SARtale Guide

Questions and limits

The questions people ask most, with straight answers.

Is SARtale a flood warning system?#

No. It is indicative screening on provisional data. It shows whether a lake's water area has changed against its own noise. Official warnings come from PMD, NDMA, GBDMA and PDMA.

Why these lakes and not all 131 that SUPARCO tracks?#

Each lake needs a coordinate with a published source and enough NISAR passes on one track. The first ones were chosen for their sources and their history, with Attabad as a check on the method. A location check against 2026 imagery then removed points that were not lakes, and marked sites whose lake has drained as No lake now. The method is the same for any lake, and anyone can put one forward with the "Add a lake" form on the story page. The team adds it once its source checks out.

Can radar really see through cloud?#

Yes. NISAR's L-band microwaves pass through cloud and need no sunlight, so it sees the lakes day or night. From 17 June to 4 August 2026, while cloud limited optical images, NISAR kept looking at the watched lakes; How we know and the guided story give the count of valid looks from the current data.

Do mountain shadows hide lakes?#

They can, for both kinds of satellite. Radar looks sideways, so a slope facing away from it falls in radar shadow and looks as dark as water, and a slope facing it folds over (layover). Both fall on the valley walls, almost never on a flat lake, so SARtale masks nothing inside a lake's zone and leaves the walls, mapped from the Copernicus terrain model, out of the window median that water is measured against. (An earlier rule set aside anything dark from only one viewing direction; it removed 15–42% of the real water and is no longer used.) If less than half of a lake's zone can be seen, it reads Not measured ("Radar cannot see enough of the lake on this track") instead of a guess.

Why do radar and Sentinel-2 disagree so much?#

How we know gives the overlap (IoU) at each lake checked, from the current data. Radar counts wet shore and sediment flats as water and is limited by speckle at small lakes; optical water indices miss shaded or very dark lakes and flag wet ground, radar misses wind-roughened water, and even on the same day the two satellites pass at different hours. Neither is a clean reference, which is why the app calls its status indicative.

How current is the data?#

As current as the last pipeline run, which the weekly refresh repeats each Monday. The story page and every lake show the date of their latest pass.

What does "provisional" mean?#

Passes up to 21 Oct 2026 are NASA's provisional release (CRID P05023), put out before validation. NASA will reprocess it, and numbers may change when it does. Later passes are NASA's standard GCOV product (CRID P05036); we have no statement that it is validated. The site keeps the "Provisional NISAR data" label while any provisional pass is used.

Why does a lake say its track mixes processing versions?#

Its status compares passes from the provisional release with passes from the standard product. A calibration change between the two can shift an area by itself, so a change across 21 Oct 2026 deserves extra care.

Known limits#

  • A level is about radar water area only, not volume, depth or hazard.
  • A lake drawing down onto wet sediment is nearly invisible to the radar (a tenth to a fifth of the real decrease).
  • An Alert comes 5–12 days after a change begins in Hunza and 8–12 days in Chitral: slow-growth monitoring, not outburst warning.
  • False-alarm and miss rates are modelled from one season, not observed; they will be estimated again after one more season. Lakes below about 0.05 km² are limited by speckle.
  • Zones are approximate: an OpenStreetMap outline where a lake has one, a search circle elsewhere.
  • Frozen lakes, lakes under 0.05 km² and wind-roughened water are not validated.
  • Lake requests need a GitHub sign-in; anyone signed in can open one (see Lake requests).