Ramps, Risks & Returns: How Forecast Blind Spots Drive Ramp Errors, Volatility and Market Outcomes

Candice Thompson:
Good afternoon, and a warm welcome to today’s Meteomatics and Montel webinar, “Ramps, Risks & Returns.”

I’m joined today by Jean-Paul Harreman, Director at Montel EnAppSys, and Rob Hutchinson, Energy & Utilities Lead at Meteomatics.

Today, we’ll be discussing price volatility trends and how weather-driven uncertainty has shaped market behavior in recent years, and how things are changing now.

We’re heading into quite an open discussion today. We’ll look at several case studies that illustrate the volatility we’ve experienced, and will continue to experience, in energy markets. We’ll also share a brief summer outlook to wrap things up.

That’s a lot to get through, so let’s jump straight in.

JP, why is increasing market volatility such a relevant topic right now?

Jean-Paul Harreman:
We’re seeing the energy market become more and more weather-dependent. We’ve installed a lot of renewable capacity, and where in the past people would ask, “How is this going to work if the wind isn’t blowing?”, that conversation has changed quite a bit.

Today, weather-dependent generation is actually pushing conventional generation out of the merit order.

So it no longer only makes sense to look at weather for renewable generation itself. It also matters because it determines how much fossil or other flexible generation is still going to be around. The more flexibility there is, the more stable the market becomes. The less flexibility there is, the more volatility you get.

If you see deviations in forecasts, you need to ask a few questions. Is it the weather? To what extent is it the weather? Is it price-related? When does it start? When does it end? And if it’s a combination of weather and price effects, you can end up with a period of blindness, for example during curtailment.

That’s why both Meteomatics and EnAppSys focus on real-time data, real-time forecasts, and nowcasts, so you get a better picture.

Looking at what drives volatility means understanding what sets the price. Is it fossil generation, or is it renewables?

When renewables are high, fossil flexibility becomes more expensive. When prices go negative, a lot of flexibility leaves the market already at the day-ahead stage. Renewable assets may be price-dependent, but when they are not generating, they cannot provide flexibility. At the same time, fossil fuel units may check out of the market because of negative spreads. You can’t ramp down something that isn’t running.

So what remains is largely non-dispatchable renewables and fossil units already contracted to provide reserves and stability. The ingredients for volatility are clearly there.

Candice Thompson:

Rob, take us through some of the high-impact weather phenomena that can have a major influence here.

Rob Hutchinson:
Absolutely. As JP said, the market is increasingly dependent on weather. That’s one of the main drivers behind our strategy at Meteomatics: investing in the next generation of high-resolution weather modeling that is specifically tuned to energy market requirements, particularly short-term power markets.

A lot of this comes down to renewables.

If we start with solar PV, some of the most important high-impact weather phenomena are:

• low cloud,
• fog,
• mist,
• snow and ice on panels,
• foehn conditions,
• and Saharan dust.

Low cloud, fog, and mist are notoriously challenging for traditional weather models. A lot of this comes down to vertical resolution as much as spatial resolution. Sometimes a relatively thin layer of moisture in the atmosphere determines whether it’s cloudy or foggy, and that can be very binary. Increased resolution can make a real difference there.

We also see issues with snow and ice accumulating on PV panels, and then uncertainty around when that snow or ice melts off again. That can create volatile pricing in winter.

Dense morning fog is another classic issue. The timing of when it burns off can be hard for weather models to capture. That’s especially relevant in shoulder months, and we’ve seen some recent situations where fog and low cloud had a clear impact on markets because reality diverged from forecasts.

Then there are foehn conditions, especially in and around the Alps. These involve the interaction of wind, moisture, and terrain, and they can be difficult for models to resolve unless the resolution is high enough.

And finally, Saharan dust. Earlier this month we saw quite a significant Saharan dust event. In Spain, the TSO solar forecast was overpredicting PV output by about three gigawatts because the dust impact had not been properly accounted for.

Moving on to wind, key phenomena include:

• thunderstorms and localized convective winds,
• icing on rotor blades,
• nocturnal jets,
• major wind storms,
• rapid wind direction changes associated with weather fronts,
• and wake effects.

Thunderstorms can cause abrupt generation spikes followed by lulls. One of the case studies we’ll look at is a classic example from Germany.

Icing on rotor blades is a major forecasting challenge, especially in the Nordics, but also in Eastern Europe and parts of Germany.

Nocturnal jets — narrow ribbons of fast-moving air at low levels — can interact with wind turbines in localized ways that are difficult to capture.

Major wind storms are another example. If a large offshore wind cluster suddenly reaches cut-out thresholds around 25 m/s, generation can drop quickly. Those events can materially affect prices and volatility.

And weather fronts can bring rapid changes in wind direction, with wake effects further complicating the picture.

Sometimes several of these factors combine, which is why we now hear more terms like Dunkelflaute and even Hellbrise — literally bright wind — where strong solar and windy conditions together create oversupply and negative pricing. One of the case studies we’ll show is a classic example of that.

Heatwaves are another important theme. Last summer we saw periods where solar output dropped rapidly in the evening just as cooling demand remained high, creating fast-moving and volatile pricing situations.

Candice Thompson:
Now that we’ve covered volatility and the impact of weather conditions, there are also important uncertainties that arise on any given day. That’s why energy and weather modeling are becoming increasingly important.

JP, do you have a question for Rob on this? And Rob, perhaps you have one for JP too.

Jean-Paul Harreman:
One of the questions we get a lot from customers is: what if everyone uses the same forecast? I have some good examples of where that happened. Rob, maybe you can tell us a bit about that.

Rob Hutchinson:
Yes, it’s an interesting one. There’s definitely a consensus trap. If everybody is following the same signal — the same traditional weather model — then everybody can be wrong in the same direction.

We often evaluate forecasts using standard metrics like mean absolute error or root mean squared error. But it’s also useful to evaluate them in terms of profit and loss. What would my P&L have been if I had taken a position based on a differentiated weather model?

So there is value in being accurate, but there is also value in being different.

Jean-Paul Harreman:
Exactly. I remember a day in April 2018, just before May 1, when a heavy storm was predicted to arrive on a Monday morning when much of Central Europe was off work. The storm didn’t arrive, and the whole of Central Western Europe ended up very short for almost the entire day.

That was a very clear example of consensus risk. A frequently updated weather model would probably have helped a lot, but at the time that just wasn’t as usable as it is today.

Rob Hutchinson:
That’s exactly how we see it. Commonly used models like ECMWF have limitations, not only in terms of resolution, but also update frequency. A lot can change in six hours, and many traditional models update only every six hours. In intraday markets, that can be a very long time.

JP, you mentioned periods of blindness during curtailment. How can weather data help address that, perhaps together with synthetic power output?

Jean-Paul Harreman:
It’s very useful to have raw weather data feeding into a generation forecast so you can estimate what the generation potential would actually have been during a specific period.

We can estimate how much volume is curtailed on an economic basis, but if you lose the feedback signal from live measurements, then you become blind. You don’t necessarily know whether the wind picked up or not, because economic decisions are now driving the output, rather than the physics alone.

That’s why it’s so important to know what wind speed and radiation are actually doing in relation to the forecast. And obviously, you can’t wait six hours for the next model run if you want that feedback in time.

Rob Hutchinson:
There’s a lot of observational weather data out there, but is the observation alone enough for this use case? Or do you also need to know what the next few hours will look like?

Jean-Paul Harreman:
If you’re an intraday trader, you definitely want to know what the next few hours look like. If you’re hedging a portfolio of assets spread across different geographies, then it becomes a bit less about the next hour and more about the broader picture.

Audience Question:
Do day-ahead forecasts try to integrate phenomena like blade icing and fog?

Rob Hutchinson:
Yes and no. Fog is certainly accounted for in day-ahead forecasts, and all numerical weather models attempt to resolve fog.

Blade icing is more complicated. We’ve recently invested quite a bit of time and effort in building highly specific blade icing forecasts. But it can be risky to directly modulate a day-ahead forecast based on those icing forecasts.

Our view is that it’s often better to alert clients to the risk of those events rather than trying to force them directly into the forecast itself.

Jean-Paul Harreman:
I agree. Having a good view of the risk helps you make a better economic decision. You may choose not to sell your full portfolio, or to be more conservative in your forecast. These are uncertain events, so having the warning and the situational awareness is crucial. It’s more about awareness than trying to force certainty into a forecast.

Candice Thompson:
Before we go too deep into the case studies, JP, I know there was a lot of discussion around Easter and Montel being on high alert. Can you tell us what you were seeing ahead of that period, and then what actually happened?

Jean-Paul Harreman:
Sure. What we were seeing was that Easter would be sunny and also quite windy. On the first day, we posted a warning on LinkedIn saying that with Easter demand being low and wind and solar expected to be relatively high across Europe, we were likely to see oversupply.

That means low prices, low fossil generation, and low flexibility. In that type of system, you can’t really absorb any large deviations.

That was the warning. Saturday looked relatively quiet. On Sunday we saw strong negative prices and a bit more volatility. But Monday is where things really became interesting.

Belgium got the short end of the stick — by being very long. There was massive oversupply, very strong solar, and also quite a significant wind ramp in the evening.

Normally, during the solar peak, fossil generation is pushed out of the merit order by solar, then ramps back up for the evening. In this case, that didn’t happen. It became very unprofitable to run gas assets through the solar peak just to preserve flexibility for later.

The Belgian system went very long for several reasons. Solar was higher than expected, but demand was also forecast too high in the day-ahead forecast — as if it were a normal working day. The intraday forecast later corrected that, but by then it was too late. That incorrect signal had already fed into market models, and people overbought because they were expecting higher demand.

Belgium tried to export the surplus, but neighboring countries were also long on solar. France still had some flexibility from nuclear ramp-down, but that too was limited.

So we ended up in what I called a downward dispatch domino effect. Flexibility was being ramped down everywhere. In Belgium there was basically nothing left. Solar was satisfying a huge part of demand. In France, nuclear couldn’t ramp down much further. In the end, France and Germany moved into a major surplus situation.

Just before delivery, the French TSO dumped power across the border to Germany using replacement reserve. That forced the German TSO to activate mFRR reserves, because the automatic reserves had already run out in Belgium and Germany. Then the mFRR merit order also ran out, reaching an activation price of minus €15,000.

There was also confusion because Belgium has an aFRR price limitation of plus/minus €1,000, so some participants assumed the price couldn’t go lower than that. But the mFRR price fed into the balancing price, which resulted in 1 hour and 15 minutes at minus €15,000.

That’s life-changing if you’re on the wrong side of it.

Candice Thompson:
Rob, can you take us through some of the meteorology behind that event?

Rob Hutchinson:
Yes. This is a classic example of a market that becomes extremely sensitive to relatively small forecast differences.

When we looked back at how different weather models were projecting the day-ahead scenario, we compared the Meteomatics EURO1k model, ECMWF IFS, and ICON-D2.

What we saw was that ECMWF and DWD ICON both maintained more persistent high and medium-level cloud across Belgium during that period, whereas EURO1k showed relatively clear skies, apart from one patch of high cloud crossing during the day.

Our working hypothesis is that this was what caused the dominoes to fall. The day-ahead solar forecast ended up around 1.5 GW lower than what actually occurred, and in a market that sensitive, that was enough to trigger the cascade.

If you aggregate solar irradiance across all of Belgium, the differences between the models may appear small, but when the market is that fragile, it doesn’t take much to start the domino effect.

The same pattern showed up when we looked at some of Belgium’s largest solar farms on a site-specific basis. EURO1k generally captured what actually happened better than the broader-scale models.

Let’s move to the next case study.

Jean-Paul Harreman:
This one is very different. On June 14 last year, we saw a massive imbalance spike in Germany.

What happened was that, until about an hour before delivery, nothing particularly dramatic seemed to be happening. Then intraday prices suddenly shot up. People started realizing that their weather model was wrong — or simply looking out the window and seeing thunderstorms approaching.

That’s an underrated aspect of power trading, by the way: people looking out the window.

These thunderstorms had a big impact because they suddenly reduced solar generation and increased demand for the evening. In June, it’s still relatively light late in the day, but when storm clouds come in it gets dark very quickly.

Earlier in the day there had already been curtailment because of high solar and wind, so there wasn’t much flexible generation left online for the evening peak. When the market suddenly went short that quickly, it put major strain on the remaining reserves.

A lot of aFRR was used, a lot of cross-border support was activated, and then mFRR as well. As we saw in the Belgian case, mFRR merit order curves are very steep, so the imbalance price became extremely high.

Rob Hutchinson:
Meteorologically, this was another example of extreme sensitivity.

On the right-hand panels, we could see thunderstorms moving in from the west, associated lightning strikes, and shifting wind fields. We compared EURO1k with ECMWF.

From a wind perspective, what happened was that as the thunderstorms passed through northwestern Germany, there was first a sudden ramp in wind associated with convective outflow. Then, after the storms, there was a lull in wind speeds — especially in northwestern Germany — and that hit the German wind fleet quite hard.

ECMWF didn’t really capture that well, and that’s not surprising. It’s not designed to resolve those localized convective wind fields. EURO1k did a much better job of capturing the localized dynamics.

So the combination was likely:

• dark skies that had not been forecast properly, reducing PV,
• and a wind lull in northwestern Germany, reducing wind output.

That combination likely drove the spike in prices.

Jean-Paul Harreman:
A more recent example was just two weeks ago, on April 13, when persistent morning fog affected Germany.

You could see it clearly in the fuel mix. The solar curve had a big dent in it instead of a normal bell shape. That usually points to morning fog taking longer than expected to clear. It kept balancing prices between €300 and €600 for quite some time.

You could also see the intraday trading evolution: as delivery approached, traders started pushing prices up because they realized the fog wasn’t going away.

Again, you can sometimes see that by looking out the window, but in a large market like Germany that’s obviously not enough. Having a better model that captures this is very useful.

We also forecast the shortage in the balancing market. That combined renewable nowcasting with the delta versus the latest major weather model runs, plus the assumption that market participants would start trading themselves out of trouble.

That kind of forecast constantly updates itself. Four hours before delivery it may not be as accurate as 15 minutes before delivery, but it gives you a combination of fundamentals and market sentiment. That is basically how you need to look at energy markets now. Fundamentals alone are no longer enough.

Rob Hutchinson:
In that case, the market expected around 22 GW of PV, while reality delivered just under 20 GW.

When we looked retrospectively at what the weather models were showing, EURO1k, ICON-D2, and ECMWF IFS differed substantially. Across multiple locations in Germany, ECMWF and IFS showed less cloud cover than EURO1k. EURO1k better captured the persistent low cloud and fog — and probably some afternoon convective cloud as well.

When we compared day-ahead forecasts against actual German PV output, the forecast derived from EURO1k tracked actuals much more closely than ECMWF-based alternatives. It’s a good example of where a higher-resolution model is better able to handle fog and low-cloud complexity that traditional models struggle with.

Candice Thompson:
Thanks, both. These are excellent examples.

We’re now heading into summer. Do you have a sense of the main trends or what we might expect from a summer outlook, and what that could mean for markets?

Rob Hutchinson:
Seasonal forecasting is notoriously difficult, so that needs to be said upfront.

What I’ve done here is use the Copernicus Climate Change Service multi-system seasonal forecast and break it down into the main European trading regions.

The main signal is heat. The strongest heat anomaly appears in Southeastern Europe. There is also a dry signal.

I should note that the Copernicus seasonal forecast is not fully detrended relative to a climate-change-adjusted baseline, so there is a bit of an artifact there. But even so, the signal is clearly pointing toward:

• heat,
• dry conditions,
• and associated risks.

That probably also implies higher-than-normal solar and lower-than-normal wind. So it could be a very interesting summer.

It’s not only the seasonal models pointing in that direction; some teleconnections are as well. It’s still early, and these things can change, but it’s definitely something to watch closely as we move further into the season.

JP, based on those signals, how do you think about the risks and opportunities for market participants?

Jean-Paul Harreman:
It’s a big question — and it’s not even summer yet.

I’d actually start with the rest of spring. Last weekend we saw a major negative price event, with prices dropping below minus €350 across many countries, and some countries reaching minus €490 to minus €500.

If we see this again within 30 days of the first event, that will trigger a recalibration of the minimum price for the day-ahead market. The floor would then move down by another €100, to minus €600/MWh.

That’s significant, because anyone using a hard floor of minus €500 in their systems would have to update that within 28 days of the trigger. It could also affect things like collateral requirements posted at TSOs for balancing exposure.

Looking ahead to summer: the Nordics have already been dry, and those markets — which usually provide cheap power — have already been among the higher-priced regions in Europe. If we get a warm and dry summer there as well, we could see the Nordics importing from continental Europe, at least during daytime hours.

More broadly, renewables will continue pushing fossil and flexible generation out of the market during the day. That reduces the amount of flexibility available to the system. If those flexible assets are not in the market, they cannot provide flexibility later.

Then in the evening peak, prices may be higher than expected. When the sun goes down, power has to come from somewhere. Some of it will come from French nuclear ramping back up, but that flexibility is limited — around 15 GW — and that does not cover all of Europe. So gas assets will have to ramp up for the evening peak.

Given the current uncertainty around gas prices, that could result in very high evening prices. If you run a gas asset, you may only have a few evening hours to recover the start-up cost of the unit, and that feeds directly into marginal cost and therefore price formation.

There’s also the risk that if temperatures rise enough, parts of the French nuclear fleet may have to reduce output as well, which would add even more volatility.

Candice Thompson:
That also has implications for gas injection over the summer, doesn’t it?

Jean-Paul Harreman:
Yes. You also lose efficiency. Some gas assets, for example in the Netherlands, use shallow-roof cooling systems, and as water temperature rises, efficiency falls.

But for batteries, this is good news. The top-bottom spreads — which is the term everyone uses now — are likely to be high. You could see negative prices during the solar peak and high prices in the morning and evening peaks. So there are plenty of opportunities for people with flexible assets that can respond quickly.

Candice Thompson:
Let’s open things up to audience questions.

Audience Question:
What makes the EURO1k model so accurate? Is it just spatial resolution, or does it use other meteorological data and techniques?

Rob Hutchinson:
Spatial resolution is part of it, but it’s not the whole story.

Other important factors include:

• vertical resolution,
• update frequency,
• the physical parameterizations we use,
• wake-effect representation,
• and the fact that the model is highly tuned toward the variables that matter most for power markets, especially solar and wind.

It’s also a non-trivial investment, both in the computing hardware needed to run it and in the R&D behind it.

Unlike broader public models like ECMWF, DWD, or the Met Office — which have to be generalists and perform well across a wide range of applications — we can focus our efforts specifically on energy-market-relevant forecasting.

We’re also quite unique in operating our own network of Meteodrones, and that data feeds into the overall system as well.

Audience Question:
How realistic is it to use weather forecasts beyond short-term power trading — for example, for mid-term hedging in the futures market? Given that forecast accuracy drops beyond three days, is it realistic to trust weather forecasts for the next month?

Rob Hutchinson:
Beyond three days — and sometimes even before that — you need to think in terms of probabilistic or ensemble forecasts.

The key is not just asking what the most likely scenario is, but also understanding the spread of risk around it. That allows you to take a more informed position.

There are various probabilistic weather models available, and we make several of them available through the API. But the short answer is: yes, you can use weather information at longer horizons, but you need to think in terms of confidence and risk, not just a single deterministic forecast.

Jean-Paul Harreman:
Exactly. A probabilistic approach is the best way to go.

What helps is defining the economic effect of different scenarios. If you define a most-likely case, a best-case case, and a worst-case case, and understand the revenue implications of each, then you can hedge around those uncertainties — assuming you have the right hedging instruments available.

There will also be scenarios where the ensemble points strongly in one direction, and you can be more confident in taking a position. In other scenarios, uncertainty will be very high and a more risk-off approach makes sense.

The good news is that modern weather modeling makes all of this quantifiable. The challenge is how you use that information.

Audience Question:
What about recent EDF changes around dampening nuclear curtailment? Does that reduce flexibility?

Jean-Paul Harreman:
Yes, it does.

France doesn’t have much other flexible generation running besides nuclear. Battery deployment is still limited, and solar and wind are already reflected in the day-ahead market. So if nuclear curtailment is dampened, that does reduce flexibility.

I can understand why EDF would do it, because more aggressive ramping increases wear and tear on equipment. But from a market-flexibility perspective, it’s not good news.

That said, EDF might also decide to shut down some assets proactively, especially plants located on rivers facing low water levels or high temperatures. If there’s no money to be made, why keep them running?

The problem is that if per-asset flexibility decreases, then you need more assets to provide the same system response. Otherwise, you just end up exporting surplus to neighbors who do not need it.

Candice Thompson:
Thank you. We’ve covered a huge amount today: volatility, weather, pricing, case studies, seasonal outlook, and audience questions.

Thank you to everyone who joined us, and thank you for your questions and participation. If you’d like to reach out to Rob or JP afterward, both are active on LinkedIn.

We look forward to seeing you again at a later stage for another session. Thanks everyone, and have a great rest of your day.