an hourly forecast for the past period

[Michael-Sharma]

1. Is it possible to get a forecast for the past period? I need to use weather forecast as features in my task
2. Is it possible to get an hourly forecast?

[andrewlkd]

Hi Michael,

1. As in our blog post

We'll soon be releasing real-time and historical forecasts from GenCast, and previous models, which will enable anyone to integrate these weather inputs into their own models and research workflows.

so do keep an eye out for this!

2. GraphCast and GenCast currently only operate at 6h and 12h resolutions respectively.

Hope this helps,

Andrew

[Michael-Sharma]

@andrewlkd, Thank you very much! I have a few questions left:

1. I'll rephrase the first question: is it possible to get forecasts for the past period using GenCast, for example, by training it on data up to 2023-12-01 and get the GenCast's forecast for the next N periods (2023-12-02, 2023-12-03, 2023-12-04, etc.)?
2. Did I understand correctly that at the moment the model makes a forecast in real time?

[andrewlkd]

1. Yes, absolutely. This process of training a model up until time T and then producing forecasts from T onwards is the causal mechanism we refer to, e.g. from the README

GenCast 0p25deg <2019, GenCast model at 0.25deg resolution with 13 pressure levels and a 6 times refined icosahedral mesh. This model is trained on ERA5 data from 1979 to 2018 (inclusive), and can be causally evaluated on 2019 and later years. This model was described in the paper GenCast: Diffusion-based ensemble forecasting for medium-range weather (https://arxiv.org/abs/2312.15796)

2. The model makes a forecast based off of whatever data you feed it as input. Whether this is a historic frame (as your previous question is asking), or is a frame from realtime data, it will predict the following frames.

Hope this helps,

Andrew

[Michael-Sharma]

@andrewlkd, Please tell me how and where in gencast_mini_demo.ipynb to fix the date from which a forecast is needed in the predictions df?