HN Explorer

Show HN: Skyrim – Open-source Large Weather Model (LWM) Zoo (github.com)

19 points by sureklix ↗ HN
Hi HN, I'm Efe from Secondlaw AI. We are building physics-informed large AI models. Currently, we are focusing on weather modelling.

To benchmark SOTA, we had to build a forecasting infra for all available large weather models and we could not find a solid tooling to do so, so we built Skyrim. Within <5 mins and <5 LOC you can run forecasts on par with global weather models that are run on 100K+ CPU HPCs!

We are implementing more models & fine-tuning capabilities. Let us know if anything more we can add, also happy to answer any questions!

10 comments

[ 2.3 ms ] story [ 34.5 ms ] thread
luskira
Lol skyrim? hahah
sureklix
this is what happens when you listen to the soundtrack on repeat for weeks while developing this...
u2x
Thanks for releasing this.

Always wanted to play with these models!

It's great to be able to use modal and get predictions without a gpu.

sureklix
It shines especially when you need to run ensembles!
JojoFatsani
Are you just entirely unconcerned with being sued for trademark infringement?
sureklix
OK so feels like this needs some explanation. It is in an entirely unrelated field, just like Palantir (also directly from LOTR). Not to distract the discussion, but Skyrim trademark is registered at USPTO in the field of computer game related anything.
Pwnguinz
What's the granularity of the prediction (I'm not sure if that's the correct word? I'm not a meteorologist)? Region level (100s km)? City level (a few 10s of kms)? Block level (a few kms)?

When you say it's on par with global weather models, how is "weather prediction accuracy" measured?

Cool, none-the-less!

sureklix
Very good question! These models are trained on ~40 years of ERA5 data –you can think of past forecasts from numerical models integrated with real observational data to have a continuous distribution of weather parameters (temperature, wind etc.) Therefore model resolution is 0.25 degrees (28km x 28km at the equator).

The way accuracy is measured is through picking targets (say temperature at 2 meters, at x,y lat & lon and forecasted 24h ago) and comparing them on RMSE and ACC (anomaly correlation co-efficient). For instance, in Google Graphcast paper they pick 1380 targets and the model out performs NWPs in 90% of them.

m13uz
To add to this, there are other ML models with higher resolution. For example, Google's MetNet-3 uses satellite radar images and ground measurements, and its resolution is 1km x 1km. And we are currently working on training a "nano" version of this!

Metnet-3: https://news.ycombinator.com/item?id=38122631