Harvey article / data science workup

Harvey floodplain stress test: the data workup behind the map.

This page is the technical companion to the article. The article tells the story; this page shows the evidence pipeline, scoring method, caveats, and what still needs to be tested before the project can claim more.

535Houston-region Harvey high-water marks scored

170intersect Zone X / minimal context

92+ ft Zone X marks at least 0.25 mi from SFHA

42+ ft Zone X marks at least 1 mi from SFHA

Short version:

The tested Variant C result does not say “the maps failed.” It says most outside-SFHA marks were boundary-near, but a smaller material tail was not: among Zone X/minimal-context marks with at least 2 ft above ground, 9 were at least a quarter mile from current SFHA geometry and 4 were at least a mile.

Open the working map or read the narrative article.

Pipeline

Acquire

Pull USGS Harvey peak high-water marks from the CUAHSI HydroShare WFS service. Preserve the raw GeoJSON and a source manifest.

Normalize

Keep coordinates, county, waterbody, peak stage, height above ground, peak date, datum, estimated flags, and notes. Filter to Harris, Brazoria, Fort Bend, Liberty, Montgomery, Chambers, Galveston, and Waller counties.

Score

Query FEMA NFHL Flood Hazard Zones at each high-water mark point. Classify the result as current SFHA context or Zone X/minimal context.

Distance

Fetch current FEMA SFHA polygons for the Houston-region bounding box, project points and polygons to EPSG:3083, and calculate distance from each Zone X/minimal-context Harvey mark to the nearest SFHA polygon geometry.

Visualize

Aggregate the observed marks into coarse stress cells so the reader sees the pattern first, then can toggle individual marks as receipts.

What is defensible now

The workflow can reproduce the 535-mark Houston-region subset.
Current FEMA context can be queried and cached for each mark.
The distance calculation scored 170 Zone X/minimal-context marks against current SFHA geometry.
The material subset is 54 Zone X/minimal-context marks with at least 2 ft above ground.
Among that material subset, 9 marks are at least 0.25 miles from current SFHA geometry and 4 are at least 1 mile away.

What is not proven yet

This is not yet matched to the exact effective pre-Harvey FIRM panel for each point.
It does not yet incorporate terrain/elevation or official flood-depth rasters.
It should not be used as insurance, engineering, emergency, or property advice.

Variant C result

The surprise-distance test is now built:

Zone X mark + height above ground + distance to nearest SFHA boundary

The median Zone X/minimal-context distance is 0.032 miles, which means many misses are boundary-near. The engaging part is the tail: 9 material marks are at least 0.25 miles from current SFHA geometry, and 4 are at least 1 mile away.

Site	County	Waterbody	Above ground	Distance to current SFHA	FEMA context
TXGAL20912	Galveston	cloud bayou	2.6 ft	1.469 mi	Zone X, minimal hazard
TXCHA21578	Chambers	Oyster Bayou	2.9 ft	1.278 mi	Zone X, 0.2 pct annual chance
TXGAL20911	Galveston	cloud bayou	3.58 ft	1.214 mi	Zone X, 0.2 pct annual chance
TXLIB21725	Liberty	Ditch	5.8 ft	1.085 mi	Zone X, minimal hazard
TXHAR22784	Harris	Buffalo Bayou	2.9 ft	0.741 mi	Zone X, minimal hazard

Full generated table: harvey-surprise-top10.json.

Version E context comparison

The current FEMA point-intersection split is not balanced: 365 marks fall in current SFHA context and 170 fall in Zone X/minimal context. The outside group is shallower on median, but still contains 54 marks at or above 2 ft above ground.

Current FEMA context	Marks	At least 2 ft above ground	Median above ground	Max above ground	Zero-height marks
Inside SFHA	365	181	1.97 ft	11.85 ft	132
Zone X/minimal context	170	54	1.07 ft	7.5 ft	67

Version E sensitivity check

The 2 ft material threshold is a screening choice, so the distance tail was checked against lower and higher height cutoffs. The tail persists at 1 ft and tightens as the height threshold rises.

Zone X/minimal height cutoff	Marks	At least 0.05 mi from SFHA	At least 0.25 mi from SFHA	At least 1 mi from SFHA
0+ ft	170	76	37	8
1+ ft	88	39	19	4
2+ ft	54	23	9	4
3+ ft	34	12	4	2
4+ ft	16	7	1	1
5+ ft	9	5	1	1

Version E near-vs-far check

Among the 54 material Zone X/minimal-context marks, 45 are less than 0.25 miles from current SFHA geometry and 9 are at least 0.25 miles away. This keeps the main claim narrow: most outside marks are boundary-near, while the far tail is small enough to inspect site by site.

Distance bucket	Material marks	Median above ground	Max above ground	Median distance	Max distance
Less than 0.05 mi	31	3.54 ft	6.2 ft	0.008 mi	0.046 mi
0.05 to 0.25 mi	14	3.865 ft	7.5 ft	0.102 mi	0.226 mi
0.25 to 1 mi	5	2.9 ft	3.7 ft	0.307 mi	0.741 mi
1 mi or more	4	3.24 ft	5.8 ft	1.246 mi	1.469 mi

Version E data handling

Zero-height records are retained as source observations with height_above_ground_ft = 0. They remain in the 535-mark scored set and in context counts, but they are excluded from material-threshold tables such as the 2+ ft outside-SFHA test.

Estimated flags are preserved as audit fields and are not used as filters in this version. Counts below are for the 535 scored Houston-region marks.

Field	Definition used here	True	False
`is_peak_estimated`	Peak value includes an estimated component in the normalized source record.	530	5
`is_peak_stage_estimated`	Peak stage was flagged as estimated in the normalized source record.	393	142
`is_peak_time_estimated`	Peak timing was flagged as estimated in the normalized source record.	532	3

Measurement plan

For the public site, track the funnel without collecting more personal data than needed:

LinkedIn post variant using UTM tags: utm_content=harvey_a or utm_content=harvey_b.
Article page view and scroll-depth completion.
Click-through from article to this data science workup.
Time on this page and outbound clicks to source manifest, map, notebook, or repo.
Aggregate counts by page and referrer only; keep raw IP/user-agent logs siloed from the public proof packet.