LandTrendr change
Per pixel trajectory segmentation of annual Landsat NBR. Draw a period, pick an area, and the burn scar of the 2021 Bootleg Fire separates from the forest around it. One recipe id, two places to run it: here in the browser, and on the fleet.
Where these boxes come from
contrastAuthored framing box, centred on the hero AOI with roughly 20 km of margin. Not a measured fire perimeter.hero-lockdemos/cng-nyc-2026/hero-lock/MINT.md, --set bbox.
-121.9, 42.4, -120.8, 43.05
period 2013-06-01 to 2026-09-30
The fleet run replays the LOCKED bbox (-121.55, 42.6, -121.2, 42.85), not the framing box on screen. A lock is what makes a plan priceable, and none exists for other bindings: minting one resolves a catalog, which is the engine's job rather than a browser's.
View the recipe
Everything the render above was given, and every place it differs from the recipe as authored. Filled in by the runner once resolution completes.
Bound parameters
Overridden from the recipe defaults
Interim source substitutions
Resolved spec
Waiting for the recipe to resolve.As authored
The band on screen is band 4, peak_to_trough
LandTrendr emits four bands. The one this recipe styles and thresholds is
peak_to_trough, the signed depth of the disturbance, on a one sided ramp:
peak to trough cannot be positive, so half of a diverging ramp would be spent on values
that cannot occur.
Measured on this recipe over the MTBS Bootleg perimeter through the release engine, one
run, threshold -0.20: band 4 scores recall 0.7772, precision 0.9474, kappa 0.6836 and
clears every floor. Band 1 magnitude scores 0.5292, 0.9631, 0.4502 and fails
recall and kappa. Band 1 is a signed whole period trend, which is a different question.
Those scores are the engine's, not this browser render's. They are quoted here because
this is the recipe they were measured on, and they say which band to read. Source:
registry/recipes/landtrendr-change/folia.yaml outputs.change.