How these water-loss numbers are produced

CA Urban Retail Water System Water Loss Annual Loss Audit Data, published by the California Department of Water Resources on the state open-data portal. Each row is one AWWA M36 water-audit submission for one supplier and one accounting period. The audit year is taken from the reporting start date, because the source year column is not populated.

View the source dataset

The state file is checked once a week. Each check is logged: the source is downloaded, fingerprinted, and compared with the file behind the currently published data. If it is byte-identical the run is recorded as “no change” and nothing is republished.

When the file has changed, rows are normalized into a staging area, validated (required columns present, at least half the rows usable, and no more than a 20% drop against the currently published audit count) and only then promoted in a single atomic publish that also rebuilds trends. A failed or partial refresh leaves the previously published dataset untouched, so the public pages never show half-loaded data.

Suppliers report volumes in either acre-feet or million gallons, declared per audit. Everything is converted once, in one place, at 1 acre-foot = 325,851 gallons. Displayed volumes always carry their unit; acre-feet and gallons are never mixed in a single figure.

• Real loss per service connection per day = real-loss gallons ÷ service connections ÷ days in the reporting period. The primary comparison measure, because it is independent of system size.
• Real loss per mile of main per day = real-loss gallons ÷ miles of main ÷ days in the reporting period. Used for systems with few connections per mile.
• Real loss as % of water supplied and apparent loss % and total loss % are shares of the volume of water supplied in the same audit. Labels always name the denominator.
• Infrastructure leakage index is reported by the audit itself and passed through unchanged — it is never recalculated here.

A metric is only produced when its denominator is valid: connections above zero, mains length above zero, and a reporting period between 300 and 400 days. Otherwise it is null, never zero.

Trends compare real loss per connection per day across a system's own audits: latest versus prior year, latest versus three years earlier, and latest versus earliest available. A change is described as rising or falling only when it exceeds 1 gallon per connection per day, so audit noise is not reported as a trend.

Marker colour is a descriptive band, not a compliance grade. Thresholds are the statewide quartiles and top decile of real loss per connection per day across every system's most recent audit.

Systems are located by public water system identifier (PWSID). Service-area polygons come from the EPA ORD Service Area Boundary model; centroids come from California SABL drinking-water service areas where available, otherwise from the same PWSID-matched service-area data. Matching is PWSID-exact — no name or fuzzy matching is used — and each system profile states which source and match method produced its location.

Suspicious records are flagged, never deleted or silently corrected. Flags include negative volumes, losses exceeding water supplied, non-positive connection or main counts, implausible reporting periods and non-standard identifiers. Flagged audits still appear in the record table so anyone can see exactly what the utility filed.

Open the internal data check for source provenance, completeness and metric eligibility counts.

Water loss is not the same as waste, and a higher number is not automatically mismanagement: pressure, pipe age, soil, metering practice and audit data quality all shape it. These pages report what utilities filed and make it comparable; interpretation belongs to the reader.