What atmospheric correction does to a satellite chlorophyll-a number

You pull up two passes of the same reservoir three weeks apart. Nothing happened in the lake that you know of, no bloom report, no upstream discharge, no big rain event. But the chlorophyll-a number jumped by a third. Before you call the lab for a grab sample, it's worth knowing where that jump probably came from, because there's a decent chance it's not the lake at all.
What the sensor is measuring
A satellite sensor doesn't see your lake directly. It sees whatever light makes it back up through roughly 100 km of atmosphere to the instrument. That signal, called top-of-atmosphere reflectance, is a mix of three things: light reflected off the water column (the part you want), light scattered by air molecules and aerosols before it ever reaches the surface, and light scattered back up without touching the water at all.
Over open ocean that last term can account for 80 to 90 percent of the total signal. Over inland water it's usually lower but still substantial, and it varies pass to pass with haze, humidity, dust, and smoke. The part you want, the light that actually bounced off the water and carries the chlorophyll signature, is called water-leaving reflectance. It's a small fraction of what the sensor records.
Atmospheric correction is the processing step that estimates and strips out the atmospheric contribution, leaving an estimate of water-leaving reflectance behind. Every chlorophyll-a, turbidity, or clarity algorithm downstream of that assumes the atmospheric term has already been removed. Run an algorithm on raw top-of-atmosphere data and you're measuring the lake plus whatever was in the air that day.
Why readings drift between passes on the same lake
Analysts tracking the same lake week over week run into this pattern constantly: readings swing even when nothing changed in the water. Three things move the atmospheric term independently of algae biomass.
Aerosol loading changes day to day. Wildfire smoke drifting through from a few states over, regional haze on a humid summer afternoon, even dust events, all add scattered light that a correction algorithm has to account for and sometimes gets wrong.
Sun angle and view geometry shift with the date and the satellite's orbit position. A pass at a slightly different time of year, or a slightly off-nadir view, changes how much atmospheric path the light travels and how much sun glint lands on the water surface.
Thin cirrus and sub-pixel cloud contamination can leak into pixels that look cloud-free in a quick-look image but still carry extra scattered light.
Stack those three together and you get the pattern a lot of analysts have run into: the raw signal wobbles between passes in a way that has nothing to do with the lake's actual trophic state. A few points of noise introduced by the atmosphere on a given day can look exactly like a real bloom signal, especially to anyone downstream who just sees the number go up.
What a calibrated reading buys you
A correctly atmospherically corrected product gives you water-leaving reflectance that's comparable across passes, across seasons, and ideally across sensors. That's the baseline requirement for a chlorophyll-a number you can trend rather than just eyeball. Without it, you're comparing apples to whatever the atmosphere happened to be doing that week.
This matters more once you're trying to cover a whole region instead of one lake you can sanity-check with a boat trip. A single reservoir with a known baseline, you'll catch an atmospheric artifact because you know what normal looks like. Forty lakes you've never sampled by boat, an uncorrected spike reads the same as a real bloom unless the correction has already been done before the number reaches you.
Water Quality Retrieval delivers a weekly, calibrated chlorophyll-a, turbidity, and clarity read per lake, with the atmospheric correction already handled before the number reaches you.
If you're tired of chasing down whether a reading is a bloom or a bad pass, it's worth seeing what a calibrated weekly memo looks like for your region.