Solar noon sweeps west at 15° of longitude per hour. The day side gains energy; the whole planet loses it to space all the time.
The air holds about 1% of the surplus. It is the thin, fast-moving part of the system; the ocean is the reservoir. (IPCC AR6, 1971–2020.)
Same latitude can mean very different budgets. Compare a coast with an interior, or a desert with a forest at the same hour.
Energy density here means the heat content of the whole air column above one square metre: the thermal part (mass × heat capacity × temperature) plus the latent part (water vapour × heat of vaporisation). It is the atmospheric analogue of temperature or pressure maps, but it is a stock, in joules, rather than a state. Temperature and vapour come from an embedded reanalysis climatology; height from an embedded relief grid.
One kelvin of warming adds about 10 MJ/m² of thermal energy everywhere, and up to 8 MJ/m² more in the humid tropics because warmer air holds 7% more vapour per kelvin.
The imbalance (energy in minus energy out, the number that drives warming) is anchored to the published record: about 0.4 W/m² in 1980, about 1.0 W/m² now. The map then shows where in and out happen; the global sum is pinned.
Terrain at half a degree. Relief is a real grid (SRTM15+ / ETOPO1 via GMT, 0.5°, 25 m steps). The column above it thins with height. The reanalysis sees terrain at ~2°, so its temperature is adjusted by 6 K per kilometre between its own smoothed relief and the 0.5° grid; that is what gives Tibet and the Andes their shape.
Climatology, first harmonic only. The annual mean and one sinusoid per place for 2 m temperature and precipitable water, from NCEP/NCAR Reanalysis 1 long-term means (1991–2020), at 1°. Currents, continentality, monsoon moisture and plateaus are in it. A sharp monsoon onset, a second harmonic, or anything finer than the reanalysis grid is not.
No weather. No clouds, fronts, or storms. Clouds are folded into a fixed albedo per surface type.
No transport. Winds and currents move roughly 5 PW poleward. The maps show local budgets, not where the energy ends up.
Inferred surface types. Deserts and forests are read off the climatology: how much vapour the air holds relative to what its temperature allows. Sea ice is where the annual mean is well below freezing; ice sheets where high-latitude land is colder still. These weights set albedo and the day–night swing only. The diurnal cycle and the warming pattern (land faster, Arctic ~3×) are still idealised; monthly means carry neither.
Beyond 2026 is a pathway, not a forecast. Warming follows a middle-of-the-road scenario (SSP2-4.5).
Embedded data: land mask, Natural Earth via global-land-mask, 0.5°; relief, GMT earth_relief 30′ (SRTM15+, ETOPO1), 0.5°; climatology, NCEP/NCAR Reanalysis 1 long-term monthly means 1991–2020, 2 m air temperature and precipitable water, reduced to annual mean + first harmonic at 1°. About 210 KB in the page. Everything else is derived.