Why did so many people die in the Lower Ninth Ward?

Posted on August 29, 2015 By

My last blog addressed a question that I’ve gotten the most over the last decade: how many people died because of Hurricane Katrina impacts in Louisiana? This blog post deals with the next most common question that I get: Why did so many people die in the Lower Ninth Ward?

Typically, the person asking, whether a reporter, researcher, policy analyst, or anyone else, follows the question with the suggestion that it has to do with evacuation rates being lower there because of the social-economic characteristics of the neighborhood’s population. My response is that there is no evidence to support this explanation and that the true explanation lies with hydrodynamics, specifically depth and flow velocity, and not social-economics. In what follows, I lay out the rationale for this explanation. Details are available in my dissertation, which has been published as a scientific monograph.  This specific avenue of my research has also been presented at the 2011 Basic of the Basin Workshop.

The figure below is the result of a point pattern analysis of the victim recovery locations. It shows the density of the points interpolated to a grid. The figure shows that the density of victims is indeed strongest in the Lower Ninth Ward, adjacent to the two catastrophic failures of the levees along the Inner Harbor Navigational Canal, also known as the Industrial Canal. However, this map is just based on the raw count of victims and does not take into account the distribution of the underlying population at risk.

Kernel Density of Katrina VictimsMap of the Kernal Density Point Pattern Analysis of Hurricane Katrina Victim Recovery Locations.


Following basic epidemiological practice, disaster health impacts are measured and modeled as a rate (proportion) relative to an underlying “at risk” population. The rationale is fairly simple: you need a measure of risk that is independent of the size of the population. However, measuring that population denominator gets tricky when when the population is highly dynamic, such as the case when an estimated 80% evacuated to outside of the region before a hurricane while many of remaining people relocated to official and unofficial shelters and refuges within the region.

To obtain the underlying population denominator to calculate the flood fatality rate, I started with the spatial distribution of the population based on the 2000 Census, then adjusted the number for the change in population by 2005 along with the percent of the population that evacuated and the percent of the population that registered in the official shelters. Using GIS software, the calculations where completed at the Census blockgroup level, and the resulting layer I termed the “at risk” population. Next, I identified just the blockgroups that experienced flooding, and called this layer the “flood exposed population.” Finally, I counted the total number of flood fatalities for each blockgroup, and then divided this by the flood exposed population to get what I called the “flood fatality rate” layer.

The flood fatality rate is shown in the figure below. It can be seen that after accounting for the underlying population at risk, that the Lower Ninth Ward still shows up as having the highest fatality rate. Indeed, the fatality rate here was four times the average for New Orleans. While the average flood fatality rate for New Orleans was 0.9% (588 flood deaths/61,563 flood exposed person), the flood fatality rate for the Lower Ninth Ward was 4% (80 flood deaths/1,983 flood exposed persons.)


The Flood Fatality Rate for New Orleans and St. Bernard Parish.


The above calculation assumes an uniform evacuation and sheltering rate for all of New Orleans. This assumption is necessary given that there is very little reliable data on inner-parish variability in the evacuation rate. What data is available however is consistent with my estimate of the flood exposed population. On a regional level, my estimate of the flood exposed population came in at 66,000, which agrees well with the estimated 65,000 people rescued. The second figure is based on a thorough review the after-action reports and similar documentation. For the Lower Ninth Ward, my calculations indicated 1,983 persons remained in the Lower Ninth Ward, which is very close to the 2,000 rescues there reported by the Louisiana Wildlife and Fisheries Department. So, the data simply does not support the notion that the evacuation was less effective for the Lower Ninth Ward.


GIS based estimated of the flood exposed population for Orleans and St. Bernard Parishes.


Number of people rescues in the aftermath of Hurricane Katrina, based on review of after-action and similar reports.

After calculating the flood fatality rate layer, I then sought to explain the observed patterns through statistical modelling that included the flood hazard characteristics and the population vulnerability attributes. In terms of flood hazard characteristics, I was able to use a highly precise and accurate flood depth grid (based off of LIDAR, satellite imagery, and high water marks) along with the results of the SOBEK hydrodynamic model that provided depth, velocity, depth times velocity, rate-of-rise, and arrival time. For the population vulnerability attributes, I used 2000 Census datasets.

In my statistical analysis, I found that the hydrodynamic variable mattered much more than the social-economic variables. Water depth, flow velocity matter most, and specifically the product of depth and velocity was the most significant variable explaining the patterns of the flood fatality rate. When I examined the population attributes, I found that the percent of the population that was African-American was statistically significant, but the value of coefficient indicated that it did not have a substantive impact. Increasing the percent African-American from 45% to 95% while holding all the variable constant increases the flood fatality rate by a mere 0.2%. Other social-economic variable, including car ownership, were either not statistically significant or had no meaningful impact. These results indicate that none of the factors that we might think are associated with lower evacuation and sheltering rates in the Lower Ninth Ward have any bearing in explaining the high number of flood deaths there.



Map of depth times velocity, based on the SOBEK model.

So why did the flood waters have such lethally high flow velocities there? The answer has nothing to with the characteristics of the people who lived in the Lower Ninth Ward. The answer has to do with what I call the Lower Ninth Ward’s unfortunate location at the “last exit of the MRGO storm surge highway.” The MRGO (short for the Mississippi River Gulf Outline) is a man-made navigational canal that was dug through the marsh in Orleans and St. Bernard parishes east of the Industrial Canal, another man-made navigation canal the connects the Mississippi River with Lake Pontchartrain and separates the Lower Ninth Ward from central New Orleans. During Hurricane Katrina, this configuration of man-made canals served as a storm surge conveyance channel that both conveyed the surge to the heart of New Orleans and amplified its flow velocity. This is the storm surge that bulldozed the concrete and steel floodwalls protecting the Lower Ninth Ward and then bulldozed all but one house over a 20 block area.

Most floodwaters that flow through a levee breach are pulled through the breach by gravity, and they show modest flow velocities as a result. In contrast, the flood waters that flowed into the Lower Ninth Ward were not just pulled by gravity, these waters were also pushed through the breach because of the MRGO funnel effect. The figures below show results obtained from the ADCIRC storm surge simulation. The top figure shows that the surge flow velocity was 8 ft/s in the canal system, nearly three times the value in the open water of Lake Borgne. The second figure shows velocity vectors. As you can see, the arrows point straight toward the Lower Ninth Ward. While these results are based on the assumption that the levee did not breach, they show that once the levees did breach fast moving surge waters were pushed into the neighborhood because of the storm surge funnel effect of the MRGO.


Storm Surge Velocity Values from the ADCIRC Model (assumes no levee breaches.)



Storm Surge Velocity Vectors from the ADCIRC Model (assumes no levee breaches.)

To collaborate this line of reasoning, I will move away from fatalities and talk about building damage, specifically the residential building damage assessment dataset for New Orleans. This assessment was based on a visual inspection of buildings to determine the percent of the building that was damaged. I was able to geocode 65,000 of the residences (about 60% of the total) to create a point dataset for use in GIS mapping and analysis. The first map shows each building as a color coded point based on percent damage. The concentration of red dots in the Lower Ninth Ward is obvious. The second map shows an interpolation of this dataset. An interpolation is similar to the point density analysis described above, except that the interpolated values represent both the density of points and the magnitude of the damage for each point. It shows that the apparent concentration of damage in the Lower Ninth Ward is indeed a statistically significant effect.

Now, houses do not evacuate, so the variations in the residential building damage cannot be explained in terms of variability in the evacuation effectiveness. Indeed, in looking at this data in relation to the flood hazard characteristics, my colleagues Aimilia Pistrika and Bas Jonkman found that it was the lethal and destructive combination of depth and velocity that explains the cluster of structural damage in the Lower Ninth Ward, just like my research found that these variable explain the fatality rate there that is four times the average for New Orleans.

Residential Damage Points

Map of Residential Structure Damage for New Orleans.

Residential Damage Raster

Map of Interpolated Residential Structure Damage for New Orleans.

To conclude, the patterns of the recovery locations for the deceased victims of Hurricane Katrina show that the Lower Ninth Ward experienced the highest concentration of deaths due to the disaster. There is no evidence that this pattern relates to the notion that the evacuation was less effective there. Indeed, there is no evidence to even support the notion that the evacuation of the Lower Ninth Ward was less effective than in the other parts of New Orleans. When accounting for evacuation and sheltering, the flood fatality rate for the Lower Ninth Ward is four times the average rate for New Orleans. When comparing the flood fatality rate to the flood hazard characteristics and the population vulnerability attributes, the depth and depth times velocity were the most significant variables. These two variables also explain much of the observed patterns in building damage.

Corps of EngineersDisaster GeographyDisaster MappingDisaster ResponseFloodGISHotshotshurricaneHurricane KatrinaMan-made DisasterMapsNew Orleansstorm surge     , , , , , , , ,

Leave a Reply