Meet Cloudcasting: Our newest forecasting innovation

A world first innovation that predicts cloud movements through satellite imagery, dramatically improving solar generation forecasts.

The Challenge of Solar Forecasting

With the rise of intermittent energy sources such as wind and solar, the electricity grid has become more weather-dependent than ever before. Accurate renewable energy generation forecasts are essential for electricity grid and solar farm operators, as well as smart home owners. These forecasts allow the energy system to operate with high levels of solar and wind generation, despite the changeable nature of the weather.

However, there are significant challenges when it comes to traditional weather forecasts - one of the most difficult is accurate short-term prediction of cloud movements, as they are volatile and subject to rapid change. Not surprisingly, clouds passing overhead hugely impact the effectiveness of solar panels, slashing the amount of electricity generated by eighty percent.

The AI-Powered Solution: Cloudcasting for Solar Forecasting

At Open Climate Fix, we have developed a solution that forecasts solar energy hours into the future, enabling users to quickly adjust power generation based on predicted demand changes. It outperforms competitors, and is in use by the National Energy System Operator (NESO) in the UK, as well as system operators in India and the world’s largest solar park. 

At the heart of our solar forecast is PVNet, an AI model we developed two years ago that uses recent satellite imagery to predict solar energy generation. When working on the model we noticed a pattern - the forecasts became more accurate when we used satellite data closer to the forecasting time. 

That got us thinking: What if we could use satellite images from the future?

That question led to the research and development of our current AI for weather breakthrough - Cloudcasting. By leveraging AI and satellite imagery, Cloudcasting generates forecasts of cloud movements and atmospheric conditions before they happen. 

This allows PVNet to make even sharper, more reliable solar energy predictions, helping the UK’s electricity grid run cleaner and cheaper.

This solution is transformative. When compared to traditional cloud forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) and the UK Met Office, Cloudcasting improved solar energy forecast accuracy by 100%.

AI for Weather Forecasting

There has been an explosion of interest in using AI for weather forecasting in recent years, with models released by Google, ECMWF and NVIDIA, amongst others. However, these models focus on long timescales, predicting days ahead instead of hours, without specifically considering cloud movements. We are also not aware of any weather services currently predicting cloud movements through satellite imagery in a live environment, meaning we are confident that our innovation is a groundbreaking development within AI for weather forecasting.  

This project started in partnership with the Alan Turing Institute. Together, we proposed using AI to solve the ‘problem’ of cloud forecasting in a submission for the Manchester Prize, run by theUK Department for Science, Innovation and Technology.

We were delighted to be selected as one of ten finalists for the Prize in 2024, where we were awarded £100,000 to put towards developing our solution. Since then, we have created a fully functioning AI model which not only forecasts clouds, but significantly enhances the accuracy of our solar energy generation forecast. 

How It Works:

1. Satellite-Based Cloud Forecasting
   
   • Our cloudcasting AI model analyses multi-spectral satellite imagery from EUMETSAT, which includes 11 different spectral channels.
   • These channels provide detailed information about cloud height, thickness, and composition, crucial for estimating their impact on solar generation.
   • The AI predicts future satellite images up to 3 hours ahead, with updates every 15 minutes, offering unparalleled precision in short-term forecasting.
2. Enhanced Solar Power Forecasting
   
   • The cloud movement predictions feed into PVNet, our advanced solar power forecasting AI.
   • PVNet integrates weather forecasts from multiple providers and historical solar generation data to deliver highly accurate solar output predictions.
   • The synergy between cloudcasting and PVNet results in a 5% improvement in solar forecast accuracy at two hours ahead, surpassing ECMWF’s 2.5% improvement when integrated into PVNet.

Real-World Impact: Reducing Costs and Carbon Emissions

We plan to integrate Cloudcasting into our live solar forecasting service, Quartz Solar, by summer 2025. 

The benefit will be felt immediately by NESO, leading to lower bills for UK consumers and a decreased reliance on fossil fuels. We estimate Cloudcasting can cut carbon emissions in the UK by 100,000 tonnes annually. With the UK government target to quadruple solar power by 2035, Cloudcasting has the potential to deliver carbon savings equivalent to taking one million cars off the road. 

But the impact doesn’t stop there. With solar energy expected to make up more than half of all new electricity generation worldwide by 2030, this innovation could play a pivotal role in reducing global carbon emissions. 

Looking globally, our ambition is to establish our AI Cloudcasting as an essential input to the best solar forecasts across the world, while making the service free in the Global South, democratising access to the best AI generated information. 

A Smarter, Greener Future with AI

Quartz Solar AI Cloudcasting represents a transformative leap in solar energy forecasting, and we see it as a leading practical example of AI for good. 

—

Have questions or want to collaborate? Reach out to us.

Find out more about our leading solar forecast, Quartz Solar.

‍