EVA Archives - Reduce Flooding

3/31/26 – Since Hurricane Harvey, one of the dominant themes of more than 3,000 posts on this website has been outdated rainfall assumptions and how flood-infrastructure design lags those assumptions. That design lag contributes to flooding nationwide.

How Infrastructure Design Lags Rainfall Estimates

Almost nine years after Harvey, Harris County finally has new “draft” flood maps issued by FEMA based on Atlas-14 rainfall assumptions. But they won’t take effect for another two to three years. Meanwhile, parts of the region have built and will continue to build infrastructure, homes and businesses based on long-outdated rainfall predictions.

Before Atlas-14-based flood maps become effective in Harris County, FEMA will issue Atlas 15 predictions throughout the U.S. in 2026.

Different regulatory agencies and jurisdictions update at different speeds.

Imagine members of a choir each singing from different songbooks.

These rainfall predictions affect many aspects of flood control and building codes, including:

Stormwater detention basin sizing
Floodplain mapping
Subdivision drainage criteria, i.e., size of storm sewers
Elevation of homes and businesses above the 100-year floodplain and street level
Acceptable foundation types in flood- hazard areas
Building in floodways including width, depth, bracing and other requirements for piers
Where fill can and cannot be used

In 2009, Harris County revised its floodplain development standards. After Harvey, Harris County Engineering compared damage found in subdivisions built before and after the new standards.

Subdivisions built with the updated standards experienced one twentieth the amount of damage.

EVA and the Difficulty of Developing Rainfall Estimates

So getting the rainfall right is important. But why do we get it wrong so often? Rainfall predictions stem from an obscure branch of mathematics, known as EVA (Extreme Value Analysis). With EVA, forecasters try to predict the probability of unobserved future events based on the frequency of somewhat smaller past events.

But there’s a problem. All predictions (500-year storms, etc.) are based on extremely small data sets. EVA may produce the best numbers possible, but predicting 500-years into the future based on 100 years of data takes a lot of guess-work. And we design entire cities based on these probabilities!

And that’s what I mean by the “design lag” problem based on outdated rainfall assumptions. Assumptions change with major new storms, such as Harvey. But we can’t just wave a major wand and change trillions of dollars of infrastructure already in the ground. We are systematically under-designed for current rainfall statistics.

This isn’t just a Houston problem. And it’s not just “in the ground” infrastructure. For example…

Mississippi Levees

Engineers typically design levees to contain a 100-year (1% annual chance) storm. Even small increases in in rainfall when levees constrain a river can produce huge increases in water surface elevation and push crests above levees. We’ve seen this happen recently in 2008 and 2019.

Design lag manifested itself as systemic underestimation of river stage, leading to overtopping of levee defenses.

Street Flooding in New York City

In 2023, extreme rainfall (up to 2.5 inches per hour) snarled the New York metropolitan area. It knocked out subways, commuter rail lines, flooded basements and closed a terminal at LaGuardia. It was the City’s wettest day since 1960 and came just two years after Hurricane Ida killed at least 13 people.

Engineers from previous generations designed storm sewers for much lower peak intensities. Usually, they design urban systems around peak intensities for short time periods (minutes to a few hours). But pre-Atlas 14 standards underestimated those short duration bursts.

Storm sewers were simply designed for much lower peak intensities. Systems failed almost instantly. Subway entrances became inlets for street flooding.

The Lesson Learned

Flooding results not just from rainfall. It results from under-designed infrastructure for rain that falls. For example:

Homes not high enough.
Channels not wide enough.
Storm sewers not big enough.

As a result, NOAA hopes to update rainfall frequency estimates much more frequently in the future. NOAA developed Atlas 14 region-by-region from approximately 2000–2018.

Texas lagged other regions by years. Before Atlas 14, the prior standard (TP-40) dated to 1961. And during that time, Houston grew exponentially. So most of our infrastructure is built to older standards.

NOAA’s new Atlas 15 is designed to be the nation’s first unified data set. And it is built from the ground up to incorporate climate trends and be updatable as new data becomes available.

NOAA is trying to eliminate the “outdated by the time it’s adopted” issue. But:

Local adoption will still likely lag availability
Regulatory inertia will remain the bottleneck.

See Lessons Page For More Information

For more information about other causes of flooding, see the Lessons page of this website.

Posted by Bob Rehak on 3/31/26

3136 Days since Harvey

In the School of Hard Knocks, there’s an introductory course called, “Money Has a Short Memory.” Most students fail this free course and, as a consequence, are still paying “tuition” years later. The irony was never more visible than last week. As I reviewed a Houston Public Media Story about how the City of Houston was not attempting to curb development in the 100-year flood plain – despite everything we learned from Harvey – I had a presentation about the 1994 flood waiting for review on my desktop.

1994 Flood Should Have Taught Us Lessons We Still Haven’t Learned

The presentation, “Rain by the Cubit: The Great Southeast Texas Flood of 1994,” brought back memories. That was the year I started my company. I was supposed to move into my first commercial office space when this flood hit.

Kingwood received 29″ of rain that week. Rainfall averaged 19.5 inches over the entire 2,880-square mile San Jacinto River watershed. The event lasted four days. It started on Saturday, October 15, 1994 when Pacific Hurricane Rosa met a gulf coast warm front over Texas. It affected 38 Texas Counties, an area as large as Maine.

1.9 million acre-feet of runoff passed through Lake Houston: almost 12 times the volume of the entire lake! The lake crested 8.3 feet above the 3,160-foot spillway.

*Homes under construction on Atascocita Point. HCFCD Photo from presentation by Yung and Barrett on 1994 flood.*

Stunning Photos of 1994 Flood

The presentation contains photos of flooding:

On Atascocita point, where new construction was just beginning at the time.
In Forest Cove townhomes that would flood at least four more times before buyouts
In Banana Bend below Lake Houston, which is also just now being bought out
Around Toys ‘R Us on 59 – before an entire strip center of big box stores surrounded it
That collapsed the 59 bridge
That downed power lines over Lake Houston
That went up to the roofline of what was then Reeves Furniture on the southbound 59 feeder just north of the West Fork
That ruptured pipelines across the San Jacinto and started a toxic blaze
That buried downstream areas in sand and gravel.

Sound familiar? It should. Virtually all those things happened during Harvey, with the exception of the pipeline fire. However, toxic waste pits were involved during Harvey.

What are the Chances?

At the time, experts opined about how rainfall exceeded the expected 100-year levels. But the new Atlas-14 data released by NOAA, now advises that a four-day flood averaging 19.5 inches would have an average recurrence interval of 50 years.

*The latest NOAA Atlas-14 Rainfall Data for the Lake Houston area*

After Harvey, people dazed by the devastation, solemnly concluded that the storm must have been a 500-year, a 1,000-year, or even a greater storm. They had absolute faith in the numbers that developers, engineers, bankers, insurers, and government agencies certified. They assumed storm intensity had to be greater than expected. It never occurred to them that perhaps the numbers could be off…in the other direction.

How Average Recurrence Interval is Determined

All these numbers (500-year, etc.) are based on extremely small data sets. Forecasters use a branch of mathematics called Extreme Value Analysis (EVA). With EVA, they try to forecast the probability of unobserved future events based on the frequency of somewhat smaller past events. EVA may produce the best numbers possible, but predicting 500-years into the future based on 100 years of data takes a lot of guess-work.

Limitations of Numbers

Complicating things, most people are oblivious to the nuances of probabilities. The naming convention (100-year storm) misleads them into thinking that if we had a 100-year storm last year, “we must be good for another 99 years.” Wrong. Theoretically, if you tossed a coin and it came up heads 99 times in a row, you have a 50:50 chance of getting heads on the hundredth toss, too.

How many people read…or understand…the fine print in tables like the one above? Did you read the footnotes? If not, please go back and read them now. It’s important for your own safety and the safety of your investment.

They’re trying to say, “We can’t predict extremes with accuracy.”

Conclusions of 1994 Flood Presentation

Yung and Barrett conclude with several warnings. They include.

Extreme rainfall events will continue to occur.
The adoption of criteria that exceed FEMA minimum requirements should be considered by communities to guard against severe events.

So until the City learns this lesson, what’s someone without a PhD in math supposed to do when buying a home? Forego the river or lake view and buy on the highest ground you can find. Buyer beware! There are huge markups on floodplain property. And money has a short memory.

Posted by Bob Rehak on 6/9/2019, based on a presentation by Andy Yung and Duange Barrett

649 Days since Hurricane Harvey

Tag Archive for: EVA

Outdated Rainfall Assumptions: A Systemic Design-Lag Problem