How dawn thermography narrows the search for canal seepage

You have more canal than you have money to line

Ponding gives you a real loss number. It also needs the water off, a plugged reach, and a crew for a day or two. You can’t do that to twenty miles of canal to find the bad two.

Walking the bank is cheap. It finds what shows from the bank.

A dawn thermal flight covers ground fast. It does not measure loss. What it does is tell you where to point the methods that do.

Nationally, conveyance loss is not a small line item. In 2019 it accounted for more than 14 percent of the water entering irrigation organizations’ delivery systems. Your reach is its own question. But the scale is why the question keeps coming up.

This piece is about how to use a thermal flight honestly. What it reads. When it works. What the published numbers mean. And where it sits in a decision a board has to defend.

If you came for one thing:

• Why a wet spot isn’t always cold
• Why we fly at dawn, and the timing rule nobody has published
• What “91 percent” actually means
• Everything else that makes a warm spot
• What confirms a finding
• What a board and a funder actually score

What the camera actually reads

A thermal camera does not see water. It does not see moisture, or a void, or how fast something is leaking.

It reads long-wave infrared coming off the exposed surface. Then it converts that reading into a temperature. The conversion runs on assumptions. Emissivity, which is how readily a surface gives off infrared. Reflected temperature. Air, distance, and the calibration of the instrument. Change an assumption and you change the number. We set emissivity and reflected temperature before the flight, and we log both.

So what does seepage do to a surface?

Three things, and they overlap.

Thermal inertia. Wet soil stores more heat than dry soil. It warms slower and it cools slower.

Thermal conductivity. Water in the pore spaces moves heat between soil grains better than air does.

Advection and evaporation. Moving water carries the canal’s own temperature into the bank. A damp surface also sheds heat as that water evaporates.

Here is the part most explanations skip.

The sign of the anomaly is not fixed.

After a clear night, a damp seep zone can read warmer than the dry bank beside it. The dry bank shed its heat fast. The wet ground held on.

After a day of sun, that same ground can read cooler. Thermal inertia kept it from heating. Evaporation pulled more heat off the surface.

This is established soil physics, not a field quirk. Wetter ground runs cooler by day and resists cooling at night. Laboratory work on scale models of earthen dams found the same reversal in a seepage setting. The flip tracked time of day and solar load. Those were dam models, not canals — but the mechanism is the one at work on a canal bank.

So anyone who tells you leaks look cold is describing one time of day and calling it a rule.

We don’t read a color. We read a difference, under conditions we wrote down, at a time we can repeat.

Why dawn, and the rule nobody has published

Dawn is a contrast window. It is not a law.

By first light the bank has had a full day of sun and a full night of cooling. That cycle is what builds the difference between wet ground and dry ground. Fly before the sun starts working the bank directly, and you read that difference.

Once direct sun hits, slope and aspect and every shrub start writing their own temperature patterns across the bank. Those patterns have nothing to do with water.

So the window is tight, and we’re off the reach before it warms.

Now the part worth asking any vendor about.

There is no published rule that says fly thirty minutes before sunrise. There is no published temperature difference that means seepage. We went looking for both. Neither exists in the peer-reviewed or agency literature for canals.

Where a rigorous method does use a threshold, it comes with conditions attached. One study of temperature sensing in river embankments worked to a 0.2 degree Celsius difference. The authors were explicit about what it depended on. How deep the flow path sat. How long it was. The season it was measured in.

That’s the honest shape of it. A threshold belongs to a site, a season, and a setup.

Canals aren’t the only structure people point a thermal camera at, and the findings don’t all agree. Work on grassed earth dams points at afternoon and night as the better windows. Different structure, different cover — and another reason to treat a window as site-specific rather than settled.

So our window and our severity bands are an operating spec. We develop them on the ground we’re flying and we log them with the flight. They are not a published standard, and we won’t present them as one.

Before you rely on a number, ask where it’s published.

What has to be true on the morning we fly

A dawn flight works under conditions. Here’s what we check before one is worth doing.

Water in the channel. Seepage needs head behind it. An empty channel can still show residual moisture or a change in material. It can’t show you what’s leaking now.

A day of sun before it. The contrast we read is built by daytime heating and overnight cooling. A stretch of overcast flattens that cycle.

Clear or documented sky. Clouds change how the ground sheds heat overnight. Broken cloud can leave one stretch of bank with a different thermal history than the next.

Low wind. Wind pulls heat off the surface and can wash out a small difference.

Stable operation. A changing stage changes the seepage gradient underneath it.

Time for water to arrive. Water has to move through the bed or the bank and reach the surface. Until it does, the surface has nothing to show.

Two of those come with an admission. Nobody has published a maximum wind speed or a cloud limit for canal work. Both effects are well understood. Neither has a number attached to it. So we work to conservative limits, we log the conditions, and we tell you when a morning wasn’t good enough.

Then the Idaho part.

Seasons vary by district. Settlers Irrigation District puts its normal season at April 15 through October 15, depending on water supply.

But water in is not the same as stable operation. During filling the front is still moving and the stage is still climbing. Bank that was dry last week is wetting for the first time. That’s real moisture. It isn’t a leak. Early-season flights read hardest, and we’d rather fly a reach once it settles.

Every flight carries a condition log: air temperature, wind, cloud history, canal water temperature, and exact capture time. That isn’t paperwork. It’s what lets someone interpret the result, and it’s what makes next year’s flight comparable to this one.

Seasons and operating windows vary by district. More on who we work with at irrigation districts.

What “91 percent” actually means

Two numbers get quoted constantly in this field. Both are real. Both are usually described wrong.

In 2002, a Texas A&M evaluation flew thermal imagery over an irrigation district’s canals. The imagery flagged 45 candidate sites. District crews went to 11 of them. Ten had leaks.

That’s 91 percent. Here’s what it means, and what it doesn’t.

It means the flags were usually worth walking to. When the imagery marked a spot, someone who went and looked usually found something.

It does not mean the flight found 91 percent of the leaks. Nobody walked the full canal to count what the flight missed.

A 2009 study in the Lower Rio Grande Valley reports the same shape of number. Twenty-six of 28 checked sites confirmed, or 93 percent. Read it the same way. That work also used a camera reading red, near-infrared, and thermal bands from a manned aircraft — not thermal alone.

The oldest study is the most revealing.

A 1979 flight over a central Oregon district flagged 39 sites. Twelve were confirmed. That’s 31 percent. It was flown with the instruments of its era. So it shows you what the method’s history looks like, not what a calibrated radiometric sensor does today.

Most of those flags were not confirmed. Note the wording. Twenty-seven sites weren’t proven to be nothing. Crews looked and found nothing, which is a different claim.

Field work also turned up no leaks outside the flagged set. That’s weak evidence the flight missed little. Weak, because nobody went hunting through the ground the imagery hadn’t marked.

So all three numbers are precision figures. They tell you how often a flag is worth the walk. They tell you nothing about what a flight misses.

No published source gives a detection rate for canal thermography. Not a probability of detection, not a false-positive rate, not a minimum leak size. We went looking. It isn’t there.

Anyone quoting you a detection rate is quoting a precision figure with the wrong label on it.

We use these numbers the way the studies support them. A flag is worth checking. That’s the claim.

Everything else that makes a warm spot

Seepage is one reason a patch of bank reads different from the bank around it. It is not the only reason, and the list is long.

Sun and shade. A shaded slope and a sunlit slope are different temperatures. So are a north bank and a south bank.

What the bank is made of. Soil texture, compaction, rock, concrete, asphalt, liner thickness — all of it changes how the surface holds heat.

Plants. Vegetation, roots, irrigated landscaping next to the canal, and the cooling that comes with plants losing water.

Water that isn’t from the canal. Recent rain, a sprinkler, road runoff, a drain, or shallow groundwater sitting close to the surface.

Ground that’s been disturbed. Rodent burrows, fresh cuts, maintenance work.

Things buried or built. Culverts, pipe, utilities, structures, concrete transitions.

Surface conditions on the morning. Standing water, dew, frost, snow, moisture left over from maintenance.

The physics of the reading itself. Smooth water reflects. A pixel that covers soil and plants and concrete at once averages them. And an emissivity assumption that’s wrong for the surface produces a temperature that’s wrong with it.

Every one of those can look like seepage in a thermal image.

This is why our anomalies ship as candidates. Each one gets a station, an offset, a temperature difference, and a severity rank. And a flag saying it needs a ground check. A flag is a place to look. Nothing in the register is a conclusion, and we don’t write it as one.

It’s also why the visible imagery comes with the thermal, registered to it, from the same pass. Half the list above shows up plainly in the visible image. The culvert, the burrow, the sprinkler, the patch of asphalt. You rule those out before anyone walks the bank.

What a thermal flight cannot tell you

Plainly, so it’s settled before you hire anyone.

It cannot tell you how much water you’re losing. Not in cfs. Not in acre-feet per season. Not per mile.

It cannot tell you where the water goes. No depth, no flow path.

It cannot tell you the water is yours. Canal water, irrigation water, rain, groundwater, and a leaking utility all wet the ground the same way.

It cannot tell you what failed. A liner defect, permeable soil, a bad joint, rodent damage, internal erosion — the surface expression can look alike.

It cannot tell you whether it’s getting worse. One flight is one morning. Change needs a second flight against the first.

It cannot tell you the bank is going to fail. No factor of safety. No breach prediction. Thermal is not a geotechnical evaluation.

Reclamation’s framework for investigating seepage out of unlined canals treats these as separate jobs. Locating candidate areas, measuring what a reach loses, and diagnosing what’s below the surface are different jobs. They use different methods, in sequence, from lower cost to higher. Thermal sits at the front of that sequence. It is a screening layer.

Screening layers are worth money because the layers behind them are expensive. That’s the entire argument for flying first.

What thermal does give you is a place to start. And a documented record of where you looked and what you found.

What confirms a finding

A thermal flight hands you candidates. Here’s the chain that turns a candidate into something you can spend money on.

Fly it again. Same reach, same geometry, similar weather and operating conditions. An anomaly that shows up twice is worth more than one that showed up once. One that doesn’t repeat was probably the morning, not the bank.

Walk it. Your ditch rider or watermaster goes to the station and looks. Saturation, changed vegetation, sloughing, cracking, settlement, boils, burrows, a liner defect, a drain someone forgot about. Much of what the register flags gets resolved standing on the bank.

Measure at the spot. Contact temperature, soil moisture, elevations where the geometry matters. This is where a candidate becomes a real wet area with a size.

Tie it to the canal. Compare canal stage against when the wet area responds. If the ground wets when you raise stage and dries when you drop it, the canal is the source. If it doesn’t track, look elsewhere — a sprinkler, a drain, shallow groundwater.

Put a number on the reach. This is the step thermal can’t do. NRCS lists four accepted ways to estimate conveyance loss. Measuring inflow and outflow across a reach. Controlled ponding. Seepage meters. Estimates from the bed material. NRCS calls controlled ponding one of the most accurate. It also notes it has to be done during a non-operation period.

That trade-off is the whole reason screening is worth paying for. Ponding gives you the defensible number. It also costs you the water off and a crew on site. You want it aimed at the right two miles, not the whole twenty.

The ponding method itself is documented in published extension guidance. Isolate a reach. Measure the water level drop. Correct for canal geometry, evaporation, and any rain. Your engineer will know it. It isn’t our work.

Look below the surface, if that’s the question. If what you need to know is the pathway or the condition of the embankment, that’s a different toolkit. Geophysics, borings, piezometers, or an observation well. Electrical methods, ground-penetrating radar, and instrumented monitoring all answer questions a camera can’t reach.

Real quantification of seepage using temperature looks nothing like a flight. A USGS study in Nevada’s Walker River Basin buried calibrated probes at multiple depths. It tracked canal stage, monitored groundwater head, and took sediment cores. Then it ran the whole thing through a numerical model of coupled water and heat flow. It produced local seepage rates. On the order of 0.01 to 2.5 feet per day across the Smith Valley transects.

That is what it takes to convert temperature into a loss rate. Probes in the ground and a model behind them. Not a camera above it.

Reclamation’s framework says the same thing in the order it puts the methods. Cheaper and broader first, expensive and specific after. Thermal is the front of that line. Steps two through six are where the answer gets built.

What a board and a funder actually score

Districts don’t rank reaches by seepage. Seepage is one input.

What goes into the decision:

• How much water the reach is losing
• What’s below it — homes, roads, businesses — and what failure would cost
• Whether the reach limits delivery or capacity
• How old the liner is and what maintenance it’s taken
• Access, crossings, utilities, easements
• Whether it can be built during the winter outage
• Where conserved water goes and what that means for water rights
• What the effect is on shallow groundwater
• Cost per acre-foot conserved over the life of the project
• Whether match funding exists and whether you’re ready to apply

A high thermal anomaly on a reach with nothing below it and a good liner is a maintenance item. A moderate anomaly on a reach perched above a subdivision is a different conversation. The imagery doesn’t rank them. You do.

Then the funding side, stated straight.

Grant records ask for the annual acre-feet a project will conserve, and the method behind that estimate. They ask for a defined reach, a design, and existing conditions. Cost and cost share. A board resolution, a schedule, environmental readiness. And where the conserved water ends up.

Thermal does not produce the acre-feet number. That comes from ponding or a reach water balance. If a vendor implies their imagery fills that line on your application, they don’t know what’s on the application.

What thermal contributes to the file is real, and it’s this:

• A documented reason you selected this reach and not another
• Georeferenced imagery of where you looked and what you found
• A written record of the conditions and method, so the work is repeatable
• A defensible basis for aiming ponding or geophysical work, which is what makes the loss estimate affordable
• Before-and-after imagery, if you fly the reach again after construction

That last one is worth a caution. No funding program we reviewed requires thermal imagery before or after a project. It’s supporting evidence, not a checkbox. Anyone who tells you a grant requires it is selling.

The honest summary: thermal strengthens the record behind your decision. It doesn’t fill the line the funder scores.

If you’re weighing which reaches to line, the canal decision page lays out where a thermal flight fits.

Lining isn’t automatically the right answer

Here’s the part a vendor selling lining justification would leave out.

Water that leaves an unlined canal doesn’t disappear. It goes into the ground. In an irrigated valley, some of that water comes back. Into shallow groundwater, into drains, into return flow, into wells that people are pumping.

So a leak is a loss to the district and, sometimes, a supply to somebody else. Seal it and that water stays in the canal. It also stops arriving wherever it had been arriving.

How much that matters depends on the ground under the specific reach. Shallow aquifers and deep regional aquifers don’t get recharged the same way, and the answer is local. It’s a hydrogeology question. Your engineer or the state can answer it for your reach better than any camera can.

Four questions worth answering before a reach gets lined:

• Who is using the water that currently seeps out, and when do they get it?
• Does it support wells, drains, or river return flow — or does it just raise local groundwater where nobody wants it?
• Where does the conserved water go once you stop the leak? Downstream? Additional deliveries? An obligation you owe somewhere else?
• Would deliberate recharge, at a place and time you choose, be worth more than the leak you have now?

None of that is an argument against lining. Plenty of reaches should be lined, and some should have been lined already.

It’s an argument for knowing which kind of reach you’re looking at first. Before you spend a million dollars on a mile of it. A thermal flight tells you where the water is leaving. It doesn’t tell you whether that’s a problem.

What you actually receive

We publish a full specimen set so you can see the deliverable before you talk to us. It’s demonstration data throughout — every plate is stamped, and no value in it came off a real canal.

Six plates.

The stationed hotspot map. Candidates marked by station and offset along the flown reach. PDF and KMZ, so the points open in your GIS.

The anomaly register. One row per candidate, worst first. Each row carries an ID, station, offset, the temperature difference in degrees Celsius, a severity rank, and a status. In the specimen set, all four rows read verify on ground. That isn’t a formality. A flag is a place to look.

Thermal and visible orthomosaics. The same reach in both bands, from the same pass, registered to one another. This is what lets you rule out the culvert and the sprinkler before anyone drives out.

A photogrammetric surface model. Bank and channel geometry, for cross-sections and for comparison against a later flight. Built from overlapping camera capture — photogrammetric, not LiDAR. The plate says so on its face, and states shape only: no datum, no elevations.

The method report. Cover, method, conditions on the day, findings, limitations. Sensors stated by resolution class: radiometric thermal at 640×512, visible at 48 MP.

The delivery manifest. What lands in the folder and in what format. Thermal and visible orthomosaics as GeoTIFF. Surface model as DSM and LAZ, register as CSV and GeoPackage. Map package as KMZ, report as PDF.

A last page states what the set is. Every anomaly ships flagged for a ground check. Models from our capture are photogrammetric. No output is a boundary or a benchmark. Legal elevation and boundary work is done by a licensed Idaho professional land surveyor. Every flight is flown under written authorization. And we don’t use generative tools to produce or alter findings.

Two things about the register worth saying plainly. The severity ranks are a project operating spec, developed for that site and logged with the flight. They are not a published standard, because none exists for canal work. And the archive is kept, so a later flight can be registered against this one and read as change.

If a reach like this is what you’re looking at, the canal seepage service page has scope and timing.

If you want to talk about a reach

A short call covers three things. Which reach and how long it is. When the window works for your season and your operations. And what your own people will need to do on the ground after the flight. That part doesn’t go away.

We’re owner-operated. The person who plans the flight is the person who flies it and signs the report.

If a reach is a bad fit for thermal, we’ll tell you on the call rather than after the invoice.