Skip to main content

Demand during an eclipse

A measurement drawn over the model that predicts it — and an annotation layer saying why they disagree. When the two curves track for thirty hours and split for ninety minutes, the chart's job is to make the split unmissable and name its cause.

The chart​

On 12 August 2026 a total solar eclipse crossed northern Spain in the last hours of daylight. This is what the country's power demand did: the solid line is what peninsular Spain actually drew, the muted one is what Red Eléctrica's demand model said it would draw. Hover through the shaded region — at 20:40 CEST, eight minutes after the umbra left the peninsula, demand ran 2.13 GW under forecast, 5.8% of the number the model had written down.

The data​

Real measured data — Red Eléctrica's Seguimiento de la demanda de energía eléctrica (demanda.ree.es), peninsular demand at the operator's native 5-minute cadence, retrieved the morning after the eclipse. Source: Red Eléctrica. 361 samples, 21:00 CEST on 11 August to 03:00 CEST on the 13th, converted MW → GW in the fixture:

ColumnMeaning
demandMeasured peninsular demand, GW — 24.90 min, 40.23 max
forecastREE's own demand forecast, GW

Outside the eclipse evening (taking 19:00–22:00 CEST as the window) the model is good: the gap between the two columns averages +63 MW with a standard deviation of 305 MW over the other twenty-seven hours. Inside it, the gap is a wall: every sample from 19:30 through 21:10 — twenty-one of them — runs more than a gigawatt under forecast. The first sample to break −1 GW is 19:30, the same minute the Moon's disc first touched the Sun over western Spain (Madrid's first contact: 19:31). The deepest gap lands at 20:40, eight minutes after the umbra left the peninsula; it is back under half a gigawatt three minutes after last contact (21:25), and gone entirely by 21:55.

The eclipse times on the chart are Madrid's, from Spain's national observatory (IGN): partial phases 19:31–21:22 CEST, the umbra crossing the north of the country 20:27–20:32. They are astronomy, not something derivable from the demand columns — which is exactly why they're drawn as annotations rather than data.

The grid operator had spent weeks preparing for the generation side of the event — an expected loss of around 5 GW of solar as the shadow passed (the measured version is the eclipse-solar card, this card's companion). What this export shows is the quieter story on the demand side: a country stepping away from its appliances to watch, and stepping back afterwards. The time axis renders in your browser's timezone; clock times here are CEST.

Build it​

1. A measurement over its model. Two lines from one series, and one prop carrying the semantics: the forecast is drawn as="muted" — the theme's hairline neutral, not a data hue — so the measured curve is unmistakably the subject. Model first, measurement second: layers paint in JSX order, and the subject goes on top.

const series = eclipseDemand();

<ChartContainer range={eclipseDemandRange()} width={width}>
  <ChartRow height={220}>
    <YAxis id="gw" label="GW" format=".1f" width={48} />
    <Layers>
      <LineChart series={series} column="forecast" axis="gw" as="muted" />
      <LineChart series={series} column="demand" axis="gw" />
    </Layers>
  </ChartRow>
</ChartContainer>;

…which gets you the divergence but leaves the reader to guess its cause.

2. Name the cause with annotations. A <Region> spanning the partial phases and a <Marker> at totality. These are the annotation register — they never share a hue with the data register, so a placed mark can't be mistaken for a series:

<Region
  from={ECLIPSE_MARKS.partialsBegin}
  to={ECLIPSE_MARKS.partialsEnd}
  label="eclipse"
/>
<Marker at={ECLIPSE_MARKS.totality} label="totality" />

3. Choose the window honestly. The chart at the top of this page shows eclipse day as a civil day — 03:00 to 03:00 CEST — so the dip sits inside the whole diurnal shape: the overnight trough, the 40.23 GW afternoon peak, and a forecast that tracks all of it until 19:30. On the gallery card a 7-hour window sweeps the record instead, because at card width the full day squeezes the story into a sliver. Pass full to open all thirty hours:

<GalleryEclipseDemand width={width} full />

Options to try​

OptionWhat it doesReach for it when
as="muted"Draws a curve in the theme's neutral, off the data rampAny reference curve — a model, a baseline run, yesterday
A second <Region>A narrower band inside the firstMarking totality's 20:27–20:32 span rather than a single instant
selectable={false}Makes a mark inert background contextThe annotation is context, not something a click should select
cursor="crosshair"Reticle + pill snapped to the nearer curveReading the gap sample by sample; nudge across it to flip between the two
rangeWhich slice of the record is on screenThe zoom-vs-context tradeoff in step 3 — there is no right answer, only an honest one
curve="monotone"Smooths both linesCoarser exports (REE also publishes 10-min and hourly views)

See also​