Why one Idaho well record claims 15,187 gallons per minute

One record, 15,187 gallons per minute

A 1997 well record in Canyon County reports a yield of 15,187 gallons per minute. The next highest in that county is 80.

Fifteen thousand gallons a minute is close to 22 million gallons a day — municipal supply for a small city, from a single domestic well log. It is almost certainly a daily total typed into a per-minute field.

Nobody is at fault. The record is a scan of a form filled in by hand at a drill site nearly thirty years ago, and the state publishes what was filed. The interesting question is what a website does with it.

What one bad record does to a range

Published as a maximum, that single figure would define an entire county. "Yields in Canyon County run from 1 to 15,187 gpm" is technically traceable to the source and completely useless — worse than useless, because it makes every other number on the page look unreliable.

This is why these pages report the 10th to 90th percentile rather than the outright minimum and maximum. The extremes are where data entry errors live; the middle 80% is where the information is.

The same problem in three other places

• About 4.8% of Idaho well records carry a construction date of 9999-12-31, a placeholder meaning "not reported". Rendered literally it prints as a well drilled in the year 9999.
• A static water level or yield of exactly 0 is ambiguous — it can mean artesian flow, or that nobody wrote a figure down. Excluding zeros moved one median from 12 ft to 29 ft.
• Twenty-two records statewide report a depth shallower than 18 ft, which is Idaho’s own threshold for what counts as a well. They were setting the minimum on county pages.
• FEMA flood data uses -9999 for a base flood elevation that does not apply, which reads as an elevation below sea level.

Why this is published rather than quietly handled

Every number on this site states how many records it came from and what was excluded, because a figure without its denominator is a claim rather than a measurement.

The errors above are not a reason to distrust the dataset. They are a reason to read anyone presenting it who does not mention them.