CAMExpedition
Members
First name (team leader)
Hisu
Last name
Kim
Organisation name
GIST
Organisation type
Academic (Student)
Organisation location
Korea (South)
First name
Wooseok
Last name
Jang
Organisation name
GIST
Organisation type
Academic (Student)
Organisation location
Korea (South)
First name
Minseo
Last name
Kwon
Organisation name
GIST
Organisation type
Academic (Student)
Organisation location
Korea (South)
First name
Jihun
Last name
Ryu
Organisation name
Utah State University
Organisation type
Research Organisation (Academic, Independent, etc.)
Organisation location
United States of America
Models
Model name
TxT
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 TxT
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)
We standardize the input by 40-yr ERA5 statistics.
If any, what data does your model rely on for real-time forecasting purposes?
This model consumes 10 IFS ensembles.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
We trained model using IFS ensembles as input and ERA5 EDA as target.
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)
We applied the method proposed in the following paper on U-net. Each IFS ensemble (for now we used 10 members) was used for training corresponding U-net and repeated 10 times while initializing model weight randomly. As a result, 10 * 10 = 100 U-nets are trained.
Detailed method can be found in P. Leeuwen, et al., (2025), Available: https://doi.org/10.1017/eds.2025.10027
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
U-net is based on J. Ryu, et al., (2026), Available: https://doi.org/10.5194/gmd-19-27-2026.
This model was used for superresolution and we adjusted for IFS postprocessing.
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
We didn't validated before submission yet.
Did you face any challenges during model development, and how did you address them?
None. Model size reduced but the performance does not decrease significantly so we decided to fix the model size as large as our GPU can handle.
Are there any limitations to your current model that you aim to address in future iterations?
We assume there is no uncertainty in ERA5 even though that is not the fact, for now. So we are planning to expand the source of uncertainty covering data (IFS, ERA5), model structure and its weight.
Are there any other AI/ML model components or innovations that you wish to highlight?
We used relatively simple model but employing the method from uncertainty quantification raised the performance.
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.
Hisu Kim designed and conducted overall development of this model including data preparation. Wooseok Jang and Minseo Kwon assisted data preparation. Jihun Ryu reviewed the process and corrected some errors while developing the 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)
We standardize the input by 40-yr ERA5 statistics.
If any, what data does your model rely on for real-time forecasting purposes?
This model consumes 10 IFS ensembles.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
We trained model using IFS ensembles as input and ERA5 as target.
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)
We applied the method proposed in the following paper on U-net. Each IFS ensemble (for now we used 10 members) was used for training corresponding U-net and repeated 10 times while initializing model weight randomly. As a result, 10 * 10 = 100 U-nets are trained.
Detailed method can be found in P. Leeuwen, et al., (2025), Available: https://doi.org/10.1017/eds.2025.10027
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
U-net is based on J. Ryu, et al., (2026), Available: https://doi.org/10.5194/gmd-19-27-2026.
This model was used for superresolution and we adjusted for IFS postprocessing.
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
We didn't validated before submission yet.
Did you face any challenges during model development, and how did you address them?
None. Model size reduced but the performance does not decrease significantly so we decided to fix the model size as large as our GPU can handle.
Are there any limitations to your current model that you aim to address in future iterations?
We assume there is no uncertainty in ERA5 even though that is not the fact, for now. So we are planning to expand the source of uncertainty covering data (IFS, ERA5), model structure and its weight.
Are there any other AI/ML model components or innovations that you wish to highlight?
We used relatively simple model but employing the method from uncertainty quantification raised the performance.
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.
Hisu Kim designed and conducted overall development of this model including data preparation. Wooseok Jang and Minseo Kwon assisted data preparation. Jihun Ryu reviewed the process and corrected some errors while developing the 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)
We standardize the input by 40-yr ERA5 statistics.
If any, what data does your model rely on for real-time forecasting purposes?
This model consumes 8 IFS ensembles.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
We trained model using IFS ensembles as input and ERA5 as target.
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)
We applied the method proposed in the following paper on U-net. Each IFS ensemble (for now we used 8 members) was used for training corresponding U-net and repeated four times while initializing model weight randomly. As a result, 8 * 4 = 32 U-nets are trained.
P. Leeuwen, et al., (2024), Available: https://arxiv.org/pdf/2405.20550
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
No.
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
We didn't validated before submission yet.
Did you face any challenges during model development, and how did you address them?
Computing resource is limited especially VRAM size, so we employed lots of GPU memory optimization such as gradient checkpointing and seperated model to train partially. Also,
Are there any limitations to your current model that you aim to address in future iterations?
The boundary of the output still has weird lines due to miscalculation of dimension yet so this is the first fault we will deal with. Next, the RPSS improves as we exploit more IFS ensembles so we plan to increase the number of it.
Are there any other AI/ML model components or innovations that you wish to highlight?
We used relatively simple model but employing the method from uncertainty quantification raised the performance.
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.
Hisu Kim designed and conducted overall development of this model including data preparation. Wooseok Jang and Minseo Kwon assisted data preparation. Jihun Ryu reviewed the process and corrected some errors while developing the model.
Model name
SBCDiff
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 SBCDiff
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)
Z-standardization and normalization
If any, what data does your model rely on for real-time forecasting purposes?
IFS outpus, and ERA5
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
IFS outpus, and ERA5
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)
In here, I used diffusion model.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
No
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
No
Did you face any challenges during model development, and how did you address them?
The model struggle to improve precipitation RPSS. I tested a lots of pre-processing.
Are there any limitations to your current model that you aim to address in future iterations?
I want to improve RPSS, but I don't know.
Are there any other AI/ML model components or innovations that you wish to highlight?
No
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.
Jihun Ryu - all
Hisu Kim - comment
Wooseok Jang, Minseo Kwon - discussion
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)
Z-standardization and normalization
If any, what data does your model rely on for real-time forecasting purposes?
IFS outpus, and ERA5
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
IFS outpus, and ERA5
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)
In here, I used diffusion model.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
No
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
No
Did you face any challenges during model development, and how did you address them?
The model struggle to improve precipitation RPSS. I tested a lots of pre-processing.
Are there any limitations to your current model that you aim to address in future iterations?
I want to improve RPSS, but I don't know.
Are there any other AI/ML model components or innovations that you wish to highlight?
No
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.
Jihun Ryu - all
Hisu Kim - comment
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)
Z-standardization and normalization
If any, what data does your model rely on for real-time forecasting purposes?
IFS outpus, and ERA5
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
IFS outpus, and ERA5
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)
In here, I used diffusion model.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
No
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
No
Did you face any challenges during model development, and how did you address them?
The model struggle to improve precipitation RPSS. I tested a lots of pre-processing.
Are there any limitations to your current model that you aim to address in future iterations?
I want to improve RPSS, but I don't know.
Are there any other AI/ML model components or innovations that you wish to highlight?
No
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.
Jihun Ryu - all
Hisu Kim - comment
Model name
VAEtherCast
Number of individuals supporting model development:
1-5
Maximum number of Central Processing Units (CPUs) supporting model development or forecast production:
8-48
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 VAEtherCast
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.
- Hybrid model that integrates physical simulations with machine learning or statistical techniques.
- An empirical model that utilises historical weather patterns.
- Ensemble-based model, aggregating multiple predictions to assess uncertainty and variability.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
Z-standardization using ERA5.
If any, what data does your model rely on for real-time forecasting purposes?
ECMWF IFS ensemble Z500 (control + perturbed)
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
ERA5 reanalysis Z500 (1986–2015), ERA5 daily total precipitation (tp)
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)
Regime-based statistical post-processing of IFS ensemble forecasts using ERA5 calibration to produce precipitation quintile probabilities.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
No.
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
No.
Did you face any challenges during model development, and how did you address them?
We focused on robust, interpretable post-processing by standardising preprocessing across datasets and ensuring stable probabilistic outputs.
Are there any limitations to your current model that you aim to address in future iterations?
Current limitations include limited predictor diversity and sensitivity to regime or teleconnection shifts. We plan to expand inputs and strengthen calibration for different seasons and regions.
Are there any other AI/ML model components or innovations that you wish to highlight?
No.
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.
Wooseok Jang - All
Minseo Kwon - All
Hisu Kim - All
Jihun Ryu - Conceptualization, Model architecture
Which of the following descriptions best represent the overarching design of your forecasting model?
- Post-processing of numerical weather prediction (NWP) data.
- Hybrid model that integrates physical simulations with machine learning or statistical techniques.
- An empirical model that utilises historical weather patterns.
- Ensemble-based model, aggregating multiple predictions to assess uncertainty and variability.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
Z-standardization using ERA5.
If any, what data does your model rely on for real-time forecasting purposes?
ECMWF IFS ensemble Z500 (control + perturbed)
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
ERA5 reanalysis Z500 (1986–2015), ERA5 daily total precipitation (tp)
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)
Regime-based statistical post-processing of IFS ensemble forecasts using ERA5 calibration to produce precipitation quintile probabilities.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
No.
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
No.
Did you face any challenges during model development, and how did you address them?
We focused on robust, interpretable post-processing by standardising preprocessing across datasets and ensuring stable probabilistic outputs.
Are there any limitations to your current model that you aim to address in future iterations?
Current limitations include limited predictor diversity and sensitivity to regime or teleconnection shifts. We plan to expand inputs and strengthen calibration for different seasons and regions.
Are there any other AI/ML model components or innovations that you wish to highlight?
No.
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.
Wooseok Jang - All
Minseo Kwon - All
Hisu Kim - All
Jihun Ryu - Conceptualization, Model architecture
Which of the following descriptions best represent the overarching design of your forecasting model?
- Hybrid model that integrates physical simulations with machine learning or statistical techniques.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
Z-standardization using ERA5.
If any, what data does your model rely on for real-time forecasting purposes?
IFS and ERA5.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
Reanalysis data (ERA5).
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)
The model combines a variational autoencoder (DIVA) and a Bayesian Gaussian Mixture (DPMM) for probabilistic post-processing.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
No.
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
No.
Did you face any challenges during model development, and how did you address them?
We focused on improving mathematical rigor and revising the model to ensure interpretability throughout the development process.
Are there any limitations to your current model that you aim to address in future iterations?
Our model is limited by the dependence on a single reanalysis-forecast dataset. Future work will focus on improving physical consistency.
Are there any other AI/ML model components or innovations that you wish to highlight?
No.
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.
Wooseok Jang - All
Minseo Kwon - All
Hisu Kim - All
Jihun Ryu - Conceptualization, Model architecture