Remote Sensing Labs
Browse all docs

Optical Water Quality

Light that carries news of a lake’s contents has been inside the lake. It crossed the surface, traveled some way down, was absorbed and scattered by whatever is dissolved or suspended along the path, and a small remainder found its way back out. Water color is that remainder — measured as water-leaving radiance, or as remote-sensing reflectance once normalized by the incoming light — and everything anyone wants to know about the contents of the water has to be recovered from it.

That remainder is also a small part of what the sensor records. Water is among the darkest things the optical bands ever look at — the absorption responsible is part of the water signature in Spectral Bands — so the light carrying news of its contents arrives buried under the air in between, glare off the surface skin, and the glow of nearby land. The quantity of interest is a minority of the number recorded.

What is in the water, and what each one does to the color

Three kinds of thing change the color of natural water, by different optical means that cannot be collapsed into one quantity.

Phytoplankton pigment — chlorophyll is the figure usually quoted — acts mainly by absorbing. It takes blue and red out of the light passing through and leaves green comparatively untouched, so as a bloom builds the water slides from blue toward green. In dense blooms the cells’ scattering starts to show too: just beyond the red absorption, near 700 nanometers, a reflectance peak appears that inland-water methods lean on, and a bloom floating as a surface scum can look more like vegetation than water.

Suspended mineral sediment does the reverse. Clay and silt scatter far more light than they absorb, so sediment-laden water is emphatically bright and warms toward the red as the load rises. It also lifts the return at the longer wavelengths where clean water is dependably black — which matters below.

Colored dissolved organic matter — the tea-brown stain leached from soils and wetlands — scatters almost nothing. It absorbs steeply in the blue and less toward the red, darkening that end without brightening anything, so it strips blue by a wholly different route from pigment. Any reading that treats missing blue as algae will call a peat-stained lake a bloom.

Turbidity stands apart, being not a substance in the water but a property of it: how strongly the water scatters light — related to, but not the same as, how far light gets, which absorption also limits. Sediment usually dominates it, but a dense bloom raises it too. It is the quantity nearest to what an optical sensor responds to, and the least specific about cause.

Color is therefore not a one-to-one code for content. In the open ocean pigment tends to dominate and the others roughly track it — waters often called Case 1 — so one relationship carries a long way. In lakes, estuaries, and river plumes — where nearly all the practical questions live — the three vary independently: a plume can be silty and nearly sterile, a stained lake dark and clear. A relationship fitted where they move together does not transfer, and does not announce that it has stopped working.

What else can make water look that color

Before any of that applies, the signal must survive four impostors, each capable of outweighing the thing measured.

The first is sunglint, specular reflection of the sun off the wave-roughened surface: light that never entered the water and carries nothing about it. Where the geometry lines up it swamps the pixel, and since it follows sun angle, view angle, and wind roughness, it wanders rather than sitting still. A shallow bottom is the second: where light reaches the bed and comes back up, part of the brightness belongs to sand, weed, or rock rather than the column above, and read as sediment it becomes a plume that is not there.

The atmosphere is the third and the biggest: accounting for the air between sensor and ground is a moderate adjustment over bright land and the dominant operation over dark water. The common shortcut assumes the water returns nothing at the longer wavelengths and charges all of it to the air — which holds until sediment makes those wavelengths genuinely bright, so it gives way precisely in the turbid water it was needed for. The fourth is the land next door: light off bright terrain, scattered sideways by the air into a dark water pixel, so a narrow river carries a share of its own banks. Cloud and its shadow come before all four. Cloud Masking handles the screening; the shadow is the part that gets through, because it darkens a pixel without brightening it, and dark is already what water looks like.

Why the honest answer is usually a pattern

Even a clean signal reports only a slab. An instrument sees down as far as light penetrates and returns from — centimeters in a turbid estuary, meters in a clear lake — never the full column, and never a fixed depth, since the constituents measured set how far the view reaches. A bloom settled well below it barely registers.

Water bodies are also awkwardly shaped for a grid. A river or a lake margin falls partly across its own bank, and the bank is bright, so the contamination runs in the direction that reads as sediment: a mixed pixel here is not merely imprecise, it is biased. Pixel size itself is the subject of Resolution. Motion compounds it. Wind can push a surface bloom into one corner of a lake in an afternoon, so two images days apart may show the same water in two arrangements rather than more or less of anything — what a sequence of them can be read to mean is taken up under Time Series.

What this imagery answers well, then, are questions of shape and direction: how far a plume reaches, which arm of a lake is worst, whether this August is greener than the ten before it. What it usually answers less well is what a regulatory threshold asks, which wants a number with a unit attached. Knowing which of the two is being asked matters more than any refinement of the retrieval. And a ramp will render a shaky one with the authority of a solid one — Visualization sets out what a display does not vouch for.

What makes a water-quality product worth believing

Monitoring programs act on these numbers; Environmental Monitoring covers who is asking and what for. Because so much of the raw signal is taken away before interpretation starts, the opening question is how much, and by what reasoning. A product that states its corrections, and says how the outcome was checked against something outside the model that generated it, is making a different order of claim from one that publishes a concentration.

Three questions do most of the rest. Was the method built on water like this one — the same mixture of pigment, silt, and stain — or imported from a setting where those ingredients keep different company? What was it compared against — how many field samples, how near in time and place, given that a set gathered on calm bright days certifies only calm bright days? And does it mark where it declines to answer, over glint, shallow beds, or close to a shoreline, instead of returning a number everywhere it has data?

One more question sits underneath: was the instrument built for this? The broad bands of general-purpose land missions were placed mainly to separate land covers, though recent ones added bands that help over water. Landsat and similar land missions resolve small lakes and rivers that dedicated ocean-color sensors are far too coarse to see, and can support sediment and turbidity estimates, but their few broad bands and modest signal over dark water leave pigment concentrations harder to pin down than an instrument designed for water color would.

Water color was never a direct reading of water condition. It is an inference from a faint remainder, made against assumptions the image cannot check. Held that way it supports a great deal; mistaken for a measurement of the water, it misleads where the answer matters most.

Sources