#FourCastNet
Yash Diggikar, Milton Halem, Aishwarya Raman: An AI-Based OSSE Framework to Evaluate Learned Physical Consistency: A Multi-Model Study Using FourCastNet, Aurora, Prithvi WxC, and GraphCast https://arxiv.org/abs/2611.22024 https://arxiv.org/pdf/2611.22024 https://arxiv.org/html/2611.22024
November 18, 2026 at 6:42 AM
This is literally my day job and it still blows my mind sometimes.

ECMWF now shows ML weather forecasts from FourCastNet, Pangu-Weather, and GraphCast alongside the operational IFS on our public charts. Same verification, same standards. Plus we open-sourced ai-models so anyone can run them.
This is literally my day job and it still blows my mind sometimes.
You can compare physics-based and ML forecasts yourself, right now.
amplt.de
November 3, 2026 at 10:58 AM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=89)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
November 3, 2026 at 6:45 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=89)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
November 2, 2026 at 8:59 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=89)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
November 1, 2026 at 5:12 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=89)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
October 31, 2026 at 4:16 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=89)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
October 30, 2026 at 5:17 PM
Anyone else looking at AI weather prediction models and wondering how we move from 0.25-degree deterministic outputs to actually useful local ensembles?

GraphCast and FourCastNet are cool, but running inference on your own hardware without melting a GPU is the real frontier. Thoughts?
October 28, 2026 at 12:41 AM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=88)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
October 29, 2026 at 5:39 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=87)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
October 28, 2026 at 5:14 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=86)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
October 27, 2026 at 4:32 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=85)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
October 26, 2026 at 5:20 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=84)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
October 25, 2026 at 3:07 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.3 °F MAE, n=67)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
October 8, 2026 at 6:01 PM
How do we build AI weather systems that work for the people who need them most?

@pedramh.bsky.social's research sits at the intersection of climate physics, AI, and real-world impact, from co-developing FourCastNet to studying events AI models struggle to predict

youtube.com/shorts/aN1ja...
https://youtube.com/shorts/aN1jaMzlXqE
youtube.com
October 7, 2026 at 4:04 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.4 °F MAE, n=50)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
September 21, 2026 at 5:01 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.4 °F MAE, n=49)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
September 20, 2026 at 2:44 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.4 °F MAE, n=48)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
September 19, 2026 at 2:25 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.4 °F MAE, n=47)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
September 18, 2026 at 3:03 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.4 °F MAE, n=46)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
September 17, 2026 at 3:41 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.4 °F MAE, n=45)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
September 16, 2026 at 3:35 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.4 °F MAE, n=44)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
September 15, 2026 at 3:39 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.4 °F MAE, n=43)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
September 14, 2026 at 5:01 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.4 °F MAE, n=42)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
September 13, 2026 at 3:02 PM
Best raw 2 m temperature forecast of the models with n≥30 at lead day 1 over the last 90 days: ECMWF AIFS Single (2.4 °F MAE, n=41)
Instantaneous values at 00/06/12/18 UTC vs the observation at the same instants. Raw 0.25° model output, no post-processing.
https://castcheck.zifanzhang.com
September 12, 2026 at 2:07 PM