# Anemoi dataset creation from forecasts with overlapping

**URL:** <https://forum.ecmwf.int/t/anemoi-dataset-creation-from-forecasts-with-overlapping/14887>\
**Category:** Software and Tools\
**Tags:** anemoi\
**Created:** [22 April 2026 14:49 UTC](https://forum.ecmwf.int/t/anemoi-dataset-creation-from-forecasts-with-overlapping/14887 "2026-04-22T14:49:15Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Alessio\_Golzio](https://forum.ecmwf.int/user_avatar/forum.ecmwf.int/alessio_golzio/32/751_2.png) [@Alessio\_Golzio](https://forum.ecmwf.int/u/Alessio_Golzio)\
**Post date:** [22 April 2026 14:49 UTC](https://forum.ecmwf.int/t/anemoi-dataset-creation-from-forecasts-with-overlapping/14887/1 "2026-04-22T14:49:16Z")

</div>

Dear all,

I am struggling to create a dataset using anemoi-datasets. I aim to create a dataset starting from daily forecasts (00 UTC initialisation) at 3h step from step 3 to 90. My dataset should contain the overlapping of all these forecasts (I have a 2-3 years of daily forecasts), so it should be a hypercube with 4 dimensions, I suppose, (valid-date, step, params, latlon). In detail, I have a grib file for each calendar date containing the forecasts of 2t up to 90 h. The analysis step is always missing. How can I configure the .yaml to achieve my goal? And is it possible with Anemoi (I hope so!!)?  
My project is to create many other datasets (I have 6 different meteorological models), and an observation dataset, then create a GNN and train it over this period. Afterwards, deploy the model to predict 2t (and in the future expand to 2td, wind, tp, etc) up to 90-96 h over a regular grid. Am I missing something about the framework. Any help is appreciated!