How to tell a seasonal lake from a permanent lake in satellite imagery

A playa in the western half of a basin can be bone dry in August and a quarter-mile across by March. Catch it on the wrong pass and you'll log it as a lake. Catch it on the wrong pass six months later and you'll log it as dry ground that got misclassified. Neither call is wrong, exactly. They're both just one frame of something that only makes sense as a sequence.

That's the core problem with telling seasonal water from permanent water off a single satellite image: you're asking a snapshot to answer a question about behavior over time. A lake's permanence isn't a property of its pixels on any given day. It's a property of what those pixels do across a water year, or several.

The single-date problem

Open any wide-swath scene and you'll find three kinds of water bodies sitting side by side: the reservoir that holds a pool year-round regardless of season, the floodplain lake that fills every spring and draws down by late summer, and the ephemeral pan that only shows up after a wet winter or a big storm event. On the image itself, all three can read as identical blue polygons. Spectral signature doesn't carry a timestamp.

Planners who've tried to shortcut this with a single cloud-free scene usually end up over-counting permanent water in a wet year and under-counting it in a dry one. The reservoir behind the irrigation district looks the same as the depression that fills twice a decade. If your basin inventory gets built off one good pass, it inherits whatever that one pass happened to catch.

Build a presence record instead

A longer record fixes this. Pull the same water body across repeated passes over years, and permanence turns into something measurable: how often is this pixel wet versus dry across the archive. A reservoir behind a dam will show water in nearly every pass you've got, drawdown season aside. A floodplain lake will show a clean seasonal pattern tied to snowmelt or monsoon timing. Compare that to a playa, which shows water in a handful of passes scattered across wet years and nothing in between.

This is also where oxbows and farm ponds get sorted out, the ones that are genuinely hard to call from a single image. An oxbow cut off from the main channel might hold water for most of a normal year but go dry in a drought year, which puts it in a different bucket than a lake that's never gone dry in the record you have. A stock pond fed by a spring will show up as stubbornly permanent even though it's small enough to get lost in a one-off scene. None of that shows up until you're looking at the pixel's history, not its single reading.

This is roughly the approach behind a weekly surface water layer stacked against the multi-decade archive: instead of asking whether a reach is wet today, it tracks how a water body's extent behaves across every pass on file, so a planner can see the water-year pattern instead of guessing from one date.

Rules of thumb that hold up in practice

A few working rules, drawn from how basin staff sort these bodies once they have a time series instead of a single frame:

Water present in nearly every observation across multiple years, with only brief gaps for drawdown or maintenance, reads as permanent. A pattern that shows up on a predictable seasonal schedule, same months, most years, reads as seasonal rather than permanent even if it's reliable. Presence that only appears in some years, tied to wet winters or storm years rather than a calendar pattern, counts as intermittent or ephemeral, a meaningfully different category for allocation and habitat purposes than a seasonal lake that fills every spring on schedule.

None of these rules need a precise cutoff to be useful. What they need is enough passes over enough years that the pattern shows itself instead of getting inferred from one good image. A basin-wide inventory built that way holds up when someone asks why a water body got classified the way it did: the record shows the pattern, pass by pass, that produced the call.

If your basin's inventory still leans on scattered single-date imagery to sort permanent from seasonal water, it's worth seeing what the weekly extent layer and long-run archive behind Surface Water Monitoring would show for your reaches instead.