Table 1 Training data, devices, and time for different global weather forecasting models

From: A hybrid framework for global weather forecasting via low-resolution dynamical core and multigrid neural operator

Model

Years of training data

Data resolution

Pressure levels

Training devices

Training time

Pangu-Weather

39 (1979–2017)

0.25°

13

192 Nvidia V100 GPUs

16 days

NeuralGCM 2.8

39 (1979–2017)

2.8°

37

16 Google Cloud TPU v4

1 day

NeuralGCM 1.4

39 (1979–2017)

1.4°

37

16 Google Cloud TPU v4

7 days

SMgNO

12 (2004–2015)

1.4°

13

4 Nvidia 4090 GPUs

1.4 days

HMgNO (ours)

12 (2004–2015)

1.4°

13

4 Nvidia 4090 GPUs

1.6 days

Back to article page