Global solar energy potential forecasting through machine learning and deep learning models

Why sunshine and smart computers matter

As the world heats up and extreme weather becomes more common, societies are scrambling for cleaner ways to power homes, transport, and industry. Solar panels are one of the most promising options: they are getting cheaper, more efficient, and can be deployed almost anywhere the sun shines. This study asks a simple but vital question: how far could solar power go by the middle of this century, and what would that mean for our overall energy mix, if we use modern data tools to peer into the future?

Looking at the past to see the future

The researchers gathered detailed global data from 2000 to 2022, including how much electricity the world used, how much came from solar panels, and the basic weather ingredients that drive solar power: sunlight, temperature, and wind speed. Over this period, global solar capacity exploded from just 1.23 gigawatts in 2000 to more than 1,000 gigawatts in 2022, and annual solar electricity jumped from about 1 terawatt-hour to over 1,300. These numbers reflect cheaper panels, better technology, and strong policy support in countries such as China, India, the United States, and across Europe. The team used these historical patterns as the foundation for their forecasts up to 2050.

Two smart ways to forecast sunshine power

To project the future, the authors turned to two families of computer models that specialize in time-based data. One, called SARIMAX, belongs to the traditional toolbox of statisticians; it looks for repeating patterns and trends and can also take into account outside factors such as temperature or wind that influence solar output. The other, known as a temporal convolutional network, is a deep learning approach that examines long stretches of data at once and can uncover more tangled, non-linear relationships. Both models were trained and tested on the same global dataset, and their accuracy was scored using standard measures of error and goodness of fit. In these tests, SARIMAX slightly outperformed the deep learning model, giving more precise forecasts overall.

How big solar could become

Using these tools, the researchers forecast that global solar power will keep racing ahead. According to SARIMAX, installed solar capacity could rise from about 1,300 gigawatts in 2023 to more than 11,600 gigawatts by 2050. The deep learning model gives a very similar upper value, a bit above 11,500 gigawatts. Solar electricity generation shows the same dramatic climb: from under 2,000 terawatt-hours in 2023 to roughly 15,000–16,000 terawatt-hours by 2050 across the two models. In other words, the world could produce more than ten times today’s solar electricity by mid-century if present trends continue and policies remain broadly supportive.

The changing mix of world energy

The study also looks at how this surge in solar fits into the broader mix of power sources—fossil fuels, nuclear, and all renewables combined. Historically, fossil fuels have dominated, with renewables and nuclear providing a smaller share. The models suggest that by 2050, renewable electricity, led by solar, will more than triple compared with 2023, reaching around 26,000 to 31,000 terawatt-hours per year. Nuclear power stays roughly flat or declines slightly, while fossil-based electricity still grows, especially in the SARIMAX scenario, though more slowly than renewables. This means that even as clean power expands impressively, strong action would still be needed to rein in coal, oil, and gas if humanity wants to hit strict climate targets.

What this means for everyday life

To a non-specialist, the technical details of the forecasting tools matter less than their message: there is ample solar potential to power a large share of human activity, and its role is set to grow sharply. The study finds that traditional statistical methods, when carefully tuned and fed with good data, can rival or beat more fashionable deep learning techniques for long-term energy forecasts. More importantly, both approaches point in the same direction—a future in which solar energy is central to the world’s power supply. If governments pair this growth with policies that limit fossil fuel use and support energy storage, electric vehicles, and carbon removal, a cleaner, more stable climate remains within reach.

Citation: Raza, M.A., Karim, A., Altayeb, M. et al. Global solar energy potential forecasting through machine learning and deep learning models. Sci Rep 16, 10466 (2026). https://doi.org/10.1038/s41598-026-41357-x

Keywords: solar energy, renewable power, energy transition, climate change, machine learning