Progress over time
Satellite rainfall observations help an AI weather model learn
A preprint tests a different way to train rainfall forecasts. It reports improvements, but not a replacement for physical models.
Schmitt and colleagues · arXiv preprint · September 2026
Research claim
Deep learning narrows one source of uncertainty in El Nino projections
Neural networks trained on climate simulations and checked against ocean observations reduced the reported spread in projected El Nino variability under a high-emission scenario. The result constrains one climate process; it is not a precise forecast of future El Nino events.
Zhu and colleagues - Nature Communications - August 19, 2025
Research claim
A battery model learns by comparing how different cells age
BatLiNet compared early charge-discharge patterns between battery cells and improved lifetime prediction across several public datasets and ageing conditions. It is a retrospective modelling result, not proof of longer-lasting or safer batteries in use.
Zhang and colleagues - Nature Machine Intelligence - January 15, 2025
Independent validation
A hybrid AI model maps carbon changes across US cropland in finer detail
A model combining process knowledge, satellite data and field observations estimated crop and soil carbon across the US Corn Belt at 250-metre resolution. It improves research mapping, but it is not yet a field-level carbon-credit measurement system.
Liu and colleagues · Nature Communications · January 8, 2024
Research claim
4 of 4 results. End of results.