Spire experts reflect on AI trends and insights from the 105th AMS Annual Meeting

Last week, the American Meteorological Society’s (AMS) 105th Annual Meeting in New Orleans brought together experts and thought leaders throughout the Weather Enterprise with discussions on advancements in the field of weather prediction. Several Spire Weather & Climate experts participated, sharing insights and learning from other industry pioneers.?

From cutting-edge trends in AI/ML to the critical role of weather observations, here are the key takeaways our team brought back from the conference:?

1. AI and the shift toward data-driven forecasting?

?AI has revolutionized weather prediction over the past few years, moving beyond traditional physical methods. The next wave of innovation focuses on integrating raw observational data directly into AI models. ?

Mike Eilts, General Manager of Spire Weather & Climate, highlighted this evolution: “We are witnessing a paradigm shift in how weather forecasts are produced. The changes over the last two years have been phenomenal. We haven’t seen this level of industry disruption in the past 50 years. The Weather Enterprise — from research institutions to national weather services and private sector companies — is all grappling with how to utilize AI effectively.”?

“Research dollars and efforts have shifted dramatically, with around half of the papers now focusing on AI rather than traditional numerical weather prediction (NWP). Organizations are learning quickly and exploring new avenues,” he added.?

Building on this momentum, ECMWF’s upcoming operationalization of its all-AI model, AIFS, represents a significant shift in the industry. ?

As Spire Weather & Climate Senior AI Weather Scientist Dr. Nachiketa Acharya, observed, “That might be a game-changer because this is where physical model development has traditionally taken place. Now, with AIFS, the structure is all AI, even though it still uses ERA-5 based data.”?

Dr. Tom Gowan, Spire's Director of Weather Prediction and AI, added an insight that big tech — Google, NVIDIA, Microsoft —?is still leading the way with AI model innovations, also noting that Spire will soon release its AI models.?

2. Sub-seasonal to seasonal forecasting and the role of AI?

The sub-seasonal to seasonal (S2S) forecasting space is gaining traction, with AI models leading the charge. Acharya co-convened and co-chaired the AMS meeting’s AI S2S meeting. Spire is set to operationalize its AI ensemble and S2S models over the next few months, providing extended forecasts up to 45 days. These advancements are particularly valuable for renewable energy markets, where accurate long-term predictions drive wind and solar resource planning, as highlighted in an AMS talk presented by Gowan.?

“Our high-resolution and AI sub-seasonal models are tailored to support industries like renewable energy markets that depend on precision,” said Eilts. “Take power producers, for instance. Improved forecasts help producers optimize their output and maximize profits. For example, wind farm operators can better hedge their production and avoid penalties for over- or under-delivering.”??

Power buyers use forecasts to minimize costs by planning purchases days, weeks, or months in advance. “High-resolution models that extend up to 45 days are particularly valuable here,” Eilts explained, adding that commodity traders rely on these forecasts and aim to profit by predicting market dynamics better than others.?

3. Earth observations: The fuel for AI models?

Global datasets like Spire’s Global Navigation Satellite System Radio Occultation (GNSS RO) observations are instrumental in improving model accuracy. At AMS, discussions emphasized how these unique data sources enable more accurate weather forecasts. Experiments such as The International Radio Occultation Working Group’s Radio Occultation Experiment (ROMEX) project demonstrate that increased RO data significantly enhances forecast accuracy without reaching a saturation point.?

“AMS sessions highlighted initial findings about the transformative impact of radio occultation (RO) data on numerical weather prediction models. A study from ECMWF revealed that at 35,000 daily RO observations, this data source surpasses microwave temperature and water vapor as the most impactful, demonstrating the potential of scaling RO beyond 50,000 or even 100,000 observations daily without hitting saturation,” said Spire Weather & Climate Radio Occultation Expert Vu Nguyen. “These findings, which will be finalized in the next couple of months, show the promise of leveraging large-scale RO for more accurate and reliable forecasting.”?

Eilts emphasized the importance of proprietary data: “Data ownership has become a critical advantage in the AI-driven world. Companies with unique, high-quality data have a competitive edge in both AI and traditional modeling. A notable shift this year was the idea of bypassing data assimilation altogether by feeding raw data directly into AI models, enabling them to train and generate outputs independently. This approach could revolutionize how we use observational data.”?

4. Blurring the line between AI and physics-based models?

While AI models offer transformative capabilities, hybrid approaches that combine AI insights with traditional physics-based methods are emerging. This synergy ensures accuracy across scales and timeframes, addressing the limitations of standalone models.?

Acharya noted that an AMS session by Environment and Climate Change Canada (ECCC) highlighted recent research underlining the growing promise of hybrid forecasting systems combining AI and traditional physics-based models. A comparison of the physics-based GEM (Global Environmental Multiscale) model and the AI-driven GraphCast model revealed that GraphCast excels at predicting large-scale features, particularly for longer lead times. ?

Researchers used that physical model and ‘nudged’ the?GraphCast?to improve resolution and model accuracy?of 500 hPa heights in zero to seven days, Acharya explained. Nudging involves adjusting the physical model's large-scale predictions toward the AI model's insights, ensuring better alignment while allowing the physical model to refine local-scale climate details critical for accurate forecasting. ?

“By using knowledge from physics-based models in combination with AI, they were able to leverage the best from the two worlds,” said Acharya.?

5. Implications for industry and communities?

The AMS meeting reinforced the critical role of high-resolution, accurate weather data in decision-making across industries. From hedging risks in energy trading to optimizing supply chains and disaster preparedness, the applications of AI-driven weather forecasts are vast and expanding.?

“Our high-resolution model provides actionable insights for energy producers, commodity traders, and logistics planners,” Eilts explained. “These forecasts empower stakeholders to make informed decisions, safeguard assets, and maximize efficiency.”?

Trends in AI-driven weather forecasting?

?Eilts shared his thoughts on the challenges and opportunities in operationalizing AI-driven weather models: “The Weather Enterprise is at a crossroads. Research institutions, national weather services, and private sector companies are all?tackling?how to?leverage?AI effectively. In the private sector, there’s a lot of experimentation and marketing buzz, but few truly operational AI products are available yet.”?

“Companies that succeed will be those that operationalize AI to create real value, though that may still be a few years away for many applications,” Eilts continued. “The investment across the public and private sectors is immense, with everyone striving to figure out how to generate tangible benefits.”?

One emerging trend is the growing reliance on private-sector data collection and processing. “Instead of investing billions in government-led satellite programs, many are turning to private companies for faster, less risky, and cost-effective solutions,” Eilts noted. “This shift draws attention to the importance of unique, high-quality data in driving AI advancements.”?

Eilts also revealed exciting developments on the horizon: “Over the next few months, Spire will roll out new AI models focused on day 1 to day 15 forecasts and extended predictions up to 45 days. These models will leverage our proprietary observational data and generative AI techniques. We’re optimistic about the value they’ll deliver across various markets, and we’re eager to see their impact in the real world.”?

The AMS’ 105th Annual Meeting showcased the immense potential of AI, ML, and Earth observations in transforming weather prediction. As industries navigate increasing volatility and climate challenges, advancements in these fields will play a pivotal role in driving informed decision-making and resilience. ?

If you would like to talk with any of our experts further about AI and weather prediction, please contact Sarah Freeman, Spire Communications Manager, [email protected]