ERA5
The fifth-generation global atmospheric reanalysis dataset released by the European Centre for Medium-Range Weather Forecasts (ECMWF), widely used in meteorological research and in the training of machine learning models.
Timeline (last 90 days)
Mainstream AI meteorological models trained on ECMWF ERA5 reanalysis data have approached or even surpassed traditional numerical weather prediction (NWP) on multiple metrics.
On a simulator trained on historical ERA5 reanalysis data, under counterfactual settings with fixed CO₂, ReRoute retains more surface warming signals implied by observed boundary conditions, outperforming the baseline.
ERA5 reanalysis data is a global atmospheric reanalysis dataset released by the European Centre for Medium-Range Weather Forecasts (ECMWF), providing hourly three-dimensional atmospheric state estimates from 1940 to the present.
The confounding reduction rates of SSU-LSF on the three benchmarks CropHarvest, NDVI-LST, and ERA5 are 0.773, 0.821, and 0.859, respectively.
On ERA5, the worst-case RMSE degradation after SSU-LSF forgetting is 4.2%.
Using ERA5 simultaneously as a training target and evaluation reference introduces a risk of circular reasoning, because ERA5 itself is a product of model inference rather than direct observation.
The core logic of AI weather models is to treat the initial atmospheric state as token input to the model, predict the next state (e.g., 6 hours later), and continue the derivation autoregressively, similar to the token-by-token generation of LLMs.
AI weather models such as FourCastNet, Pangu-Weather, and GraphCast learn the statistical patterns of atmospheric dynamics by training on long-term reanalysis datasets such as ERA5
The spatial resolution of publicly available meteorological data (such as NASA POWER and ERA5 reanalysis data) is typically on the order of tens of kilometers, which limits the matching accuracy with township-level administrative units.
All Facts (9)
Mainstream AI meteorological models trained on ECMWF ERA5 reanalysis data have approached or even surpassed traditional numerical weather prediction (NWP) on multiple metrics.
50%UnverifiedOn a simulator trained on historical ERA5 reanalysis data, under counterfactual settings with fixed CO₂, ReRoute retains more surface warming signals implied by observed boundary conditions, outperforming the baseline.
50%UnverifiedERA5 reanalysis data is a global atmospheric reanalysis dataset released by the European Centre for Medium-Range Weather Forecasts (ECMWF), providing hourly three-dimensional atmospheric state estimates from 1940 to the present.
50%UnverifiedThe confounding reduction rates of SSU-LSF on the three benchmarks CropHarvest, NDVI-LST, and ERA5 are 0.773, 0.821, and 0.859, respectively.
50%UnverifiedOn ERA5, the worst-case RMSE degradation after SSU-LSF forgetting is 4.2%.
50%UnverifiedUsing ERA5 simultaneously as a training target and evaluation reference introduces a risk of circular reasoning, because ERA5 itself is a product of model inference rather than direct observation.
50%UnverifiedThe core logic of AI weather models is to treat the initial atmospheric state as token input to the model, predict the next state (e.g., 6 hours later), and continue the derivation autoregressively, similar to the token-by-token generation of LLMs.
50%UnverifiedAI weather models such as FourCastNet, Pangu-Weather, and GraphCast learn the statistical patterns of atmospheric dynamics by training on long-term reanalysis datasets such as ERA5
50%UnverifiedThe spatial resolution of publicly available meteorological data (such as NASA POWER and ERA5 reanalysis data) is typically on the order of tens of kilometers, which limits the matching accuracy with township-level administrative units.
50%