLES and GUIDELINES, so I guess we can't copyright it.

Note that PIEVC figure above describes infrastructure-climate interaction. Our figure substitutes infrastructure with development, meaning the property impacted by climate (weather). Where does infrastructure come into it? I suppose here:

The Media and Public Opinion

Pielke notes that the reporting on extreme weather has increased considerably, which feeds a public perception that extreme events are increasing in frequency or severity. For example in the New York Times the phrase 'extreme weather' has jumped in popularity in newspaper articles since the mid 1990's.  In this blog we have commented on the 'availability bias' that such frequent reporting can cause, skewing the public's perception on the true probability of events. A common statistical sin in such reports is to declare a record rainfall ... umm ... for a particular calendar day (July 8, 2013 in Mississauga, Ontario, or September 28, 2016 in Windsor/Tecumseh, Ontario). Once engineers start design infrastructure to operate differently on different calendar days of the year, such reporting of calendar day rainfall records will be worthwhile, saintly even. But sorry, that's nuts and not going to happen any time soon.

Key Take Away - Let's Get Back On Track, Put First Things First, and not be so ideologically driven to predetermined conclusions that we miss the obvious stuff.

I love this quote in the Pielke's book :

"There is such a furor of concern about the linkage between greenhouse forcing and floods that it causes society to lose focus on the things we already know on floods and how to mitigate and adapt to them."

Greater Toronto Hamilton Area Urban Growth Affecting Urban Flood Risk

This is from Flood risk and climate change: global and regional perspectives, Zbigniew W. Kundzewicz, Shinjiro Kanae, Sonia I. Seneviratne, John Handmer, Neville Nicholls, Pascal Peduzzi, Reinhard Mechler, Laurens M. Bouwer, Nigel Arnell, Katharine Mach, Robert Muir-Wood, G. Robert Brakenridge, Wolfgang Kron, Gerardo Benito, Yasushi Honda, Kiyoshi Takahashi, and Boris Sherstyukov, Hydrological Sciences Journal Vol. 59 , Iss. 1,2014.

Overland flow explains basement flood risks,
from "flood plain to floor drain".

That reference abstract also notes "Economic losses from floods have greatly increased, principally driven by the expanding exposure of assets at risk. It has not been possible to attribute rain-generated peak streamflow trends to anthropogenic climate change over the past several decades. Projected increases in the frequency and intensity of heavy rainfall, based on climate models, should contribute to increases in precipitation-generated local flooding (e.g. flash flooding and urban flooding)."

Along those lines, in this blog we have tried to expose some of those existing urban flood risks related to:

i) proximity to overland flow paths beyond valley systems (sometimes encumbered major drainage systems), and

ii) increased imperviousness cover in Ontario municipalities (e.g., mid 1960's to late 1990's).

BONUS - Yoda, Twerking, Extreme Weather Correlation

Maybe I didn't learn anything from Pielke's book, like the importance of data and statistical analysis to advance science and public policy. I did learn that the New York Times has a great online tool to analyze the frequency of extreme weather references in their publication. I have used that to help establish some undeniable correlations between extreme weather, twerking and Yoda, as shown in the graphic below.

New York Times Chronicle Tool - Correlation of Extreme Weather, Twerking and Yoda

Spurious correlation between temperature and disaster damages you have found?

*****

Before considering Hurricane Harvey, a category 4 hurricane, we can review 2017 trends from Pielke Jr. - there had been a large gap before Hurricane Harvey, globally hurricanes are less frequent, and in the US the losses have been decreasing as a proportion of GDP:

And some more recent hurricane frequency statistics from Pielke Jr.'s blog:

Decreasing frequency of hurricanes making landfall suggests no climate change effects despite 2017 events.

***

Dr. Pielke Jr. and atmospheric scientist Ryan Maue have recent reviewed hurricane (tropical cyclone) trends over the past 50 years - see article in Forbes. While the variability in hurricanes making land fall is quite large, making it hard to distinguish a clear trend the counts of events using the Saffir-Simpson (S/S) hurricane scale use by the U.S. National Oceanic and Atmospheric Administration is shown below:

A recent summary notes the consensus on a lack of anthropogenic influence trends, but predicted future increases (Pielke, 2019 - Forbes):

"NOAA [the National Oceanic and Atmospheric Administration] concludes ‘an anthropogenic
influence has not been formally detected for hurricane precipitation,’ but finds it likely that
increases will occur this century. Similarly, the WMO concluded, ‘no observational studies have
provided convincing evidence of a detectable anthropogenic influence specifically on hurricane-related precipitation,’ but also that an increase should be expected this century. The U.S. National
Climate Assessment concurred, explaining that there is agreement on predictions for a future
increase in hurricane-related rainfall, but ‘a limiting factor for confidence in the results is the lack
of a supporting detectable anthropogenic contribution in observed tropical cyclone data."

Despite this, damages are increasing. Klotzback et al. (2018) noted:

“While neither U.S. landfalling hurricane frequency nor intensity shows a significant trend since
1900, growth in coastal population and wealth have led to increasing hurricane-related damage
along the U.S. coastline.”

The following chart shows losses normalized, considering changes in inflation and wealth at the
national level as well as changes in population and housing units at the coastal county level in the
US: