Skip to main content

Solar during an eclipse

The other half of the eclipse-demand story. A solar curve near sunset is already falling, so eclipse day alone just looks like a steep evening — the chart only works against the days that didn't have an eclipse in them. Here the three previous days are the control, and one collapse turns them into a baseline the interesting day can be judged against.

The chart​

Two views, one toggle. Four days is the raw material: three grey evenings — 9, 10 and 11 August, each plotted forward to eclipse day's clock — and the solid one the Moon interrupted. Anomaly vs baseline collapses the grey days into their per-quarter-hour mean and redraws every curve as a distance from it: eclipse day dives to −4.34 GW at 20:15 CEST, while the grey residuals hug zero through the whole evening ramp. The empty vertical space between them is the point: sunset is the most repeatable thing a solar fleet does, and this one broke the pattern by gigawatts.

The data​

Real measured data — Spain's grid-scale solar at 15-minute cadence from energy-charts.info (Fraunhofer ISE), CC BY 4.0, upstream ENTSO-E — the same source family as the German Easter-weekend fixture behind the track's first three cards. Retrieved the morning after the eclipse. Four civil days on one shared index:

ColumnMeaning
eclipseDaySolar generation on 12 August, GW — 29.67 peak
aug09–aug11The three days before, shifted to the same clock time

The shift is the device, so here it is in plain words: the grey curves did not happen when the axis says — each happened one, two or three days earlier. Plotting reference series at a false time is only honest when the page says so, every time. (The Niño 3.4 card does the same thing with 45 years; this is the long-weekend version.)

The baseline is not in the fixture — it's computed in the example, and that's deliberate: it is one collapse (below), and teaching it is this card's job. Judged against it, the days have two personalities. The afternoon is weather: cloud swings individual days up to 3.57 GW from the pack's mean (at 17:00), and eclipse day itself rides above baseline from late morning until 18:30, peaking at +3.53 GW at 17:00 — it was the sunniest of the four (29.67 GW peak against the baseline's 27.22). The evening ramp is physics: from 19:45 onward the three ordinary days sit within ±0.53 GW of their mean, tightening to ±0.21 by 20:30. Into that repeatability the eclipse cuts a −4.34 GW anomaly at 20:15 — at 20:00 the gap is twelve times the ordinary envelope (−3.86 GW against ±0.31) — then fades with the daylight: −0.45 GW by 21:00, gone by 21:30. Red Eléctrica had pre-announced a loss of up to 5 GW.

One more quirk worth reading off the chart: none of the four days ever touches zero. Each enters the night around 0.5–0.65 GW and drains to 0.12–0.22 GW before dawn — ENTSO-E reports concentrated solar plants discharging stored heat after dark inside the same category as PV.

The eclipse marks are the same ECLIPSE_MARKS constants the demand card draws (IGN's Madrid circumstances) — one fixture module, two cards, one set of facts. The time axis renders in your browser's timezone; clock times here are CEST.

Build it​

1. Put all four days in one series. The overlay happens in the fixture: sample i of every column is the same clock time on a different day, so one TimeSeries keyed by eclipse-day timestamps carries the whole comparison. Drawing the raw spaghetti is four <LineChart>s — the reference pack muted, the subject on top:

const days = eclipseSolar(); // eclipseDay, aug09, aug10, aug11

<Layers>
  <LineChart series={days} column="aug09" axis="gw" as="muted" />
  <LineChart series={days} column="aug10" axis="gw" as="muted" />
  <LineChart series={days} column="aug11" axis="gw" as="muted" />
  <LineChart series={days} column="eclipseDay" axis="gw" />
</Layers>;

2. Collapse the pack into a baseline. collapse runs one reducer across named columns, row by row; { append: true } keeps the inputs alongside the result. The baseline is the three ordinary days' mean, and every curve's anomaly is just another collapse against it:

const series = days
  .collapse(
    ['aug09', 'aug10', 'aug11'],
    'baseline',
    (v) => (v.aug09 + v.aug10 + v.aug11) / 3,
    { append: true },
  )
  .collapse(
    ['eclipseDay', 'baseline'],
    'delta',
    (v) => v.eclipseDay - v.baseline,
    { append: true },
  );

The example repeats the delta step for each grey day (aug09Delta, …), so the ordinary days can be drawn as residuals too — they're the envelope that tells you whether the anomaly is signal. (In the real source the mean's inputs come from the exported ECLIPSE_BASELINE_DAYS list rather than being spelled twice, so the column set has one home.)

3. Re-centre the view. The anomaly mode is the same four layers reading the delta columns, with the axis relabelled as a distance and a <Baseline> naming the new zero:

<YAxis id="gw" label="GW vs baseline" format="+.1f" width={56} />
<Layers>
  <LineChart series={series} column="aug09Delta" axis="gw" as="muted" />
  {/* …aug10Delta, aug11Delta… */}
  <LineChart series={series} column="delta" axis="gw" />
  <Baseline value={0} axis="gw" label="3-day baseline" />
</Layers>

The toggle between the two is ordinary React state — swap the column props and the axis; the pond pipeline above already carries both readings of the same data.

Options to try​

OptionWhat it doesReach for it when
More baseline daysA deeper pack behind the subjectThree days is a long weekend; thirty is a climatology — see Niño 3.4
Median, not meancollapse with a different reducerA baseline that shouldn't move when one control day was cloudy
as="muted"The reference pack in the theme's neutralAny shifted or historical comparison curves
format="+.1f"Signed tick labels on the anomaly axisAny re-centred view — the sign is the reading
min={0} on YAxisPins the raw view's floor at zeroQuantities-from-zero (generation, traffic); meaningless once re-centred
curve="monotone"Smooths all the bells15-minute and coarser cadences, where the ramp shouldn't look like line segments

See also​