What a Solar Eclipse Does to a Power System

An old test, on a new grid

On 11 August 1999, the last total solar eclipse to touch the UK crossed Cornwall. The electricity system barely noticed the missing sunlight — there was almost no solar generation to lose — but it noticed the people. Demand on the England & Wales system fell by around 2,200 MW as the country stopped to watch, then rebounded with a 3,000 MW pickup as everyone went back indoors, put the kettle on, and returned to work. That surge is still the largest single demand pickup in British grid history, bigger than any royal wedding or World Cup penalty shootout.

Almost exactly 27 years later, on Wednesday 12 August 2026, the Moon passes in front of the Sun again. This time the eclipse is partial over Britain — but deep, covering up to 95% of the solar disc. And this time the grid underneath it is a completely different machine: Britain now has roughly 23 GW of installed solar — essentially none of which existed in 1999, and only about a quarter of which existed in 2015.

So the interesting question has flipped. In 1999 the eclipse was a demand event. In 2026 it is a supply event — and, because of when it falls, a much smaller one than the headline numbers suggest.

What actually happens on Wednesday

The eclipse arrives in the early evening, moving roughly north-west to south-east across the country:

(All times BST. Ireland sees the deepest partial in these islands, 93–98%.)

Two features of that table matter more than the coverage percentage.

First, it is late. Maximum eclipse lands between 19:05 and 19:16, with the Sun only about 10–16° above the western horizon. Solar output at that hour in mid-August is already a small fraction of its midday value — GB solar peaked at 15.4 GW at 13:00 on 23 April this year, but by 19:10 on an August evening the whole fleet is typically producing a gigawatt or two.

Second, it is on the falling limb. The Sun sets in London around 20:25 — some twenty minutes after the eclipse ends. There is no recovery ramp to speak of. The Sun sets before it finishes coming back.

That is the opposite of the March 2015 eclipse, which ran mid-morning while the Sun was climbing. In 2015 the difficult part was the recovery: solar coming back from an eclipse and rising with the morning, stacked on top of each other into one steep upward ramp. In 2026 the difficult part, such as it is, is the descent.

Why an eclipse is a strange kind of problem

Grid operators deal with variable renewables every day. An eclipse is different in three ways.

It is perfectly predictable. Orbital mechanics have no forecast error. Operators know the obscuration profile at every substation to the second, years in advance. That is a luxury they never get with weather.

It is perfectly correlated. Cloud is patchy — some farms dim while others don't, and a national fleet averages out. The Moon's shadow does not average out. Every panel in the country dims at nearly the same time, in the same direction, by a similar amount. An eclipse removes the geographic diversity that normally makes a large solar fleet well-behaved.

It is fast, in rate terms. Volume and rate are different things. NESO's expected loss at maximum is modest, but the steepest instantaneous ramp is not: one independent modelling estimate put Britain's worst-case ramp at roughly 3.8 GW/hour, with about 920 MWh of solar energy foregone across the whole event. That rate blends the eclipse with the natural evening decline — which is precisely the point. The two ramps add.

The numbers for Great Britain

NESO's published expectation is a drop of around 700 MW at maximum eclipse, with historical modelling suggesting up to 1.3 GW is possible under clear skies — roughly the output of a large nuclear unit, or, in the framing NESO used, the consumption of about a million homes.

For scale: National Grid expected to lose around 850 MW to the 2015 eclipse, on a fleet roughly a quarter of today's size. Four times the panels, a comparable loss — because the hour of day matters more than the installed capacity.

There is a wrinkle specific to Britain. The large majority of GB solar is distribution-connected and embedded — it sits behind the transmission boundary and is invisible to NESO as generation. It shows up instead as reduced demand. So from the control room's point of view, the eclipse is not primarily "1 GW of generation disappearing". It is national demand rising by about a gigawatt, on a summer evening when demand is already climbing towards the daily peak, and then falling back as the shadow passes — all superimposed on the normal evening ramp.

That is the honest description of Wednesday's event: a modest, entirely foreseen bump in net demand, arriving at a moderately awkward time of day.

How NESO and the European TSOs prepare

ENTSO-E — the body coordinating Europe's transmission system operators — has stood up a dedicated eclipse task force for the event, as it did in 2015. The playbook is unglamorous and effective:

• Extra forecasting. Operators are supplied with additional PV forecasts from weather providers, built specifically around the eclipse's obscuration profile rather than the standard irradiance model.
• Control-room briefing. Shift teams are briefed in advance so the ramp is anticipated rather than reacted to.
• Outage postponement. Planned grid outages are pushed out of the eclipse window to maximise operational flexibility — no lines or transformers deliberately unavailable while the shadow crosses.
• Reserve positioning. Additional reserve is held and fast-response plant readied. Portugal's REN, for instance, has said it will mobilise extra hydro and/or thermal reserve for its expected 450 MW loss.
• Market messaging. Participants are told what to expect, so the market itself closes most of the gap through day-ahead and intraday trading before the control room has to do anything.
• Seven-day adequacy checks. ENTSO-E's Short Term Adequacy assessment is used to confirm generation-versus-consumption balance across the interconnected system ahead of the event.

The last point is the crucial one. The vast majority of an eclipse response is scheduling, not real-time control. If it is done well, frequency barely moves and nobody notices — which is exactly what happened in 2015.

On the demand side, Britain is running an unusual experiment this time: Octopus Energy has asked customers to shift dishwashers, washing machines and other flexible loads out of the 18:00–20:00 window, offering an hour of free electricity the following Sunday in return. It is voluntary, and it is also marketing — but it is a real-world test of consumer flexibility against a known, scheduled system event, which is a rehearsal worth having.

The European picture

Across the continent, ENTSO-E expects PV output to fall by up to 9.7 GW at the deepest point, between 19:15 and 21:30 CEST. Spain and Germany are expected to see the largest reductions — Spain because the path of totality crosses it, Germany because of its sheer 125 GW of installed capacity. Published country estimates vary between sources (Spain has been quoted anywhere from about 3 GW to 5 GW, Germany around 2–3 GW); RTE's estimate for France, around 1,800 MW, is consistent across reports.

Set against 2015 — when the equivalent European figure was in the tens of gigawatts — this is a smaller event, for the same reason it is smaller in Britain: it happens near the end of the solar day rather than the middle of it.

For GB, the European context matters mainly through interconnectors. Britain's evening peak coincides with a continental solar dip, which tends to firm up evening prices on both sides of the cables. If you want to see the eclipse in the data, the price and flow panels may show it more clearly than the GB generation mix does.

What this means for inertia and frequency

This is where an eclipse gets counter-intuitive, and where it is worth watching on the dashboard.

Solar is inverter-based: it contributes no synchronous inertia. So losing a gigawatt of solar does not reduce system inertia — it removes non-synchronous generation. Whatever replaces it determines what happens to stability:

• If the gap is filled by CCGTs or pumped storage, synchronous inertia goes up slightly, because spinning mass is added to the system.
• If it is filled by interconnector imports or batteries, inertia is roughly unchanged — HVDC links and grid-scale batteries are inverter-based too.

The realistic expectation is inertia flat to modestly higher through the eclipse window, alongside a small rise in gas share and imports. On GridHz, that would show up in the inertia readout and the generation mix rather than in anything dramatic.

Frequency should be dull. A scheduled, forecast ramp is met by scheduled, forecast generation; frequency response reserves exist for unforeseen events, and a solar eclipse is the least unforeseen event in the calendar. If you see a RoCoF excursion on Wednesday evening, the eclipse almost certainly isn't the cause.

The real uncertainty is cloud

The one thing nobody can schedule is the British weather, and it cuts both ways.

If Wednesday evening is overcast across the country, solar output is already suppressed and the eclipse does essentially nothing electrically — a non-event with a large amount of reserve held for nothing. If it is unusually clear, the loss sits at the top of NESO's range. The error bar between 700 MW and 1.3 GW is not eclipse uncertainty; it is cloud uncertainty.

That is a small but real illustration of a bigger point about a high-renewables system: the deterministic astronomy is the easy part. The atmosphere is the hard part.

What to watch, and when

If you want to follow it live on Wednesday:

• 18:10–19:10 — the descent. GB solar share falls faster than the normal sunset curve; gas and imports pick up the difference. This is the interesting hour.
• 19:05–19:16 — maximum. Solar contribution close to zero across the fleet; the daily evening demand ramp is well underway on top of it.
• 19:16–20:10 — the non-recovery. The shadow clears but the Sun is setting, so solar stays low. The eclipse quietly merges into an ordinary summer evening.
• Through the window — watch inertia and the gas/import share together. The eclipse's clearest fingerprint on this dashboard is compositional: less inverter-based generation, more synchronous plant and interconnector flow.

The honest forecast is that nothing much will happen, and that this is the whole point. An eclipse is one of the few system events that can be planned for with total certainty about its cause and complete uncertainty about nothing except the weather. Britain's grid handled a 3,000 MW demand surge in 1999 with far cruder tools. A 700 MW solar dip in 2026, flagged weeks in advance and coordinated across an entire continent, ought to pass without incident.

Which will make it a good evening to look at the sky instead of the data — with proper ISO 12312-2 eclipse glasses, never with the naked eye.

Data: eclipse impact figures from NESO and ENTSO-E published statements, August 2026; eclipse timings via the Royal Observatory Greenwich. GridHz tracks GB inertia, frequency, generation mix, interconnector flows and prices in real time, refreshed every few minutes.