Environmental and gas monitoring guidance for audited field teams
Application note

The Reading Looked Fine: Four Instruments, Four Levels of Verification

2026-11-06 Luc Moreau
Measurement team reviewing calibrated environmental monitoring data

There's no standard answer to 'how much should I trust this reading?'

I've been specifying and verifying instruments for environmental and process monitoring for about eleven years. In that time I've made — and written down — a dozen or so mistakes that added up to roughly $30,000 in wasted budget, one rejected dataset, and a very awkward call with a plant manager. (The dollar figure is an estimate; I stopped keeping an exact running total after the first few.)

Almost all of those mistakes had the same shape. I applied the wrong level of verification to the wrong instrument. I over-tested things that would have announced their own failure, and I under-tested things that fail silently while continuing to output respectable-looking numbers.

The four situations below are where I've personally gotten it wrong. They're not organised by instrument type so much as by what a wrong number actually costs you.

  • Continuous water quality monitoring — high consequence, almost zero visibility into failure.
  • Pressure gauges — low consequence individually, but you own fifty of them.
  • Insulation resistance testing — the number is meaningless without the conditions it was taken in.
  • Reading a utility water meter — trivial to read, easy to misread, and rarely the actual problem.

If you're dealing with something else entirely — flow meters, calipers, torque wrenches — the reasoning probably transfers, but the specifics won't. I can only speak to the four I've broken myself.

1. Continuous water quality monitoring: clean more, calibrate less

This is the one that cost me a dataset.

We were running multiparameter sondes — pH, dissolved oxygen, turbidity, and an optical nitrate channel — at six stations in a reservoir. Calibration sat on a fixed 30-day schedule, because that's what the maintenance contract said. In August 2021 the nutrient data came back for review and one station's nitrate series had gone flat-ish for about three weeks. Not obviously wrong. Just smoother than the other five.

The cause wasn't drift. It was biofouling on the optical path. Algae had grown across the measurement window and the sensor kept logging numbers that were plausible, internally consistent, and wrong. We flagged the affected period, re-sampled what we could, and ended up excluding roughly six weeks from the report. That's the expensive version of this mistake — the data still exists, it just can't be used for anything.

Here's the part that runs against how most monitoring programs are set up: calibration frequency is usually not the lever you think it is. The optical channels I've worked with — the YSI product line included, EXO NitraLED and the rest — are pretty stable over a month. What isn't stable is the window the light passes through. Fouling, not drift, is the dominant error source on any optical channel sitting in biologically active water.

So we moved to: clean every two weeks (weekly through summer), calibrate every four to six, and record the reading before and after every cleaning. That last part matters more than either interval. If the post-cleaning reading is within a few percent of the pre-cleaning one, your fouling rate is under control. If it jumps 15%, you've been reporting bad numbers for two weeks and had no way of knowing.

Standards help but don't settle it. ISO 15839 covers performance testing for on-line water sensors — check you're on the current version — but no standard can tell you how fast your particular water fouls a lens. Only your own log can.

Context caveat: this worked for a temperate reservoir with a monthly grab-sample cross-check against a lab method. If you're in high-productivity water, an effluent stream, or a warm-climate system, your fouling cycle might be measured in days rather than weeks. Don't take my interval — take my method for finding yours.

2. Pressure gauges: buy the cheap one and throw it away

I get the most pushback on this one, so I'll say it plainly. For most plant and process installations, you should not buy the expensive gauge.

EN 837-1 sorts Bourdon-tube gauges into accuracy classes — 1.0, 1.6, 2.5, and coarser. People spec a 1.0% gauge for a decision that is genuinely 'is this line at roughly 4 bar or roughly 6?' A 2.5% gauge answers that question just as well for a fraction of the price, and when it dies you bin it instead of raising a calibration requisition and waiting three weeks.

I went back and forth between rebuilding the existing gauges and replacing them outright for the better part of two months. Rebuilding won on sentiment. Replacing won on arithmetic, comfortably, once I actually wrote the numbers down.

The failure mode is what converted me. We had a gauge reading a steady 5.2 bar on a discharge line for what turned out to be several months. It was stuck — the tube had taken a set — but it was stuck at a number that looked exactly like normal operation. Nobody questioned it until a planned pump shutdown, when the gauge cheerfully sat at 5.2 bar on a dead line.

The check that would have caught it takes two minutes: close the isolation valve, vent the gauge, watch it fall to zero, close up, reopen. A gauge that doesn't return cleanly to zero is finished. Fifty gauges is an hour and a half a year, and it will find more real problems than any calibration program I've run on the same assets.

What I'd actually spend money on: liquid fill for anything with vibration, a snubber or restrictor on anything with pulsation, and a proper isolation valve so the two-minute check is genuinely two minutes. The gauge head itself? Mid-grade, and keep spares on the shelf.

Where this doesn't apply: if the gauge feeds a custody transfer or a safety interlock, ignore everything above and buy the certified one. A 0.25% gauge wired into a trip system isn't a gauge, it's a safety device, and the usual cost logic doesn't hold.

3. Insulation testing: the number is meaningless without the conditions

An insulation tester — the 1503 and its relatives — gives you a resistance reading. What it does not give you is a verdict. That gap is where I've made my expensive mistakes.

Two things have to be right before you read anything off the screen. First, test voltage. IEC 61557-2 covers what the tester itself must do to qualify as a compliant instrument, but it's the rating of the equipment under test that decides whether you apply 500 V or 1000 V — and pushing the higher voltage at something that isn't rated for it is a fine way to convert a test into a repair. Second, conditions. Insulation resistance falls as temperature and moisture rise, and the rule of thumb I was taught is that it roughly halves for every 10°C. I've never verified that tightly enough to defend it in an argument, but the direction is right and the magnitude is large enough that you can't wave it off.

Dodged a bullet on the second one, incidentally. We'd already written off a motor as failed and were one signature away from a rewind quote when a colleague asked what the test conditions had been. Moisture in the terminal box. Two days in a dry room and it tested fine.

The one that still annoys me is the pump we pulled out of service on a low reading. If I remember correctly it was somewhere in the low single-digit megohms, though I might be misremembering the exact figure. The number wasn't wrong. I just never asked what produced it. We lost a weekend of downtime and generated a rewind quote we didn't need.

The habit now: record test voltage, ambient temperature, and whether the equipment had been running. A reading with no conditions attached isn't data, it's a rumour. And compare against the previous test on the same equipment rather than a generic pass/fail threshold — the trend matters more than the absolute value, because it's measured under conditions close to the ones you had last time.

4. Reading a water meter: the meter is almost never the problem

Reading a Sensus meter — or any meter with a mechanical or digital register — is straightforward. The register digits are your billing units. Anything after the decimal (often in a contrasting colour on mechanical registers) is fractional and usually not billed. The sweep hand or flow indicator is the genuinely useful part: if it's moving, water is passing right now.

The step everyone skips: confirm the units on the register before doing any arithmetic. Cubic feet and cubic metres look similar at a glance and differ by a factor of about 28. I've watched a three-month consumption report get built in the wrong unit and presented to a client before anyone noticed.

But here's the lesson that actually cost me time. When someone calls and says the meter's wrong, it usually isn't. The test that settles it takes ten minutes: shut off every fixture in the building, then watch the flow indicator. If it's still turning, water is going somewhere — an irrigation valve, a toilet flapper, a slab leak, a softener regenerating at the wrong hour. I once spent the better part of a week assembling a case for a meter replacement before a plumber found a leaking irrigation solenoid. The meter had been right the whole time.

If the indicator is genuinely dead still and the bill still doesn't reconcile, then you have a meter question — and that's the point at which you call the utility, not the point at which you start.

So which one are you? Four questions

I'd rather spend ten minutes helping you work out which of these you're dealing with than six weeks explaining why a number didn't get a second opinion. So:

  1. Would a wrong reading look plausible? If yes — water quality sensors, insulation resistance — you're in scheduled-verification territory, because nothing else will catch it. If failure produces an obviously broken value (zero, pegged, dead display), you can afford to be lazy. The instrument announces its own lies.
  2. How long would a check against an independent method take? Under five minutes, do it every time and stop reading articles about it. Over a day, build the schedule and the log instead, because you will not do it spontaneously.
  3. What's the replacement cost to verification cost ratio? If a new unit costs less than a calibration, replace on a schedule and stop thinking about it. That's the pressure gauge logic and it applies to far more instruments than people admit.
  4. Is the number going into a report, a bill, or a safety decision? Then none of the shortcuts apply. Full verification, conditions recorded, second method cross-checked, every time.

Most of what I know came from getting the allocation wrong in one direction or the other. The point isn't to verify everything harder. It's to put the effort where a bad number can hide.

Luc Moreau

Luc Moreau

Luc Moreau is a large-scale dimensional systems analyst covering coordinate measuring machines, laser trackers, laser interferometers, video measuring machines, and profile projectors. He works from ISO 10360-2 and ISO 10360-10 concepts to compare probing error, length-measurement error, volumetric performance, environmental compensation, and maximum permissible error. He helps metrology laboratories and manufacturing engineers plan acceptance tests, estimate usable uncertainty, and match system capability to part size and tolerance.

Ask about this topic

Turn the article into a site-specific measurement brief.

Share the device family, range, approval region and document expectation so the follow-up can stay practical.