Mostar from orbit
Every summer since 1984 has been photographed from orbit. This is Mostar and the Neretva valley in all of them — one cloud-free composite per year, forty-three frames, every one of them the same 31 × 40 km of ground on the same grid.
The city runs north–south because the karst gives it nowhere else to go. What the resolution is good for is not the Old Bridge — at 30 m that’s about one pixel — but everything harder to see from the ground: which hillsides lost their vegetation and which grew it back, how far the built-up edge crept up the valley sides, and how the two banks changed at different rates and different times.
Each frame is a median of the least-cloudy June–September scenes from every satellite footprint overlapping the valley. Landsat 5 carries 1984 to 2011, Landsat 7 covers 2012 alone — the one year with no alternative, since Landsat 5 had retired and Landsat 8 hadn’t launched — Landsat 8 takes 2013 and 2014, and Sentinel-2 runs from 2015.
Everything is converted to surface reflectance before compositing, so a value means the same physical thing in 1984 as in 2026, and every frame then gets the same fixed stretch. The tempting alternative — histogram-matching each year to a reference — produces a beautifully smooth animation and destroys the subject, because it guarantees that any real change in the distribution of brightness has been normalised away. What does get corrected is brightness alone, anchored on pixels that genuinely shouldn’t change: bare karst, chosen where it reads as rock in at least 85% of the forty-three years. The city is never used to normalise the city.
Two corrections mattered more than expected, and both are the kind that would have quietly lied. Sentinel-2 gained a radiometric offset in processing baseline 04.00 on 25 January 2022; miss it and every recent frame comes out too bright, which reads convincingly as four hazy summers rather than as an arithmetic error. And scene selection has to draw from every footprint the area touches — Mostar straddles a boundary in both the Landsat and Sentinel tiling grids, so simply taking the least-cloudy scenes leaves whole corners of the valley unobserved.
Built from the Microsoft Planetary Computer’s copies of USGS/NASA Landsat Collection 2 and ESA Copernicus Sentinel-2. Contains modified Copernicus Sentinel data.