There are exactly three ways to find out how much water is in a hole: trace its shape, dissolve something known into it and read the change, or fill it and measure what went in. Each one answers a slightly different question, each one fails in its own particular way, and when two of them disagree, that disagreement is usually telling you something rather than just being wrong.
This is what the tracer on this site does. Trace the rim and every shelf in a photo, and the tool runs the shoelace formula on each traced layer to get its area directly from the polygon, not from a bounding rectangle. Then it sums the volume between layers as a stack of prisms and prismatoids, the same shapes you would use by hand if you cut the pond into slabs at each depth step and added them up. The result is a hole shape, not just a number, which is also why it is the only one of these three methods that hands you a liner size and a surface area for free.
Its errors come from three places. The two tape measurements you type in set the pixel-to-inch scale for the whole photo, and that scale enters the final volume roughly squared, because it applies to both the length and the width of every traced shape. Ten percent off on a tape span is closer to twenty percent off on the final gallons, which is a worse trade than it sounds like it should be. A steeply angled photo has a second problem a two-tape-span correction cannot fully undo: things near the camera read larger than things far from it, and that near-far gradient is honestly unsolved here, not quietly ignored. The straighter down the photo, the smaller that leftover error. And depth itself is a handful of point samples, typed in at the rim and at each shelf; anything the floor does between those points, a dip, a rise, a soft slope that was not marked as its own layer, the tool never sees.
Trust geometry as the default. It is free, it works at any size, and between the four combinations of the three methods it is the only one that also describes the shape you actually dug. Read the photography guide before you shoot, because most of geometry's error budget is spent well before you ever touch the tracer.
Add a known weight of plain salt, give it a day to spread through the whole pond, and measure how much the concentration rose. The salt calculator runs the arithmetic both ways: dose to a target, or invert a measured rise back into a volume with gallons = 120 × pounds added ÷ (ppt after − ppt before). Its real virtue is what it is blind to. Salt dissolves into whatever water is actually there, so it reports the volume of the water itself, with no idea whether that water is sitting in an empty basin or wrapped around a foot of river rock. Geometry cannot make that distinction; salt dilution cannot see anything else.
Its error sources are different from geometry's. A digital meter reads to about ±0.05 ppt, so a before-and-after pair carries roughly ±0.1 ppt of combined noise, and that noise is fixed in absolute terms no matter how big the pond is. The rise you are measuring, though, is not fixed: a cautious tester on a pond that might be small keeps the test dose light, to avoid accidentally pushing a pond well over a comfortable concentration, and a light dose produces a small rise. Add half a pound of salt to a pond that turns out to hold 150 gallons and the reading climbs by roughly 0.4 ppt; the same ±0.1 ppt of meter noise on that smaller rise swings the volume estimate from about 120 to 200 gallons, a band wide enough to argue with itself. The same meter on the calculator's own example, a ten-pound dose producing a 0.8 ppt rise, comes in tighter, at roughly ±13 percent. The smaller the achievable rise, the more the meter's fixed noise dominates the result, which is exactly why a cautious test dose on a small pond is also the least accurate one. Incomplete mixing pushes the same way: read too soon, before the salt has reached behind the rocks and under the shelves, and the reading comes in low, which makes the pond look bigger than it is. And any salt already in the water before you started, from a previous dose or a moderately hard tap fill, has to be part of the "before" reading or it quietly inflates the "after" one.
Trust salt dilution as a cross-check against a trace, or as the fallback when there is no usable overhead photo but you can safely add a modest, measured dose. It is worse than geometry at telling you the pond's shape, because it cannot tell you a shape at all.
Fill the pond from empty through a metered hose bib, or a garden hose timed against a bucket of known volume, and read the total off the meter when it is full. Or run it the other way: drain the whole pond into containers of known size and add them up. Either way there is no model standing between the measurement and the answer, no scale factor, no concentration curve, nothing to be wrong about except the metering itself. That is exactly why it is the reference the other two methods are judged against. Geometry should match it when there is no rock or gravel taking up room; salt dilution should match it always, rock or no rock, because both of them are measuring the same water displacement measures directly.
It is also almost never practical once a pond gets past a stock tank or a small preformed shell. Filling a few hundred gallons through a garden hose takes hours, and draining one into containers you can actually measure means finding, and emptying, a few hundred gallons' worth of containers. Above a few hundred gallons this stops being a method anyone actually runs and becomes the standard the other two are compared to instead.
When a traced volume and a salt-derived volume exist for the same pond, they will not land on the same number, and that is not a failure of either method. Geometry measures the hole; salt measures the water sitting in it. Subtract one from the other and what is left over is an estimate of whatever is displacing water without being water: rock, gravel, a folded liner, a submerged planting basket. Say a traced outline comes out to 400 gallons and a salt test on the same pond reads 350. The 50-gallon gap is not an error bar, it is a rough measurement of your hardscape, taken without draining anything. The salt calculator computes exactly this gap for you the moment both numbers are on the page.
That only holds while the gap stays plausible. The calculator treats a disagreement of more than roughly 200 percent as too large for rock and gravel to explain, and flags it instead of quietly labeling it displacement. Past that point the more likely explanation is a bad input on one side or the other: a tape span measured along the wrong edge, a camera height typed in wrong, a salt dose or a before-and-after reading that got mixed up. Recheck the inputs before trusting either number.
Start with geometry. Trace the pond from a decent overhead photo and you get a volume, a surface area and a liner size in one pass, all from the shape you actually dug. Run a salt test alongside it when you suspect real rock or gravel underneath, or when the photo was the best you could manage and you want a second, independently-derived number to check it against. Reach for displacement only on something small enough to fill from a bucket count; above that size it is the number everything else is trying to approximate, not a method you run yourself.