What happened
GraphCast is a graph neural network trained on ERA5 reanalysis that predicts hundreds of atmospheric variables at six-hour steps. The Science paper reported better deterministic skill than ECMWF’s high-resolution forecast on 90 percent of 1,380 verification targets and improved cyclone tracking and extreme-temperature metrics in the authors’ tests. Code and weights were released. The result is a research and later operational-influencing model, not a replacement of national weather services on its own.