SARtale Guide

When clouds close in, the lakes stay in view.

SARtale watches glacial lakes in Gilgit-Baltistan and upper Khyber Pakhtunkhwa with NASA–ISRO NISAR radar, which sees through cloud, day or night. Each lake gets one plain status, a chart of every pass and a printable brief.

This guide is for anyone who wants to use the website, run it, rebuild its data or work on the code.

NISAR HH radar picture of Attabad Lake, 23 October 2026
Attabad Lake, NISAR HH, 23 Oct 2026, from an earlier build (before the lake window was moved on 10 Nov 2026). Blue: water read from radar. The current picture and area are in the app.

Watch the videos

Demo, 30 seconds, 29 s. The Hunza flood, cloud, NISAR radar through it, and SARtale: dark pixels to water, pass after pass, flags only beyond a lake's own noise.
Demo, 60 seconds, 59 s. The Shisper lake outburst and SUPARCO's cloud gap, then the app in use: Lake Watch, Passu's water and history, its honest flag, and requesting a lake.
How it works, 3 min 29 s. How radar sees through cloud, how terrain shadow is left out, how dark pixels become water, how a change is tested against a lake's own track and noise, frozen lakes, honest flags and lake requests.

Captions are on by default. On the website the videos play from media/; opened from a clone of the repository, they play from the repository's media/ folder, without captions.

Why it exists#

Over 3,000 glacial lakes have developed in Gilgit-Baltistan and Khyber Pakhtunkhwa (UNDP GLOF-II). SUPARCO tracks 131 of them as potentially dangerous, mostly with optical satellites. The lakes fill in summer, when cloud hides them. SUPARCO's 2 July 2026 update said persistent cloud after 16 June had made lake changes difficult to assess. From 17 June to 4 August 2026, NISAR kept looking at the watched lakes through the cloud; the site gives the count of valid looks from the current data.

How it fits together#

Workflow: six pipeline steps from finding NISAR passes to deciding a status, which write lakes.json for the web app. A lower loop shows a lake request: the form on the site opens a pre-filled GitHub issue, a workflow comments with the NISAR pass count, a team member adds the accepted label, the data are rebuilt and committed to main, and Pages redeploys. A weekly refresh feeds the same rebuild.
The pipeline turns NISAR passes into lakes.json; the app reads it. A new lake comes in through a visitor's GitHub issue and the team's review, and a weekly refresh feeds the same rebuild (Lake requests).

What is in the repository#

FolderWhat it holds
app/The website: TypeScript, Vite, MapLibre and uPlot. Page text is in app/src/content/; the data it shows is in app/public/data/.
processing/The Python pipeline and its tests. The lake register is processing/lakes_in.json.
media/The 30 s and 60 s demos, the how-it-works video, screenshots and the workflow figure.
docs/This guide, the data sources and licences, and the AI tool use statement.

Team#

Team SARcastic Scientists: Saeed Ahmad, Shaharyar Sajid and Usama Hussain. NASA Space Apps Challenge 2026, "Dancing with the SARs", Islamabad local event.

Indicative screening, not a warning service. SARtale uses NASA's provisional NISAR product, which is not validated. Official warnings come from PMD, NDMA, GBDMA and PDMA.