STL Tornado
Aid Inequality

The Feedback Loop

How structural inequities from redlining compound across disaster cycles

What Actually Happened With FEMA

Six months after the tornado, federal recovery funding stalled in North St. Louis. As Federal policy changes restrict FEMA's funding and shift response toward a state obligation, the mere concept of response has become fractured between groups in St. Louis. Does it look like debris removal, or housing assistance, and renovation programs? How might this perception of response differ throughout the city? How might it differ between City hall and North St. Louis residents?

KSDK — St. Louis

FEMA Rejects Bulk of St. Louis Tornado Demolition Costs

February 3, 2026

Federal policy prohibits reimbursement for structures condemned prior to the disaster. FEMA considers multiple structures owned by the same entity to be “commercial properties,” regardless of the number of units, making them ineligible for debris removal programs.

“Visually, you won’t see progress in neighborhoods like mine,” McLemore said. “A lot of these vacant buildings will probably stand for years as a result.”

Action St. Louis is urging the city to ensure that demolition costs do not drain the $13 million currently earmarked for critical home repairs. For now, neighbors are left waiting, hoping the promised resources will finally reach their doorsteps.

Read full article →

NPR

Tornado Survivors Say Recovery Is a Mess Due to FEMA Changes

November 16, 2025

Six months ago today, a tornado ripped through the city. It left more than a billion dollars of destruction, including more than 5,000 damaged buildings. Now it’s become a key experiment for President Trump’s new policy on federal emergency response — shifting the burden of recovery from FEMA to the states.

Residents say they’re the guinea pigs of this experiment and many feel, at best, in limbo, at worst, abandoned.

Read full article →

Connected thread — land banking & the pre-condemned clause

The FEMA rule barring reimbursement for pre-condemned buildings hits North St. Louis disproportionately — a city where decades of land banking left hundreds of properties legally condemned before any storm arrived. Mayor Cara Spencer’s office argued these restrictions “unfairly impact property owners who are low-income and uninsured, leaving them without the personal financial means to address the hazardous debris on their own.”

Of all the spot abatement done in the St. Louis metro area over the past decades, the most notorious has involved the developer Paul McKee, who has used tax increment financing to purchase hundreds of buildings on the city’s Northside. A 2018 investigation found that of the 154 buildings McKee had specifically promised to rehabilitate, only two had actually been rebuilt… Northside Regeneration represents a huge land bank, accredited by the city of St. Louis and capitalized by the continued suffering of the poor Black population of the beleaguered Northside.

— Walter Johnson, Broken Heart of St. Louis

Where This Analysis Could Go Further

The FEMA Aid Proxy is a starting point. Here are dimensions we haven’t fully accounted for — variables that would sharpen the picture of structural recovery inequity.

• ? Factor in vacant properties? Tracts with high vacancy rates face compounded FEMA ineligibility — and our current proxy doesn’t capture how many units were already condemned before the storm.
• ? Factor in commercial land banking practices such as Paul McKee’s? Large single-owner portfolios trigger FEMA’s “commercial property” clause, shutting out entire blocks from debris removal aid regardless of damage severity.
• ? Renter vs. owner at the building level, not just the tract level? Our proxy uses owner-occupancy rate, but renter-occupied units in owner-occupied buildings can mask real eligibility gaps for individual households.
• ? What else could we add in?

Map It Yourself

Drag terms onto the canvas and draw connections. Build your own causal diagram of the structural forces at play in disaster inequality — segregation, FEMA rules, land banking, feedback loops, and more.

How to use: Drag any term from the left panel onto the beige canvas. Select Line or Arrow, then click and drag to draw connections. Use the color buttons to change stroke color. Reset clears all drawings and dropped terms.

Further Reading & Historical Context

On structural inequality, property abandonment, and the long aftermath of redlining in North St. Louis.

“

Of the city’s abandoned houses, it is perhaps fair to say that they are worth more dead than alive… Some [“brick rustlers”] will set an old house on fire, knowing that the water from the firemen’s hoses will soften the mortar, making the bricks easier to salvage and scrape clean for sale. So many of the houses in North St. Louis have been torn down that some of the neighborhoods look like rural farmsteads—clusters of houses here and there surrounded by open space.

— Broken Heart of St. Louis: Urban Decay, Crime and the Search for Meaning in America
Walter Johnson — Prologue: Mapping the Loss, p. 3

View book →

Photo Essay — BuiltSTL, 2007

Brick Rustlers in Action

Abandoned buildings stripped for salvage on the 1900 block of Montgomery Ave — the end stage of disinvestment that redlining set in motion decades earlier.

Variable Reference

Every token used in the methodology equations above, with source and construction notes.

• MSFT_DAMAGE_API
  ArcGIS REST API endpoint returning building-level damage polygons with AI-predicted scores. Paginated in batches of 2,000 in 01_fetch_damage.py.

  Source: Microsoft AI for Good Lab — model predicted_damage_St_Louis_model_364_v6
• DAMAGE_PCT_0M
  AI-predicted structural damage probability per building at 0-meter distance tolerance. Value range: 0 (no damage detected) to 1 (fully damaged). One record per assessed building polygon.

  Source: Microsoft AI for Good Lab — field damage_pct_0m in the ArcGIS feature layer
• CENSUS_TRACT
  St. Louis City (FIPS 29-510) census tract polygons used as the spatial unit of analysis. 104 tracts total; 54 contained at least one assessed building.

  Source: pygris TIGER/Line 2022 — pygris.tracts(state="MO", county="510", year=2022)
• MEAN_DAMAGE_PCT
  Tract-level mean of DAMAGE_PCT_0M across all buildings spatially joined to that census tract via a within-predicate spatial join. Computed in 02_spatial_join.py.
• DAMAGE_RATE
  Fraction of total assessed buildings within a census tract that recorded any damage (damage_pct_0m > 0). Distinct from MEAN_DAMAGE_PCT: a tract can have high DAMAGE_RATE (many buildings touched) with low MEAN_DAMAGE_PCT (most lightly damaged), or vice versa. Computed as damaged_buildings / total_buildings in 02_spatial_join.py. Selectable as a choropleth layer on the map.

  Derived from Microsoft AI for Good Lab damage data via spatial join — field damage_rate in tract_damage_summary.csv
• PCT_BLACK
  Proportion of census tract population identifying as Black or African American alone. Derived as B02001_003E ÷ B02001_001E.

  Source: U.S. Census Bureau, ACS 5-year 2022 — table B02001
• PCT_OWNER
  Proportion of occupied housing units that are owner-occupied. Derived as B25003_002E ÷ B25003_001E. Renters are structurally ineligible for FEMA IHP structural aid.

  Source: U.S. Census Bureau, ACS 5-year 2022 — table B25003
• MEDIAN_HOME_VALUE
  Median value of owner-occupied housing units in dollars. ACS sentinel value −666,666,666 replaced with NaN before use. Directly proxies FEMA IHP repair/replacement cost basis.

  Source: U.S. Census Bureau, ACS 5-year 2022 — table B25077
• MEAN_PCT_BLACK
  Tract-group average of PCT_BLACK, computed separately for above-median and below-median damage groups to produce the 77.4% vs. 47.8% comparison.
• FEMA_AID_PROXY
  Constructed composite score (0–1): equal-weight average of min-max-normalized MEDIAN_HOME_VALUE and PCT_OWNER. Proxies structural eligibility and award magnitude under FEMA's Individuals and Households Program (IHP). Not an official FEMA metric. Computed in 04_merge_and_analyze.py.

  Validated against: OpenFEMA Housing Assistance Owners v2
• OPENFEMA_OWNERS
  OpenFEMA Housing Assistance Program Data (Owners, v2) — individual-level FEMA IHP disbursements for declared disasters. Referenced for proxy validation context.

  Source: FEMA OpenFEMA
• PEARSON_R
  Pearson correlation coefficient (r) measuring linear association between two variables. Computed using pandas .corr() across 54 tracts with at least one assessed building. Range: −1 (perfect negative) to +1 (perfect positive). Computed in 04_merge_and_analyze.py.

Data Sources & Methodology

Building damage assessments from the Microsoft AI for Good Lab via ArcGIS REST API (predicted_damage_St_Louis_model_364_v6). Damage score (0–1) reflects predicted structural damage from AI analysis of post-event satellite imagery. 78,777 buildings assessed across St. Louis City.

Census demographics from the U.S. Census Bureau American Community Survey (ACS) 5-year estimates, 2022, at the census tract level. Variables: race (B02001), median household income (B19013), tenure/homeownership (B25003), housing age (B25034), median home value (B25077). Tract boundaries from TIGER/Line 2022 via pygris.

FEMA Aid Proxy is a composite score (0–1) equal-weighting min-max normalized median home value and owner-occupancy rate. This proxies structural aid eligibility and award magnitude under FEMA's Individuals and Households Program (IHP). It is not a precise aid forecast. OpenFEMA Housing Assistance Owners v2 was consulted for validation context.

Spatial analysis conducted in Python using geopandas, pygris, and the census API package. Pearson correlations computed across 54 tracts with at least one assessed building.

ESE 3590 Final Project  ·  Washington University in St. Louis  ·  Spring 2026