Texas Well Water Quality Atlas
What best estimates arsenic and nitrate in a Texas well? A study with its rules fixed in advance
To estimate a Texas well's arsenic, tests from wells within 15 miles had 10.6% less error than the county figure and 15.4% less than the USGS national arsenic model. Knowing the aquifer added nothing: with the nearby tests in hand, keeping only those of the same aquifer did not lower the error (8,715 test wells). For nitrate, adding well depth was inconclusive (9,145 test wells). The study's rules and code were fixed and timestamped before it was run, it was run once, and three independent programs gave the same result.
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.
The question
If a well's water is already estimated from the tested wells around it (up to 15 miles), does knowing its aquifer or its depth lower the error?
The protocol expected yes for arsenic, which comes from the rock, so the aquifer should matter; and that depth would matter more for nitrate, which comes mostly from the surface. The answer decides how the atlas is organized: by aquifer, or by nearness and county.
What came out
- Arsenic: knowing the aquifer did not improve the estimate made with nearby wells. An improvement of 2% or more is ruled out with these data. The protocol expected the opposite.
- Why: nearby wells are almost always from the same aquifer, so nearness already carries that information. In the 4,468 wells where filtering by aquifer does change which neighbors are used, the error did not fall either.
- Nitrate: the data cannot tell apart no improvement from an improvement of up to 2.5%.
- Consequence, by the rule fixed in advance: the atlas is organized by nearness and county, not by aquifer.
The two contrasts fixed before the run
Measure: Brier score, the squared error of a probability. Interval: 97.5%, from 2,000 resamples of 50 km blocks with a fixed seed. A relevant improvement needs an interval that excludes zero and an error at least 2% lower (a convention).
| Arsenic: add the aquifer | Nitrate: add the depth | |
|---|---|---|
| Test wells (above the limit) | 8,715 (537) | 9,145 (483) |
| Change in error vs. nearby wells only | error 0.68% higher | error 0.79% lower |
| 97.5% interval | from error 2.71% higher to error 1.32% lower | from error 0.80% higher to error 2.45% lower |
| Reading, fixed before the run | No relevant improvement | Inconclusive |
| Same result in the programs written by ChatGPT and by Grok | Yes | Yes |
before drillingerror 4.7% higher · 97.5% interval: error 7.6% higher to error 1.9% higher
- Contrast fixed before the run (97.5% interval)
- Exploratory
- Interval
- Right of the solid line: less error than nearby wells alone. Dashed line: 2% less, the bar for a relevant improvement.
before drillingerror 3.8% higher · 97.5% interval: error 6.6% higher to error 1.5% higher
- Contrast fixed before the run (97.5% interval)
- Exploratory
- Interval
- Right of the solid line: less error than nearby wells alone. Dashed line: 2% less, the bar for a relevant improvement.
Exploratory: what the headline rests on
These comparisons were not the ones fixed in advance, so they are labeled exploratory. 95% intervals unless noted.
| Comparison (leave one well out) | Arsenic | Nitrate |
|---|---|---|
| Nearby wells vs. the county figure | error 10.6% lowerfrom error 5.8% lower to error 15.5% lower | error 5.3% lowerfrom error 1.0% lower to error 10.3% lower |
| Nearby wells vs. the statewide figure | error 35.7% lowerfrom error 26.5% lower to error 43.4% lower | error 26.4% lowerfrom error 15.1% lower to error 36.4% lower |
| Nearby wells vs. the USGS national arsenic model, in 7,904 wells the model could not have seen through the federal portal | error 15.4% lowerfrom error 9.9% lower to error 20.6% lower | — |
| Before drilling: nearby wells at the county's typical home-well depth vs. nearby wells (97.5% interval) | error 4.7% higherfrom error 7.6% higher to error 1.9% higher | error 3.8% higherfrom error 6.6% higher to error 1.5% higher |
- Where there are no tests nearby, the USGS national model does help. Leaving out whole 100 km blocks, the error of the nearby-wells estimate for arsenic rises to 0.0533, against 0.0438 for the national model in the same wells. This observation was not part of the plan and has no interval.
- Wells without an assigned aquifer or without a depth were left out of the test, and they exceed twice as often: 12.3% for arsenic and 11.4% for nitrate.
The methods compared
Each method gives a probability of exceeding, (K + 0.5) / (N + 1), with K wells above the limit out of N. Wells within 500 m of the test well are never used, so one well under two names never predicts itself.
| Method | Which wells it uses |
|---|---|
| Statewide figure | All tested wells of Texas |
| County figure | The tested wells of the county; if none, the state |
| Nearby tested wells (up to 15 miles) | All wells in the first radius of 5, 10 or 15 miles that gathers 10; if none, the county |
| Nearby wells of the same aquifer | The same, only from the same aquifer; if not enough, nearby wells |
| Nearby wells of similar depth | The same, only with a depth between half and double; if not enough, nearby wells |
| Nearby wells of the same aquifer and similar depth | Both conditions; if not enough, same aquifer |
| Nearby wells at the county's typical home-well depth (before drilling) | Like similar depth, but using the typical depth of home wells in the county (what a buyer knows before drilling) |
| USGS national arsenic model | Probability that a private well exceeds 10 µg/L, from the national model (arsenic only) |
Three tests
- Leave one well out (the main one): the real use, a new well where there are tests nearby.
- Leave whole blocks of 25, 50 and 100 km out: what happens where there are no data nearby.
- Forward in time: learn from tests of 1992 to 2010, check against tests from 2011 on.
How it was protected
- A power simulation before the run: 84 scenarios per contrast, 1,000 runs each, with invented results on the real wells. Where the factor adds nothing, the procedure declared no improvement in 24,000 runs. A real improvement of 5% or more is detected about 8 times in 10 for arsenic and almost always for nitrate; one of 2 to 3%, about 4 times in 10.
- Calibration at the boundary: where two methods perform almost the same, the 97.5% interval held the true improvement 98.4% of the time for arsenic and 97.9% for nitrate.
- Rules and code were saved with a timestamp before any real result was read, on November 16, 2026, and the study was run once.
- Three independent programs: ours, and two written by ChatGPT and by Grok from the rules alone, without seeing our code. They gave the same probability for every test well and the same reading of both contrasts.
What it does not say
- It does not say the aquifer is irrelevant to arsenic. It says that once there are 10 tested wells within 15 miles, the aquifer's name does not lower the error.
- "Inconclusive" for nitrate is not "depth doesn't matter". It is a lack of evidence: the study tells small improvements apart poorly.
- It does not predict a well that has not been drilled, and it does not explain where each substance comes from.
- It holds where there are data: 97% of the test wells had 10 tested wells within 15 miles.
- The comparison with the USGS model is not between equals: the model estimates with no nearby test, for private wells; nearness uses the tests next door, of every use.
Test your own well
No map or table can say what is in the water of one well. These figures say how often the tested wells of an area went over a limit; the only way to know your well is to have its water tested by an accredited laboratory.
The drinking-water standards used here apply to public water systems. No law requires a private well to meet them: here they are only the yardstick.
Every comparison of the study (69 rows)
Only the two rows marked as fixed in advance decide anything; all the others are exploratory. Tap a heading to sort.
| Substance | Test | Method vs. reference | Test wells | Change in error | Interval | Reading |
|---|---|---|---|---|---|---|
| Arsenic | Leave one well out | Nearby wells of the same aquifer vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.7% higher | 97.5%: from error 2.7% higher to error 1.3% lower | No relevant improvement (fixed in advance) |
| Arsenic | Leave one well out | Statewide figure vs. Nearby tested wells (up to 15 miles) | 8,715 | error 55.6% higher | 95%: from error 76.6% higher to error 36.1% higher | Worse |
| Arsenic | Leave one well out | County figure vs. Nearby tested wells (up to 15 miles) | 8,715 | error 11.9% higher | 95%: from error 18.3% higher to error 6.2% higher | Worse |
| Arsenic | Leave one well out | Nearby tested wells (up to 15 miles) vs. County figure | 8,715 | error 10.6% lower | 95%: from error 5.8% lower to error 15.5% lower | Relevant improvement, 2% or more |
| Arsenic | Leave one well out | Nearby tested wells (up to 15 miles) vs. Statewide figure | 8,715 | error 35.7% lower | 95%: from error 26.5% lower to error 43.4% lower | Relevant improvement, 2% or more |
| Arsenic | Leave one well out | Nearby wells of similar depth vs. Nearby tested wells (up to 15 miles) | 8,715 | error 1.0% lower | 95%: from error 1.0% higher to error 3.0% lower | Inconclusive |
| Arsenic | Leave one well out | Nearby wells of the same aquifer and similar depth vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.7% higher | 95%: from error 3.1% higher to error 1.7% lower | No relevant improvement |
| Arsenic | Leave one well out | Nearby wells at the county's typical home-well depth (before drilling) vs. Nearby tested wells (up to 15 miles) | 8,715 | error 4.7% higher | 95%: from error 7.2% higher to error 2.3% higher | Worse |
| Arsenic | Leave one well out | Nearby wells at the county's typical home-well depth (before drilling) vs. Nearby tested wells (up to 15 miles) | 8,715 | error 4.7% higher | 97.5%: from error 7.6% higher to error 1.9% higher | Worse |
| Arsenic | Leave 25 km blocks out | Statewide figure vs. Nearby tested wells (up to 15 miles) | 8,715 | error 41.1% higher | 95%: from error 60.1% higher to error 23.3% higher | Worse |
| Arsenic | Leave 25 km blocks out | County figure vs. Nearby tested wells (up to 15 miles) | 8,715 | error 7.4% higher | 95%: from error 13.5% higher to error 1.6% higher | Worse |
| Arsenic | Leave 25 km blocks out | Nearby tested wells (up to 15 miles) vs. County figure | 8,715 | error 6.9% lower | 95%: from error 1.6% lower to error 11.9% lower | Relevant improvement |
| Arsenic | Leave 25 km blocks out | Nearby wells of the same aquifer vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.5% lower | 95%: from error 1.1% higher to error 2.2% lower | Inconclusive |
| Arsenic | Leave 25 km blocks out | Nearby wells of similar depth vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.3% lower | 95%: from error 1.0% higher to error 1.7% lower | No relevant improvement |
| Arsenic | Leave 25 km blocks out | Nearby wells of the same aquifer and similar depth vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.4% lower | 95%: from error 1.9% higher to error 2.7% lower | Inconclusive |
| Arsenic | Leave 50 km blocks out | Statewide figure vs. Nearby tested wells (up to 15 miles) | 8,715 | error 27.0% higher | 95%: from error 42.2% higher to error 13.0% higher | Worse |
| Arsenic | Leave 50 km blocks out | County figure vs. Nearby tested wells (up to 15 miles) | 8,715 | error 10.4% higher | 95%: from error 18.3% higher to error 2.7% higher | Worse |
| Arsenic | Leave 50 km blocks out | Nearby tested wells (up to 15 miles) vs. County figure | 8,715 | error 9.4% lower | 95%: from error 2.7% lower to error 15.5% lower | Relevant improvement, 2% or more |
| Arsenic | Leave 50 km blocks out | Nearby wells of the same aquifer vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.1% higher | 95%: from error 0.9% higher to error 0.7% lower | No relevant improvement |
| Arsenic | Leave 50 km blocks out | Nearby wells of similar depth vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.7% higher | 95%: from error 1.7% higher to error 0.2% lower | No relevant improvement |
| Arsenic | Leave 50 km blocks out | Nearby wells of the same aquifer and similar depth vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.6% higher | 95%: from error 2.0% higher to error 0.6% lower | No relevant improvement |
| Arsenic | Leave 100 km blocks out | Statewide figure vs. Nearby tested wells (up to 15 miles) | 8,715 | error 8.8% higher | 95%: from error 14.1% higher to error 3.6% higher | Worse |
| Arsenic | Leave 100 km blocks out | County figure vs. Nearby tested wells (up to 15 miles) | 8,715 | error 4.4% higher | 95%: from error 8.2% higher to error 0.8% higher | Worse |
| Arsenic | Leave 100 km blocks out | Nearby tested wells (up to 15 miles) vs. County figure | 8,715 | error 4.3% lower | 95%: from error 0.7% lower to error 7.6% lower | Relevant improvement |
| Arsenic | Leave 100 km blocks out | Nearby wells of the same aquifer vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.1% lower | 95%: from error 0.5% higher to error 0.7% lower | No relevant improvement |
| Arsenic | Leave 100 km blocks out | Nearby wells of similar depth vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.2% higher | 95%: from error 0.6% higher to error 0.2% lower | No relevant improvement |
| Arsenic | Leave 100 km blocks out | Nearby wells of the same aquifer and similar depth vs. Nearby tested wells (up to 15 miles) | 8,715 | error 0.0% higher | 95%: from error 0.7% higher to error 0.6% lower | No relevant improvement |
| Arsenic | Forward in time | Statewide figure vs. Nearby tested wells (up to 15 miles) | 2,909 | error 48.5% higher | 95%: from error 73.8% higher to error 27.4% higher | Worse |
| Arsenic | Forward in time | County figure vs. Nearby tested wells (up to 15 miles) | 2,909 | error 12.2% higher | 95%: from error 21.2% higher to error 4.3% higher | Worse |
| Arsenic | Forward in time | Nearby tested wells (up to 15 miles) vs. County figure | 2,909 | error 10.9% lower | 95%: from error 4.1% lower to error 17.5% lower | Relevant improvement, 2% or more |
| Arsenic | Forward in time | Nearby wells of the same aquifer vs. Nearby tested wells (up to 15 miles) | 2,909 | error 1.7% higher | 95%: from error 3.6% higher to error 0.2% lower | No relevant improvement |
| Arsenic | Forward in time | Nearby wells of similar depth vs. Nearby tested wells (up to 15 miles) | 2,909 | error 1.7% lower | 95%: from error 1.4% higher to error 5.0% lower | Inconclusive |
| Arsenic | Forward in time | Nearby wells of the same aquifer and similar depth vs. Nearby tested wells (up to 15 miles) | 2,909 | error 0.0% higher | 95%: from error 3.7% higher to error 3.4% lower | Inconclusive |
| Arsenic | One well out, wells the USGS model could not have seen | Nearby tested wells (up to 15 miles) vs. USGS national arsenic model | 7,904 | error 15.4% lower | 95%: from error 9.9% lower to error 20.6% lower | Relevant improvement, 2% or more |
| Arsenic | One well out, wells with a federal test through 2013 | Nearby tested wells (up to 15 miles) vs. USGS national arsenic model | 811 | error 9.5% lower | 95%: from error 5.4% higher to error 26.9% lower | Inconclusive |
| Arsenic | One well out, all test wells on the USGS map | Nearby tested wells (up to 15 miles) vs. USGS national arsenic model | 8,715 | error 15.1% lower | 95%: from error 10.0% lower to error 20.2% lower | Relevant improvement, 2% or more |
| Nitrate | Leave one well out | Nearby wells of similar depth vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.8% lower | 97.5%: from error 0.8% higher to error 2.5% lower | Inconclusive (fixed in advance) |
| Nitrate | Leave one well out | Statewide figure vs. Nearby tested wells (up to 15 miles) | 9,145 | error 35.9% higher | 95%: from error 57.3% higher to error 17.8% higher | Worse |
| Nitrate | Leave one well out | County figure vs. Nearby tested wells (up to 15 miles) | 9,145 | error 5.6% higher | 95%: from error 11.5% higher to error 1.0% higher | Worse |
| Nitrate | Leave one well out | Nearby tested wells (up to 15 miles) vs. County figure | 9,145 | error 5.3% lower | 95%: from error 1.0% lower to error 10.3% lower | Relevant improvement |
| Nitrate | Leave one well out | Nearby tested wells (up to 15 miles) vs. Statewide figure | 9,145 | error 26.4% lower | 95%: from error 15.1% lower to error 36.4% lower | Relevant improvement, 2% or more |
| Nitrate | Leave one well out | Nearby wells of the same aquifer vs. Nearby tested wells (up to 15 miles) | 9,145 | error 1.3% higher | 95%: from error 2.0% higher to error 0.5% higher | Worse |
| Nitrate | Leave one well out | Nearby wells of the same aquifer and similar depth vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.8% higher | 95%: from error 2.2% higher to error 0.5% lower | No relevant improvement |
| Nitrate | Leave one well out | Nearby wells at the county's typical home-well depth (before drilling) vs. Nearby tested wells (up to 15 miles) | 9,145 | error 3.8% higher | 95%: from error 6.2% higher to error 1.7% higher | Worse |
| Nitrate | Leave one well out | Nearby wells at the county's typical home-well depth (before drilling) vs. Nearby tested wells (up to 15 miles) | 9,145 | error 3.8% higher | 97.5%: from error 6.6% higher to error 1.5% higher | Worse |
| Nitrate | Leave 25 km blocks out | Statewide figure vs. Nearby tested wells (up to 15 miles) | 9,145 | error 28.0% higher | 95%: from error 48.5% higher to error 11.4% higher | Worse |
| Nitrate | Leave 25 km blocks out | County figure vs. Nearby tested wells (up to 15 miles) | 9,145 | error 3.0% higher | 95%: from error 8.2% higher to error 1.3% lower | No relevant improvement |
| Nitrate | Leave 25 km blocks out | Nearby tested wells (up to 15 miles) vs. County figure | 9,145 | error 2.9% lower | 95%: from error 1.4% higher to error 7.6% lower | Inconclusive |
| Nitrate | Leave 25 km blocks out | Nearby wells of the same aquifer vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.3% higher | 95%: from error 1.0% higher to error 0.5% lower | No relevant improvement |
| Nitrate | Leave 25 km blocks out | Nearby wells of similar depth vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.1% higher | 95%: from error 1.3% higher to error 1.2% lower | No relevant improvement |
| Nitrate | Leave 25 km blocks out | Nearby wells of the same aquifer and similar depth vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.7% higher | 95%: from error 2.1% higher to error 0.8% lower | No relevant improvement |
| Nitrate | Leave 50 km blocks out | Statewide figure vs. Nearby tested wells (up to 15 miles) | 9,145 | error 9.2% higher | 95%: from error 18.4% higher to error 1.2% higher | Worse |
| Nitrate | Leave 50 km blocks out | County figure vs. Nearby tested wells (up to 15 miles) | 9,145 | error 4.1% higher | 95%: from error 8.2% higher to error 0.2% lower | No relevant improvement |
| Nitrate | Leave 50 km blocks out | Nearby tested wells (up to 15 miles) vs. County figure | 9,145 | error 3.9% lower | 95%: from error 0.2% higher to error 7.6% lower | Inconclusive |
| Nitrate | Leave 50 km blocks out | Nearby wells of the same aquifer vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.0% higher | 95%: from error 0.4% higher to error 0.4% lower | No relevant improvement |
| Nitrate | Leave 50 km blocks out | Nearby wells of similar depth vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.2% higher | 95%: from error 1.1% higher to error 0.6% lower | No relevant improvement |
| Nitrate | Leave 50 km blocks out | Nearby wells of the same aquifer and similar depth vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.0% lower | 95%: from error 0.7% higher to error 0.8% lower | No relevant improvement |
| Nitrate | Leave 100 km blocks out | Statewide figure vs. Nearby tested wells (up to 15 miles) | 9,145 | error 4.7% higher | 95%: from error 8.8% higher to error 0.6% higher | Worse |
| Nitrate | Leave 100 km blocks out | County figure vs. Nearby tested wells (up to 15 miles) | 9,145 | error 2.1% higher | 95%: from error 4.9% higher to error 0.7% lower | No relevant improvement |
| Nitrate | Leave 100 km blocks out | Nearby tested wells (up to 15 miles) vs. County figure | 9,145 | error 2.0% lower | 95%: from error 0.7% higher to error 4.7% lower | Inconclusive |
| Nitrate | Leave 100 km blocks out | Nearby wells of the same aquifer vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.1% lower | 95%: from error 0.1% higher to error 0.4% lower | No relevant improvement |
| Nitrate | Leave 100 km blocks out | Nearby wells of similar depth vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.0% higher | 95%: from error 0.5% higher to error 0.4% lower | No relevant improvement |
| Nitrate | Leave 100 km blocks out | Nearby wells of the same aquifer and similar depth vs. Nearby tested wells (up to 15 miles) | 9,145 | error 0.2% lower | 95%: from error 0.3% higher to error 0.7% lower | No relevant improvement |
| Nitrate | Forward in time | Statewide figure vs. Nearby tested wells (up to 15 miles) | 2,988 | error 28.0% higher | 95%: from error 48.9% higher to error 12.0% higher | Worse |
| Nitrate | Forward in time | County figure vs. Nearby tested wells (up to 15 miles) | 2,988 | error 1.5% lower | 95%: from error 4.0% higher to error 6.7% lower | Inconclusive |
| Nitrate | Forward in time | Nearby tested wells (up to 15 miles) vs. County figure | 2,988 | error 1.5% higher | 95%: from error 7.2% higher to error 3.8% lower | Inconclusive |
| Nitrate | Forward in time | Nearby wells of the same aquifer vs. Nearby tested wells (up to 15 miles) | 2,988 | error 0.3% lower | 95%: from error 1.5% higher to error 2.0% lower | No relevant improvement |
| Nitrate | Forward in time | Nearby wells of similar depth vs. Nearby tested wells (up to 15 miles) | 2,988 | error 1.3% higher | 95%: from error 4.2% higher to error 1.3% lower | No relevant improvement |
| Nitrate | Forward in time | Nearby wells of the same aquifer and similar depth vs. Nearby tested wells (up to 15 miles) | 2,988 | error 1.7% higher | 95%: from error 4.8% higher to error 1.1% lower | No relevant improvement |
Sources you can check
- Texas Water Development Board, Groundwater Database. File of November 13, 2026: the core of every figure. How the state explains its database.
- Water Quality Portal (USGS and EPA), download of November 15, 2026: well sites in Texas, after removing duplicates of state records.
- USGS, national model of arsenic in private wells (data) and the article that describes it: Lombard and others, 2021, Environmental Science & Technology.
- This dataset on Zenodo: the CSV files, a README with the method, the license and checksums. DOI 10.5281/zenodo.24287393.
How to cite this
Blanco, R. (2026). "What best estimates arsenic and nitrate in a Texas well? A study with its rules fixed in advance." Texas Well Water Quality Atlas. TerraFunded. https://terrafunded.com/water-quality/study. 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, 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]
