@skyglowberlin@fediscience.org
Post #2086639
2026-04-08 15:22 UTC
For each pixel, the lead author Tian Li from the University of Connecticut fits a trend line (solid line left) and generates a prediction (dotted line left).
When something changes, the observations stop agreeing with the prediction (middle). If the change persists for at least 14 consecutive observations, then a break point in the trend is assigned. From that point on, a new model predicts future changes. And so on and so forth (right) for the following years.
Replies (1)
-
@skyglowberlin@fediscience.org 2026-04-08 15:32
This is where looking at the daily data (rather than monthly composites) becomes a REALLY BIG DEAL. Nighttime lights data are really noisy. Part of the noise comes about because the satellite can view a scene from different directions. When that happens, it's viewing LITERALLY DIFFERENT LIGHTS, as in the aerial photos here.