Introducing Solar Nowcasting: Forecasting cloud movement for short-term power trading

Solar generation doesn’t follow a predictable script. A cloud drifts across the sky, and within minutes, megawatts of production can vanish. For short-term power traders managing solar assets, these rapid fluctuations create a constant challenge. How do you optimize your intraday position when conditions shift faster than traditional forecasts can keep up?

Today, we’re launching our answer: Solar Nowcasting, a satellite-based forecasting capability that delivers accurate irradiance predictions for the crucial 0-3 hour window ahead.

Solar nowcasting vs near-time signals

Consider a typical trading scenario: you’ve submitted your day-ahead position for a solar portfolio. At 10:00, cloud cover starts to build up over part of the fleet. Solar generation responds instantly to changes in irradiance conditions. Depending on their density and type, clouds can reduce output by 50-70% within minutes.

If the assets are equipped with near-time (SCADA-based) signals, you’ll see the production drop materialize in real time and can react accordingly in the intraday market. This works well, as long as a reliable near-time signal is available, such as live SCADA data. This is an approach we also support through a near-time add-on to our Power Forecasting.

However, for assets without live SCADA data, you would be exposed. By the time the deviation becomes visible in metered data or imbalance positions, the market opportunity to react efficiently may already be gone. Every deviation between forecasted and actual production affects your imbalance position, often at unfavorable prices.

This is where Solar Nowcasting comes in. It enables traders to anticipate deviations across all assets, including those without near-time data, and adjust positions in advance.

The technical challenge

Predicting solar irradiance over the next few hours requires solving one of meteorology’s most complex puzzles: forecasting cloud cover and movement.

Traditional numerical weather prediction (NWP) models are not designed for this use case. Their update cycles and spatial resolution limit their effectiveness for very short-term solar forecasting. The clouds tend to move faster than a standard weather model can incorporate new data and complete its calculations.

The challenge intensifies with the time horizon. While predicting conditions 15 minutes ahead is relatively manageable, maintaining accuracy over 2-4 hours requires sophisticated modeling of cloud development, dissipation, and movement patterns.

Notice the difference in clarity between NWP and satellite-based data:

Our solution: Satellite data meets advanced modeling

At Dexter Energy, we’ve developed an in-house nowcasting capability that builds on our extensive weather data infrastructure and meteorological expertise.

We currently ingest and store over 500 terabytes of historical weather forecast data, collecting millions of data points daily from all major weather providers. Our approach combines traditional numerical weather prediction models with modern AI systems, continuously benchmarking them to identify specific scenarios in which certain inputs deliver optimal performance.

Solar Nowcasting builds on this work by developing models specifically designed for short-term irradiance prediction.

The basis of our nowcasting system is geostationary satellite data, which provides several advantages:

• Low latency: We receive satellite observations shortly after they’re captured, enabling near-real-time analysis of current conditions.
• High frequency: New satellite images arrive every few minutes, providing continuous updates on the sun’s movement, as well as cloud positions and characteristics.
• Comprehensive coverage: Satellite data captures cloud thickness, movement patterns, and atmospheric conditions across our entire coverage area.

Our meteorological team combines this information with our existing weather-model infrastructure to model cloud evolution with the precision required for intraday trading decisions.

At the core of Solar Nowcasting sits our in-house cloud nowcasting model. It processes successive satellite images and translates them into short-term forecasts of how cloud systems will move, evolve, and dissipate over the coming hours. This is the first weather model we run fully in-house at Dexter. It builds on years of experience ingesting, validating, and combining weather data at scale.

First accuracy improvements

Solar Nowcasting is now fully integrated into our Power Forecasting and available for customers across Europe.

Early user results show near-time accuracy improvements across solar portfolios. These include a significant decrease in the weighted Normalized Mean Absolute Error (w-NMAE) for one- to two-hour-ahead forecasts.

Get in touch with our team to learn more about integrating nowcasting into your trading strategy.