For crop farmers, weather forecasts matter a great deal, especially at planting and harvest. But if farmers also know during the growing season that a drought is coming or it will be especially humid (boosting severity of fungal disease), that is also very helpful in terms of planning and mitigation.
New technologies are helping with better forecasting, including satellites with radar technology that can penetrate clouds. AI systems are also taking weather prediction to new and unprecedented levels.
Alex Levy explains that over the last several years, AI has become better at a very fast pace, improving in its ability to fill in data gaps and making useful recommendations in a wide range of situations – including meteorology. Levy is the CEO and co-founder of Atmo, an AI company based in San Francisco that specializes in accurate, longer-term weather forecasting.
Levy also explains that the classic way of predicting weather is called numerical weather prediction. It’s rule-based, meaning it uses specific physical equations to describe the behavior of the atmosphere. It’s simple in comparison to AI. It doesn’t learn from the past or from its mistakes.
Powerful
AI, of course, excels at learning. In a weather context, it compares its past predictions to reality, resulting in ongoing improvements to accuracy. “And now it’s at a stage where it’s very powerful,” says Levy. “That’s why we’ve seen over the last couple of years massive improvement in image creation, music creation, content creation, in self-driving cars and all kinds of other analysis.”
The Atmo team uses a variety of different AI systems. One is transformers, a type of deep learning neural network which specialize in predicting sequences (recognizing a pattern and continuing it into the future). “Mostly, it’s our team members improving the code,” Levy explains, “but we have a lot of AI systems that also generate and refine the code.”
Text continues below picture
The chaotic nature of weather systems is a tough challenge. – Photo: Canva
Achieved so far
At this point, Atmo’s AI produces forecasts that are up to 50% more accurate than prior forecasts, such as that of the US federal National Oceanic and Atmospheric Administration. “The detail can also be up to 100x higher in terms of weather for specific locations,” he says. “We can now produce forecasts up to 14 days ahead and see potential for them going out further into the future.”
Atmo’s clients currently include national governments (soon to include the Philippines and Tuvalu), militaries and companies in sectors like aerospace and energy in the US, Europe and Africa. Levy notes that in the energy sector, predicting wind direction and speed is of particular interest as it of course helps estimate power generation levels for wind turbines. For aviation applications, wind forecasting is also important for take-off, landing and turbulence.
Move into ag
“We expect to move into ag over the next year,” he says. “We are mindful that in the farming context, when we provide a solution, it must be economical. It could be that the best way to make that happen is through large ag conglomerates who can collect farmers together and provide our service to them, or through a government program but the question of how we might make it available to individual farmers is one that we’re actively studying.”
Looking forward, in terms of the challenges to even longer-term accurate forecasting, Atmo is exploring how more advanced AI models could enhance long-term weather forecasting.
Levy notes that while it’s too early to promise specific improvements, AI may be able to uncover subtle correlations across longer time and spatial scales. “The chaotic nature of weather systems is a tough challenge, but we’re excited to push the boundaries of what’s possible in meteorology.”
Join 17,000+ subscribers
Subscribe to our newsletter to stay updated about all the need-to-know content in the agricultural sector, two times a week.
