IFUAIHydromet
Members
First name (team leader)
Rebecca
Last name
Wiegels
Organisation name
IMK-IFU, KIT
Organisation type
Research Organisation (Academic, Independent, etc.)
Organisation location
Germany
First name
Julius
Last name
Polz
Organisation name
IMK-IFU, KIT
Organisation type
Research Organisation (Academic, Independent, etc.)
Organisation location
Germany
First name
Selina
Last name
Janner
Organisation name
IMK-IFU, KIT
Organisation type
Research Organisation (Academic, Independent, etc.)
Organisation location
Germany
First name
Luca
Last name
Glawion
Organisation name
IMK-IFU, KIT
Organisation type
Research Organisation (Academic, Independent, etc.)
Organisation location
Germany
First name
Christian
Last name
Chwala
Organisation name
IMK-IFU, KIT
Organisation type
Research Organisation (Academic, Independent, etc.)
Organisation location
Germany
Models
Model name
ProS2St
Number of individuals supporting model development:
1-5
Maximum number of Central Processing Units (CPUs) supporting model development or forecast production:
< 8
Maximum number of Graphics Processing Units (GPUs) supporting model development or forecast production:
< 4
How would you best classify the IT system used for model development or forecast production:
High-Performance Computing (HPC) Cluster
Model summary questionnaire for model ProS2St
Please note that the list below shows all questionnaires submitted for this model.
They are displayed from the most recent to the earliest, covering each 13-week competition period in which the team competed with this model.
Which of the following descriptions best represent the overarching design of your forecasting model?
- Other: Post-processing of ECMWF sub-seasonal range forecasts
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
As input data we use ECMWF sub-seasonal range forecasts of precipitation. We are in the process of testing to include further variables.
If any, what data does your model rely on for real-time forecasting purposes?
We rely on the sub-seasonal range forecasts, as we use them in inference.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
Reanalysis data (ERA5) and ECMWF sub-seasonal range forecasts
Please provide an overview of your final ML/AI model architecture (For example: key design features, specific algorithms or frameworks used, and any pre- or post-processing steps)
A simple U-Net Architecture trained with different combinations of RPS and entropy regularization as loss function
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
Not yet.
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
We have built a validation pipeline that compares the past forecasts to ERA5 and the skill of the post-processed forecast compared to the raw sub-seasonal range forecast.
Did you face any challenges during model development, and how did you address them?
Main challenges are (1) the limited amount of sub-seasonal range hindcasts that we use for training and (2) overcoming the climatology barrier.
Are there any limitations to your current model that you aim to address in future iterations?
Insufficient amount of S2S training samples, potentially by adding old model versions.
Are there any other AI/ML model components or innovations that you wish to highlight?
-
Who contributed to the development of this model? Please list all individuals who contributed to this model, along with their specific roles (e.g., data preparation, model architecture, model validation, etc) to acknowledge individual contributions.
Data preparation - Rebecca Wiegels, Luca Glawion
Model Architecture - Julius Polz, Luca Glawion, Selina Janner, Rebecca Wiegels
Model Validation - Julius Polz, Rebecca Wiegels
Coordination - Rebecca Wiegels, Christian Chwala
Which of the following descriptions best represent the overarching design of your forecasting model?
- Other: Post-processing of ECMWF sub-seasonal range forecasts
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
As input data we use ECMWF sub-seasonal range forecasts of precipitation. We are in the process of testing to include further variables.
If any, what data does your model rely on for real-time forecasting purposes?
We rely on the sub-seasonal range forecasts, as we use them in inference.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
Reanalysis data (ERA5) and ECMWF sub-seasonal range forecasts
Please provide an overview of your final ML/AI model architecture (For example: key design features, specific algorithms or frameworks used, and any pre- or post-processing steps)
A simple U-Net Architecture trained with different combinations of RPS and entropy regularization as loss function
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
Not yet.
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
We have built a validation pipeline that compares the past forecasts to ERA5 and the skill of the post-processed forecast compared to the raw sub-seasonal range forecast.
Did you face any challenges during model development, and how did you address them?
Main challenges are (1) the limited amount of sub-seasonal range hindcasts that we use for training and (2) overcoming the climatology barrier.
Are there any limitations to your current model that you aim to address in future iterations?
Insufficient amount of S2S training samples, potentially by adding old model versions.
Are there any other AI/ML model components or innovations that you wish to highlight?
-
Who contributed to the development of this model? Please list all individuals who contributed to this model, along with their specific roles (e.g., data preparation, model architecture, model validation, etc) to acknowledge individual contributions.
Data preparation - Rebecca Wiegels, Luca Glawion
Model Architecture - Julius Polz, Luca Glawion, Selina Janner, Rebecca Wiegels
Model Validation - Julius Polz, Rebecca Wiegels
Coordination - Rebecca Wiegels, Christian Chwala
Which of the following descriptions best represent the overarching design of your forecasting model?
- Other: Post-processing of ECMWF S2S forecasts
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
ECMWF S2S Forecasts
If any, what data does your model rely on for real-time forecasting purposes?
S2S forecasts
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
Reanalysis data (ERA5) and ECMWF S2S forecasts
Please provide an overview of your final ML/AI model architecture (For example: key design features, specific algorithms or frameworks used, and any pre- or post-processing steps)
A simple U-Net Architecture trained with CRPS as loss function
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
Not yet.
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
We have built a validation pipeline that compares the past forecasts to ERA5 and the skill of the post-processed forecast compared to the raw s2s forecast.
Did you face any challenges during model development, and how did you address them?
Main challenge for us is finding time to actually work on model development.
Are there any limitations to your current model that you aim to address in future iterations?
Insufficient amount of S2S training samples - which we will address in the next period.
Are there any other AI/ML model components or innovations that you wish to highlight?
-
Who contributed to the development of this model? Please list all individuals who contributed to this model, along with their specific roles (e.g., data preparation, model architecture, model validation, etc) to acknowledge individual contributions.
Data preparation - Rebecca Wiegels, Luca Glawion
Model Architecture - Julius Polz, Luca Glawion, Selina Janner
Model Validation - Julius Polz, Rebecca Wiegels
Model name
ProAFNOS2St
Number of individuals supporting model development:
1-5
Maximum number of Central Processing Units (CPUs) supporting model development or forecast production:
< 8
Maximum number of Graphics Processing Units (GPUs) supporting model development or forecast production:
< 4
How would you best classify the IT system used for model development or forecast production:
Single node system
Model summary questionnaire for model ProAFNOS2St
Please note that the list below shows all questionnaires submitted for this model.
They are displayed from the most recent to the earliest, covering each 13-week competition period in which the team competed with this model.
Which of the following descriptions best represent the overarching design of your forecasting model?
- Post-processing of numerical weather prediction (NWP) data.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
The model only requires the relevant sub-seasonal range forecast (ECMWF).
If any, what data does your model rely on for real-time forecasting purposes?
The latest sub-seasonal range forecast (ECMWF).
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
Next to sub-seasonal range forecasts the model is trained on reanalysis data for post-processing.
Please provide an overview of your final ML/AI model architecture (For example: key design features, specific algorithms or frameworks used, and any pre- or post-processing steps)
This model includes a Adapted Fourier Neural Operator (AFNO) block (Guibas, et al. 2022) that learns in fourier space, the encoding and decoding is still based on convolutional blocks.
Guibas, J. et al. (2022). Adaptive Fourier Neural Operators: Efficient Token Mixers for Transformers, International Conference on Learning Representations. DOI: 10.48550/arXiv.2111.13587
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
This model has been adapted from the application of Deep Learning based post-processing seasonal precipitation forecasts for the Blue Nile Basin (Wiegels et al. 2026)
Wiegels et al. (2026) Improved seasonal precipitation forecasts for the Blue Nile Basin: a deep learning approach. Front. Clim. 8:1691030. doi:10.3389/fclim.2026.1691030
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
We have validated the model against historical reanalysis data of ERA5.
Did you face any challenges during model development, and how did you address them?
-
Are there any limitations to your current model that you aim to address in future iterations?
Peaks of good skill in the raw sub-seasonal range forecasts is being flattened. We aim at maintaining good skill an reducing bad skill with our post-processing approach.
Are there any other AI/ML model components or innovations that you wish to highlight?
-
Who contributed to the development of this model? Please list all individuals who contributed to this model, along with their specific roles (e.g., data preparation, model architecture, model validation, etc) to acknowledge individual contributions.
Rebecca Wiegels: Data Preparation, Model architecture, Model validation
Luca Glawion: Data Preparation, Supervision
Julius Polz: Code structure, Supervision
Christian Chwala: Supervision
Submitted forecast data in previous period(s)
Please note: Submitted forecast data is only publicly available once the evaluation of a full competitive period has been completed. See the competition's full detailed schedule with submitted data publication dates for each period here.
Access forecasts data