One chart, everything on it
Every forecast and observation, mountain and valley. Drag to pan, scroll to zoom, toggle any line. Composites are trimmed-mean averages of the sources beneath them.
What am I looking at?
Four buckets of data on one timeline. To the left of "now": observed reality (obs lines) and what each forecast said about that time N hours in advance — change Historical forecast advance to see different lead times. To the right of "now": each forecast source's latest issuance.
Mountain forecast point: 34.7335, −86.5236 (Monte Sano neighborhood, ~1500 ft). Valley forecast point: 34.7307, −86.5860 (Madison Co Courthouse / Big Spring Park, downtown Huntsville, ~600 ft).
MSWX forecast = the raw mountain forecast adjusted via a 5-year KHSV → MSWX bias lookup (the airport historically tells us the plateau is N degrees cooler/warmer at this hour/wind/sky combo, so we shift the raw model output by N). Bad mountain forecast = the raw uncorrected output for the neighborhood point. Valley forecast = raw output at the courthouse point — no bias correction (it's a normal valley location and the global models do fine there).
Composites are trimmed means: for each timestamp, drop the highest and lowest source values and average the rest. Robust to a single bad model.