Key takeaways
- There are two separate problems here. Snow sits on top of the ground, which biases a surface upward. Frost moves the ground itself, which puts a seasonal offset into elevations. The first is a measurement bias and the second is a comparability problem.
- A photogrammetric surface records the top of whatever reflected light back to the camera. Over snow that is the snow surface, so every elevation and every volume computed from it is high by the local snow depth.
- Snow depth is not a constant you can subtract. Wind strips exposed crowns and loads the lee side, traffic compacts haul routes, and a pile toe collects drift, so the bias varies across the same block by more than the survey's own uncertainty.
- Frost heave lifts frost-susceptible ground by a seasonal amount that is unevenly distributed. Compare a February surface against a July surface and part of the difference is the season rather than anything anyone moved.
- Ground control placed on heaving ground inherits the same movement. Monuments set below frost depth are what make a multi-year series hold its vertical frame through a winter.
- Winter against winter is the comparison that holds. A winter capture compared to a summer one has a seasonal floor under its differences that has to be stated rather than reported as movement.
- The schedule answer is to take the annual baseline on bare ground before the first lasting snow. Where a winter capture is unavoidable, record the snow and ground condition with it so the dataset can be read correctly later.
A winter drone survey gets discussed as an operational question. Will the batteries hold up, is there enough daylight, can the crew work safely at that temperature. Those are real and the cold weather envelope and the short December daylight window are both worth planning around.
The question that costs more money is quieter: what does frozen and snow-covered ground do to the numbers that come out the other end. The honest answer is that it does two different things, they act in different directions, and only one of them is visible in the imagery.
Two problems that get confused for one
The first problem is that snow is on top of the ground. The surface a photogrammetric capture produces is the top of whatever reflected light back to the camera, so over snow it is the snow surface. Elevations are high and volumes above a base surface are large, by the local snow depth. This is a bias in a known direction.
The second problem is that the ground has moved. Frost penetrating frost-susceptible soil draws water to the freezing front, ice lenses form, and the ground above them lifts. The amount depends on soil texture, moisture availability and how deep the frost got, which means it varies across one site. This is not a bias in a known direction so much as a seasonal pattern laid over the real topography.
A site owner who reads a winter dataset as though only the first problem existed will subtract an average snow depth, feel they have dealt with it, and still be looking at elevations that contain a frost component.
Snow depth, and why an average does not rescue it
The temptation is obvious. Probe the snow in a few places, take the mean, drop the surface by that much. It does not work on an Alberta site for a straightforward reason: wind.
- Exposed ground gets stripped. A pile crown, a pad edge or an open field rise can be close to bare while the ground twenty metres downwind holds a deep drift.
- Depressions and toes collect. The toe of a stockpile, a ditch, a borrow pit floor and the lee of every structure are where the stripped snow ends up. These are frequently the exact features a volume or a drainage decision turns on.
- Traffic compacts. A haul road or a loader track is packed to a fraction of the depth of the untracked ground beside it, and packed snow is denser as well as thinner.
- It changes between flights. One storm or one windy night redistributes it, so a depth survey taken on Tuesday does not describe Thursday's capture.
Across one working block the spread in snow depth is routinely larger than the vertical uncertainty the survey itself was specified to, which means the snow is the dominant error term rather than a correction to a small one. There is more detail on the texture and contrast side of this in the piece on surveying stockpiles and sites under snow, where fresh snow also starves photogrammetry of the detail it needs to match images at all.
On a pile specifically the geometry conspires against you. Snow thins on the crown where the wind works and thickens at the toe where it collects, so the snow envelope is not a uniform shell over the pile. It changes the pile's apparent shape, not just its apparent height, and a volume is a shape calculation.
Frost heave is a datum problem wearing a topography costume
Heave needs three things together: frost-susceptible soil, available water and frost penetration. Silts and silty clays with a water source are the classic combination, and Alberta has plenty of both. Clean granular material with good drainage heaves very little, which is one reason a properly built pad behaves better than the native ground beside it.
What this looks like in a dataset:
- A smooth regional rise. Heave on uniform ground produces a gentle lift across an area, which reads as though the whole area gained a few centimetres of fill.
- Differential movement at boundaries. The edge of a pad, the shoulder of a road and the transition from cut to fill heave by different amounts, which reads as settlement or construction at a line where nothing happened.
- Structures that did not move. A foundation on piles below frost depth stays put while the ground around it lifts, so the ground appears to have gained relative to the structure. Come spring the relationship reverses.
None of that is a survey error. The survey measured the ground correctly. The ground was in a different place.
Control on ground that is itself moving
This is where a winter capture quietly breaks a measurement series. A ground control target set on heaving native soil moves with that soil. If the control moved and the ground moved by a similar amount, the surface can look internally consistent while sitting at the wrong elevation relative to last summer. If the control moved by a different amount than the rest of the block, the error is distributed across the model in a way nobody can see afterwards.
The fix is not a winter technique, it is a decision made once at the start of a monitoring programme: permanent ground control monuments set below frost depth, a named horizontal and vertical frame, and a base coordinate that does not get re-derived every visit. Pins driven into frozen ground for a single flight are fine for that flight's internal geometry and worth nothing to the series. The coordinate system and datum have to be stated either way.
The practical version for checkpoints: put them on things that do not heave. Concrete, a pile cap, structural steel, the lid of a manhole, an established paved surface. Checkpoints on frozen native soil measure the soil's seasonal position, not the survey's accuracy.
Stockpiles with a frozen crust
Three winter-specific things happen to a pile measurement, and only one of them is about snow.
- The crust. A frozen shell on the surface of a pile is part of the measured volume and is also not saleable material in the state the buyer wants it. The volume is real, the usable tonnage behind it is a separate conversation.
- The base surface. The floor under the pile is frozen and may itself have heaved, and it is under the material so nothing in the capture can see it. If the base surface came from a summer capture of the empty pad, that mismatch goes straight into the volume. The base surface choice matters more in winter than it does in July.
- Density. Frozen material, ice in the voids and packed snow in the toe all change the relationship between volume and mass, which is the relationship a reconciliation against scale house tonnage depends on. A winter volume converted with a summer density factor will not balance, and the capture is not the thing that is wrong.
This is also a common and entirely avoidable source of the argument where two surveys of the same pile disagree. One was flown in November and one in May, both are correct, and nobody wrote down the conditions.
When a winter dataset is actually comparable
A winter capture is a perfectly good measurement of a site in winter. The question is what it can be compared against. In rough order of how much weight the comparison carries:
- Same winter, same conditions, days apart. Strong. Snow and frost are close to unchanged, so a short-interval difference is movement. This is the right way to measure a load-out or a push over a week in January.
- Winter against the same month last winter. Usable, with the caveat that snow depth and frost depth differ between years. Trend over several years is more reliable than any one annual difference.
- Bare frozen ground against bare summer ground. Reasonable for horizontal extent and for anything above the snow line, with frost heave still in the vertical.
- Snow-covered winter against summer. Weak for elevations and volumes. Report it as a difference that includes an unresolved snow and frost component, or do not report a number at all.
Which is the argument for scheduling rather than for technique. If a site is on an annual monitoring cadence, the baseline flight belongs on bare ground, after harvest or after the surface is clear and before the first lasting snow. That window is short in central and northern Alberta and it is the single cheapest decision available on the whole programme.
What has to be written down
If a winter capture happens, and plenty do because the material moved in winter and somebody has to measure it, the dataset is only as useful as its record. At minimum:
- Date, time and air temperature at capture.
- Ground condition in plain words: bare and frozen, light cover, drifted, packed by traffic.
- Measured snow depths with their locations, including crown, mid-slope, toe and any traffic route, rather than one representative number.
- Whether the ground control is on monuments below frost depth or on seasonal targets.
- The horizontal and vertical frame, the geoid model and the checkpoint residuals, with a note on what the checkpoints were sitting on.
- A statement of what was not measured: the base surface under material, anything under snow, anything under water or ice.
That record is what lets somebody in two years read the number correctly instead of treating a seasonal offset as a quantity. It is also the part that is free at the time of capture and impossible to reconstruct afterwards.
