Water hardness data & methodology

Source tiers

Each utility contributes its best available value in strict tier order (T1a, then T1b, then T2). County ambient data (T3) only covers cities with no defensible tap-tier value — it is always labeled as an estimate.

A modeled T4 layer (predicting hardness for cities no tier covers) is deliberately NOT included: it is deferred behind an explicit quality gate, and no modeled value ships until it passes. Uncovered cities stay honestly uncovered.

Formulas & thresholds

• Hardness by calculation (T2)

  hardness (mg/L as CaCO₃) = 2.497 × Ca (mg/L) + 4.118 × Mg (mg/L) — Standard Methods 2340 B, verified against USGS Scientific Investigations Report 2025-5010 (eq. 15).

• Grains per gallon

  1 gpg = 17.1 mg/L as CaCO₃ — the standard rounding of the exact 17.11806 factor in NIST Special Publication 811, Appendix B.8. Every gpg figure on the site converts at 17.1.

• Classification bands

  Soft 0–60, moderately hard 61–120, hard 121–180, very hard above 180 mg/L as CaCO₃ — the USGS classification guidelines.

How city values are aggregated

• Per utility: the best available tier wins, in the order T1a, then T1b, then T2.
• Per city: the population-weighted mean across the city's utilities (weight = people served by each utility). If no utility reports a served population, a simple mean is used.
• The city's displayed tier is the tier supplying the largest population share (ties go to the better tier).

Conflicting values: the disputed rule

County estimates (T3): the Water Quality Portal pull

• Window: results collected 2000-01-01 to present, so medians reflect modern analytical methods; every T3 citation discloses the window ("2000–present"). Current national pull retrieved 2026-07-04.
• Characteristics: direct hardness results ("Hardness", "Hardness, Ca, Mg", "Total hardness"), normalized to mg/L as CaCO₃; partial-hardness characteristics (carbonate / non-carbonate) are excluded. Elemental calcium + magnesium results are combined via the signed formula ONLY where a county has zero usable direct results.
• Value: the median of the normalized results in the county.
• Site types: unfiltered — streams, lakes, wells, springs, and facilities all contribute, because utilities draw on both surface water and groundwater. The semantics are "ambient county water hardness, all site types" — a context value, never a tap measurement.
• Sample-count display: every county estimate is displayed with its sample count (n), county, and window; county-context values on measured cities disclose their n too. A zero-sample county publishes nothing — the city stays uncovered and is counted.
• County assignment: each city maps to the majority county across its ZIP codes using a vendored US Census 2020 ZCTA-to-County crosswalk (the HUD-USPS crosswalk is preferred but requires a registered token; ZCTAs are Census approximations of ZIP codes — see limitations). A ZCTA spanning multiple counties keeps the county with the largest land-area overlap; remaining ties break to the smallest county FIPS, deterministically.

Corrections: overrides & sanity bounds

How the dataset is verified

• Signed constants: every threshold and factor above lives in a signed source-of-record file pinning its citation and live primary-source URL (USGS, NIST); the build reads the constants from there, never from memory.
• Golden sample: 27 cities spanning all four bands, tiers, and regions are hand-verified against their PRIMARY sources — the utility's own CCR or annual water-quality report, or an independent Water Quality Portal county pull.
• Automated drift gate: every test run re-verifies the committed dataset end to end — provenance completeness (value, unit, tier, source, date on every entry), sanity bounds, classification-band agreement, disputed-range coherence, county sample counts, and golden drift within tolerance. A violation fails the build; nothing ships around the gate.

Limitations

• Utility averages, not your tap: a city value is a distribution-system average; individual taps vary with treatment plant, blending, season, and plumbing.
• Report lag: T1 values reflect the most recent published utility report, which may lag current conditions — every value carries its source date.
• ZCTA ≈ ZIP approximation: county assignment uses Census ZCTAs, which approximate USPS ZIP codes; PO-box-only and some rural ZIPs resolve to no county, and those cities stay uncovered rather than guessed.
• Aggregation limits: a multi-utility city is a population-weighted mean, and one number can mask real differences between the city's utilities; conflicts beyond tolerance surface as the disputed flag and range.
• T3 is ambient, not tap: county estimates measure ambient water (streams, lakes, wells) across the county — not treated tap water.
• Honest gaps: 2,214 tracked cities have no defensible value in any tier; they are published as uncovered and counted, never filled with a guess.

Data sources

Every figure traces to one of these references. The exact citation for a given city (utility + PWSID, or county + sample count + window) is shown with that city's value.

• Utility-reported water quality data (utility testing reports & Consumer Confidence Reports)T1 city values (utility-reported hardness) and the calcium/magnesium inputs behind every computed T2 value, cited with the utility name and PWSID.
• EPA/USGS Water Quality Portal (waterqualitydata.us)T3 county ambient medians and the labeled county-context values shown alongside measured cities.
• USGS Water Science School — hardness classificationThe soft / moderately hard / hard / very hard bands (60 / 120 / 180 mg/L as CaCO₃).
• USGS Scientific Investigations Report 2025-5010 / Standard Methods 2340 BThe hardness-by-calculation formula (2.497 × Ca + 4.118 × Mg) behind every computed T2 value.
• NIST Special Publication 811, Appendix B.8The 17.1 mg/L per grain-per-gallon conversion behind every gpg figure on the site.
• U.S. Census Bureau 2020 ZCTA-to-County relationship fileThe vendored ZIP-to-county crosswalk that assigns each city a county for T3 estimates.

Changelog

• July 6, 2026 — Initial public release: 16,561 of 18,775 US cities covered (6,778 utility-reported, 836 computed from utility-reported minerals, 8,947 county-level estimates).