Skip to content

Texas Well Water Quality Atlas

How the Texas Well Water Quality Atlas was made

Every laboratory analysis of Texas well water for arsenic, nitrate, fluoride and dissolved solids already public in two sources was gathered. Before a single result was compared with a limit, the rules were written and frozen: which tests count, what a well is and which value represents it. With those rules the atlas counts how many wells exceed each limit, and a study with its rules fixed in advance tests what best estimates one well's water.

Data: Texas Water Development Board Groundwater Database, file of November 13, 2026; Water Quality Portal (USGS and EPA), download of November 15, 2026. Page published November 16, 2026.

1. Sources

SourceWhat it addsHow it is used
TWDB Groundwater Databasefile of Nov 13, 2026Most of the analyses: 56,061 sites loadedCore of every figure
Water Quality PortalUSGS and EPA, Nov 15, 2026USGS tests the state does not have: 6,998 well sites, about 6% more wells since 1992Core, after removing duplicates
NUREDepartment of Energy, 1976–1980Arsenic in 13,554 wellsLoaded as a historical contrast; never enters the figures
USGS national arsenic modelLombard and others, 2021Probability that a private well exceeds 10 µg/LExternal reference of the study
Texas Well AtlasTerraFunded, TWDB driller reportsTypical depth of home wells by countyThe "before drilling" variant of the study

In all, 76,613 sites and 416,882 laboratory results for the four substances sit in the site's database, and what was loaded was checked cell by cell against what was read: zero differences. Discarded after review: the USGS file of domestic wells (almost all already in the state database), the national arsenic compilation (locations rounded to 2 km) and EPA compliance data, which is treated water from public systems, not well water. To gather 10 tested wells around a home well takes 7.4 miles on average, hence radii of 5, 10 and 15 miles.

2. The four indicators

IndicatorThresholdWhat the threshold is
Arsenicdissolved arsenic above 10 µg/LFederal drinking-water standard
Nitratenitrate above 10 mg/L as nitrogenFederal drinking-water standard
Fluoridedissolved fluoride above 4 mg/LFederal drinking-water standard (2 mg/L is the secondary level)
Total dissolved solidstotal dissolved solids above 1,000 mg/LTexas secondary level
  • "Exceeded" means above the threshold, not equal to it.
  • These standards are for public water systems. Here they are only the yardstick: no law requires a private well to meet them. Secondary levels deal with taste, staining and similar effects rather than health.
  • Why these four: arsenic, fluoride and salts come mostly from the rock; nitrate mostly from the surface (fertilizer, manure, septic systems). Having both kinds lets the study test whether the aquifer or the depth matters more.

3. Counting rules

Which tests count

  • State database: each result carries a note on how the sample was taken. Those taken with a field protocol, by USGS programs, by TCEQ with quality control or with laboratory analysis count. Those of unknown reliability enter only a sensitivity test. Those marked as not representative of the aquifer never enter.
  • Federal portal: only approved results of routine samples. Replicates and blanks do not count.

What a well is

The same well can appear in both sources under different names. Nearness proposes pairs; it does not confirm them. Two records are the same well if they share the USGS number, or are within 500 m and share a same-day sample with the same value, or are within 100 m with the same depth. Doubtful cases stay out of the main series and enter a sensitivity test.

Which value represents a well

  • The most recent accepted test within the window of years: the rule of official reports.
  • Main window: since 1992 (the Bureau of Economic Geology also starts in 1992, because of changes in detection limits). Since 2011 and all years are also published.
  • A "less than" result with a reporting limit at or below the threshold counts as not exceeding. With a higher limit the well is indeterminate.
  • Indeterminate wells are not replaced by an older test: they leave the figure and are reported apart.

Three traps in the data, and how they were solved

  • Nitrate is stored three ways: as nitrogen, as nitrate plus nitrite as nitrogen, and as the nitrate ion, which is 4.43 times larger. The first that exists is used, in that order, and the ion is divided by 4.4268. A warning for anyone using the state file: field 71851 is named "Nitrate nitrogen, dissolved, calculated (mg/L as NO3)"; the name says nitrogen and the unit says nitrate. Read as nitrogen, it multiplies any count against 10 mg/L.
  • An analysis that contradicts itself does not decide: if two of its nitrate values fall on different sides of 10 mg/L, truly differ and cannot differ for chemical reasons, it does not decide. This happens in 17 analyses since 1992 and leaves 16 wells indeterminate.
  • State dissolved solids are not always the sum they claim to be. A value counts only within 5% of the recalculated sum of its constituents. The margin barely matters: with 2%, 5% or 10%, the statewide figure stays between 22.8% and 23.0%.

What each figure is

Among the tested and accepted wells, the share whose most recent test in the window was above the threshold. It is not the share of all Texas wells, and it is not weighted to look like it: there is no frame that would justify those weights.

  • Interval: 95% Wilson. It assumes independent wells; they are not, so the real interval is wider.
  • Under 10 wells there is no percentage: only the count.
  • No "enough data" labels: with 0 of 30 wells, the data are compatible with up to 11%.

4. How the atlas was checked

  • Three programs, the same figures: Python, SQL inside the database and a second Python program that shares no code.
  • Published reports recomputed with each author's rules: the statewide figures of TCEQ 2024 and of the Bureau of Economic Geology 2011, 2018 and 2025 are reproduced within 0.3 percentage points or less. With the rule of the Bureau of Economic Geology 2022 nitrate report, 896 more wells result than published (34,063 against 33,167); the reason was not found.
  • Eleven wells against the raw state file, one by one.
  • 6 rounds of outside review by AI models (ChatGPT and Grok, and once two unnamed reviewers). Each point was checked against the data: what proved true was corrected; what did not, was answered.

5. Sensitivity: if a rule changes, does the figure change?

A closed list of changes, one at a time, fixed in the protocol. They are split into three classes because they do not mean the same thing. For each state, aquifer and county, the lowest and highest figure under the first two classes is published.

ClassExamplesWhat it measures
Which wells and tests countBroad acceptance; doubtful pairs as separate wells; without the federal portal; without estimated values; only nitrate measured as nitrogenAnother population
How it is measuredThe higher of the two arsenic fractions; residue on evaporation instead of the sum of constituentsThe same wells, measured another way
Another question"Ever above since 1992"; the median of each well's testsNot uncertainty: a different figure

6. Descriptive comparisons

  • Home wells against public-supply wells, three ways: all wells; only where both kinds exist; and public wells spread over aquifer, depth and period the way the domestic ones are. Intervals from resampling 50 km blocks.
  • How much a test ages: wells with two tests five years or more apart, and how many crossed the threshold, by how close the first one was. It is described, not explained.

Private vs. public wells · How often to retest

7. The study

If a well's water is already estimated from the tested wells nearby, does knowing its aquifer or depth lower the error? Two contrasts fixed before the run decide; everything else is exploratory. Measure: Brier score. Interval: 97.5% from 2,000 resamples of 50 km blocks. The full design and results.

8. Limits that are always stated

  • Not a random sample. Since 1992 about 30% of tested wells are public supply and about 30% domestic.
  • Wells are counted, not people or volume of water.
  • The data are old in much of the state: the median year of the latest test is 2004 or 2005.
  • Arsenic is dissolved arsenic; the standard refers to water as it is drunk.
  • Depth is the total depth of the well, not the interval the water enters from.
  • No water chemistry or land use: the study does not explain where each substance comes from.
  • It does not predict a well that has not been drilled.
  • Amendments 2, 3 and 4 were written after the atlas figures had been seen, and they say so. Earlier versions are kept.
  • No outside human review yet: reviewed by two AI models in six rounds and checked with three independent programs.

9. Version history

Every change after the protocol was frozen is dated and says whether results had already been seen. Earlier versions are kept as they came out.

VersionDateWhat changed and why
Plan, versions 1 to 315 nov 2026Two products: an atlas that describes what was measured and a study that tests a hypothesis. Reviewed by ChatGPT and Grok; the main changes adopted: write and freeze the rules before computing anything, reproduce each published report with its author's rules, publish the study whatever it shows.
Protocol frozen15 nov 2026Rules written and saved with a date and fingerprint before any result was compared with a limit. Amendment 1, before seeing results: implementation details.
Atlas, version 115 nov 2026First figures by state, aquifer and county; reproduction of published reports; sensitivity; how much a test ages.
Version 2 (amendment 2, after seeing results)15 nov 2026State dissolved solids count only when they truly are a sum of their constituents; federal results with a contamination note are removed; the nitrate sensitivity test was redone because the first was not independent.
Version 3 (amendment 3)15 nov 2026A review found state records whose two nitrate values disagreed; a rule was added so that a self-contradictory analysis does not decide. Total dissolved solids are checked against their recalculated sum. Sensitivity split into three classes. The study was programmed and run on invented results.
Version 4 (amendment 4)15 nov 2026The rule of version 3 was narrowed: nitrate plus nitrite above 10 with nitrate below is not a contradiction, because the difference can be nitrite. The study intervals were shown to hold their level at the boundary. A sentence about time that said more than the data support was removed.
Study16 nov 2026Rules and code saved with a timestamp before any real result was read; run once; checked by two independent programs written by ChatGPT and by Grok from the rules alone.
Publication16 nov 2026These pages and the open data, archived on Zenodo (DOI 10.5281/zenodo.24287393).

Sources you can check

The full method · All open data

How to cite this

Blanco, R. (2026). "How the Texas Well Water Quality Atlas was made." Texas Well Water Quality Atlas. TerraFunded. https://terrafunded.com/water-quality/methodology. Data: https://doi.org/10.5281/zenodo.24287393

Free to use with credit under CC BY 4.0. Please also credit the Texas Water Development Board. Dataset DOI: 10.5281/zenodo.24287393.

Rodrigo Blanco

Rodrigo Blanco, Founder and CEO of TerraFunded, a Texas company that sells rural land. These pages carry no listings, prices or offers.

Reviewed by two AI models in six rounds and checked with three independent programs; not yet reviewed by outside specialists.

Press and researchers: [email protected]