Meteomatics

Delighted to be representing Meteomatics at the Montel UK Energy Day in London (thank you to the Montel team for the invitation)! My session is “Under Pressure: Climate Change, System Stress and the Forecast Gap”. ...with more wind and solar, shifting demand and volatile prices, the GB system is under pressure, and better weather and power forecasts are part of the answer. I’ll look at where forecast error really comes from. I’ll also ask whether El Niño can tell us what winter has in store, and what this means for the GB Energy. I’ve been digging into 75 years of ERA5 data, and the answers surprised me! Register here: https://lnkd.in/e4kysj8k #Energy #PowerMarkets #Weather #RenewableEnergy #Meteomatics

Utilities are being asked to do more with the same grid. Connect more renewables. Support electrification. Maintain reliability. Avoid costly infrastructure upgrades. But many operational decisions still rely on weather forecasts that miss the local conditions driving real-world outcomes. A cloud field over a solar park. A wind ramp affecting generation. A weather event that impacts one part of the network but not another. As renewable generation grows, these local weather effects have a growing impact on grid operations, flexibility requirements, and network planning. The next generation of weather forecasting gives utilities a clearer view of what’s coming—helping operators improve planning, unlock capacity, and maintain reliability in an increasingly weather-driven energy system.

Everything you need to know about our new AI NWP Hybrid upgrade in EURO1k

EURO1k leads on solar. Four metrics. 544 stations. DACH and France. Thirty days of live verification. Lowest RMSE. Lowest MAE. Highest correlation. ICON-D2 is the benchmark for solar in the DACH region. It's a high-resolution regional model, tuned for exactly this domain. We beat it. 11% lower RMSE versus raw ECMWF. 11.7% lower MAE. These aren't marginal gains. Solar imbalance costs are non-linear. A few percent of forecast error at the ramp translates directly into exposure at settlement. This is our AI-NWP hybrid system. Not a pure ML model trained on reanalysis. AI + NWP The chart speaks for itself.

Many AI weather model verifications are measuring the wrong thing. Here’s the problem nobody talks about. Verifying an AI weather model against ERA5? You’re not measuring forecast skill. You’re measuring how well the model reproduces the data it was trained on. ERA5 is not an observation dataset. It’s a reanalysis. A model-based reconstruction of the atmosphere— built by assimilating observations into an NWP system. And most AI weather models were trained on ERA5. So when you verify against ERA5, you’re essentially asking: “How well did the student reproduce the textbook?” Not: “Does it predict what actually happened in the real world?” That’s a fundamental difference. The very things that matter most for grid operations and balancing markets. The real question is always: Does the model predict what actually happened— at the location and scale that matters? That’s why we verify physics/AI hybrid EURO1k against an extensive network of high-quality ground observations. Not reanalysis. Because for AI weather forecasting to mean anything for energy production and demand— verification has to go beyond reanalysis. It has to test against the real world. #EURO1k #Meteomatics #WeatherForecasting #AI #WeatherData #Verification #EnergyTrading #ERA5 #WeatherIntelligence #EnergyTransition

If you verify an AI weather model against ERA5, are you really measuring forecast skill? Not really.... ERA5 isn't an independent observation dataset. It's a reanalysis i.e. a model-based reconstruction of the atmosphere (built by assimilating observations into an NWP system). That's a problem, especially when the AI model being tested was trained on ERA5 in the first place. You risk measuring how well it reproduces the reanalysis, not how accurately it predicts the real atmosphere. And ERA5 carries its own limits: coarse resolution, smoothed-out extremes, biases inherited from its data-assimilation system. The real question: does the model predict what actually happened, at the location(s) and scale that matters? It's why we verify Meteomatics physics-AI hybrid EURO1k model against an extensive network of high-quality ground observations, not reanalysis. For AI weather forecasting to mean anything for energy production and demand, verification has to go beyond reanalysis and test against independent, real-world data. #Weather #AI #Forecasting #Meteorology #WeatherData #EnergyTransition