Which way do solar panels in Pakistan actually face?
Fitted tilt and azimuth for 313 installations whose
historical Sentinel-2 specular reflections agree on a single fixed panel orientation --
out of 2401 installations checked, stratified by size, across
two years of imagery (a 4x-larger, chunked-tile-batch re-run of the original 500-target study, plus a 2026-08-14 targeted random top-up of 401 more targets concentrated in the latitude bands/provinces a regional re-cut found too thin to trust).
This is not a neutral sample of installed pose -- read the envelope note before trusting the shape.
Tilt (radius) x Azimuth (angle)313 measured
rooftop generator ground plant unreachable by this sensor's fixed overpass time
20.6%
of ≥ 1,000 m² installations get a pose in this plot at all -- the rest never produce two self-consistent spikes, so there's nothing to fit
≥1 spike, no fittable pose†21.3%
no glint signal at all58.0%
azimuth range observed81.7° – 180.1°
tilt IQR observed6.6° – 18.8°
figures restricted to the ≥ 1,000 m² target class (n=1051 of 2401)
Why the plot only fills part of the circle
Sentinel-2 crosses this latitude at a fixed ~10:30 local time, so across two full
years of imagery the sun's azimuth at the moment of every overpass never once swings
west of due south -- a panel facing southwest, west, or north cannot glint into this
sensor no matter how long you wait. That's a sensor limit, not a property of
rooftops in Pakistan: the shaded wedge
(262° wide) marks orientations this survey
cannot observe at all, not orientations known to be empty. Nothing is plotted there,
because nothing was measured there.
† A single spike can't be checked for self-consistency -- 21.3% of installations glint
once (or on dates that disagree on a panel geometry) and so are correctly
detected as "PV is probably here" but carry no orientation information.
Tilt distribution
Azimuth distribution (180° = geometric ceiling)
For context -- validation rate by installation size (all 2401, not just the fitted 313)