2025-10-08
☔ AERIS
First billion-parameter diffusion model for weather + climate
🌀 SWiPe
| Variable | Description |
|---|---|
t2m |
2m Temperature |
X u(v) |
u (v) wind component @ Xm |
q |
Specific Humidity |
z |
Geopotential |
msl |
Mean Sea Level Pressure |
sst |
Sea Surface Temperature |
lsm |
Land-sea mask |
RMSNorm, SwiGLU, 2D RoPESWiPeSWiPe is a novel parallelism strategy for Swin-based TransformersSP)WP)PP)| Property | Value |
|---|---|
| Racks | 166 |
| Nodes | 10,624 |
| XPUs2 | 127,488 |
| CPUs | 21,248 |
| NICs | 84,992 |
| HBM | 8 PB |
| DDR5c | 10 PB |
🌡️ S2S Forecasts
We demonstrate for the first time, the ability of a generative, high resolution (native ERA5) diffusion model to produce skillful forecasts on the S2S timescales with realistic evolutions of the Earth system (atmosphere + ocean).
Goal: We would like to (efficiently) draw samples xi from a (potentially unknown) target distribution q(⋅).
Given x0∼q(x), we can construct a forward diffusion process by gradually adding noise to x0 over T steps: x0→{x1,…,xT}.
Step sizes βt∈(0,1) controlled by a variance schedule {β}t=1T, with:
q(xt∣xt−1)=N(xt;1−βtxt−1,βtI)q(x1:T∣x0)=t=1∏Tq(xt∣xt−1)
Introduce:
We can write the forward process as:
q(x1∣x0)=N(x1;αˉ1x0,(1−αˉ1)I)
We see that the mean μt=αtxt−1=αˉtx0
This research used resources of the Argonne Leadership Computing Facility, which is a DOE Office of Science User Facility supported under Contract DE-AC02-06CH11357.