Original title: GOES-East full-disk AI nowcasting of cloud evolution in observation space
Dhamma Kimpara, Omid Bagheri, Ivette Hernandez Banos, Byoung-Joo Jung +1 · August 2026
Clouds are notoriously hard to predict because they change quickly and span tiny to massive scales. Traditional weather simulations, simple tracking tools, and existing artificial intelligence models often struggle to balance accuracy, physical area, and update speed. To fix this, researchers created an artificial intelligence system called DOP+. It predicts cloud changes by forecasting infrared heat measurements directly from satellite observations, guided by broader atmospheric weather data. It generates fresh cloud predictions every 10 minutes over roughly a fifth of Earth's surface. In tests, the model beat standard baselines and conventional weather models across 0-6 h lead times, successfully keeping realistic storm details for 2-3 h. This opens a path toward tracking clouds worldwide as fast as storms develop.
Imagine trying to predict where puddles will form on a giant parking lot after a rain. If you only look at the puddles now, you assume they stay still—but they spread and shrink as rain continues. Now suppose you also know the slope of the lot and the direction of the wind; then you can better guess how the puddles will evolve. Here, the puddles are clouds, the parking lot is the satellite image of Earth, and the extra information is weather data that helps the model see where clouds are heading.
Clouds disrupt flights, solar panels, and storm warnings, but are notoriously hard to forecast beyond a few minutes. DOP+ predicts their shape and movement across an area the size of two USAs, updating every ten minutes. That means pilots get better route warnings, solar farms can anticipate power drops, and meteorologists see storms forming before they escalate.
The abstract does not discuss performance beyond 6 hours, nor does it evaluate rare or extreme cloud events. The model is trained only on GOES-East data, so applicability to other satellites or global coverage remains untested. The method's computational cost and real-time feasibility are not addressed.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
GOES-East full-disk AI nowcasting of cloud evolution in observation space — Dhamma Kimpara, Omid Bagheri, Ivette Hernandez Banos et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.