UWAtmosNVIDIA
Members
First name (team leader)
Dale
Last name
Durran
Organisation name
U Washington Atmos & Climate Sci -- NVIDIA
Organisation type
Other (please specify)
Organisation location
United States of America
First name
Peter
Last name
Harrington
Organisation name
NVIDIA
Organisation type
Large Tech Company
Organisation location
United States of America
First name
Nathaniel
Last name
Cresswell-Clay
Organisation name
U Washington Atmos & Climate Sci
Organisation type
Academic (Student)
Organisation location
United States of America
First name
Zachary
Last name
Espinosa
Organisation name
U Washington Atmos & Climate Sci
Organisation type
Academic (Student)
Organisation location
United States of America
First name
Raul
Last name
Moreno
Organisation name
U Washington Atmos & Climate Sci
Organisation type
Academic (Student)
Organisation location
United States of America
First name
William
Last name
Yik
Organisation name
U Washington Atmos & Climate Sci
Organisation type
Academic (Student)
Organisation location
United States of America
First name
Bowen
Last name
Liu
Organisation name
Institute of Atmospheric Physics, CAS
Organisation type
Academic (Student)
Organisation location
China
First name
David
Last name
Pruitt
Organisation name
NVIDIA
Organisation type
Large Tech Company
Organisation location
United States of America
First name
Akshay
Last name
Subramanian
Organisation name
NVIDIA
Organisation type
Large Tech Company
Organisation location
United States of America
First name
Michael
Last name
Pritchard
Organisation name
NVIDIA
Organisation type
Large Tech Company
Organisation location
United States of America
Models
Please note that the model information below reflects the latest details provided by the team and does not show previous versions. Only the model summary questionnaires retain a submission history.
Model name
DLESyMS2Sv1
Model type
Data-driven
Number of individuals supporting model development:
6-10
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:
16-64
How would you best classify the IT system used for model development or forecast production:
High-Performance Computing (HPC) Cluster
Documentation
Model summary questionnaire for model DLESyMS2Sv1
The list below shows all model summary questionnaires submitted for this model. Participants may submit an updated questionnaire whenever their methodology changes.
Questionnaires are displayed from the most recent to the earliest, with the latest version opened by default.
Which of the following descriptions best represent the overarching design of your forecasting model?
- Machine learning-based weather prediction.
- An empirical model that utilises historical weather patterns.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
The model was trained on ERA5 reanalysis data. The model uses a HealPix grid, the ERA5 reanalysis data was preprocess to the HPX64 grid
If any, what data does your model rely on for real-time forecasting purposes?
ECMWF OpenData IFS, GHRSST via CopernicusMarine open data,
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
ERA5 Reanalysis. Original model was also trained on ISSCP OLR satellite data, but we switched to the ERA5 TTR field because the later is available in near realtime.
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)
CNNs in a 3-level U-Net with all variables on a HEALPix mesh with 64x64 cells on each face. Model is extensively documented in the reference below. Aside from switching to TTR as a synthetic real-time substitute for satellite observed OLR, the only difference in our current model is the training loss function. We switched from RMSE to fair CRPS via an implementation similar to that in Lang et al., 2024 treatment of the AIFS
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
Cresswell‐Clay, N., Liu, B., Durran, D. R., Liu, Z., Espinosa, Z. I., Moreno, R. A., & Karlbauer, M. (2025). A deep learning Earth system model for efficient simulation of the observed climate. AGU Advances, 6, e2025AV001706.
https://doi. org/10.1029/2025AV001706
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
We performed a variety of tests as documented in our paper above, but we did not actually try any S2S forecasts.
Did you face any challenges during model development, and how did you address them?
Our biggest challenge for this forecasting exercise was correctly evaluating the rpss categories. our submissions were seriously in error for all forecasts submitted prior to 9 October.
But this problem has been eliminated before the start of the 2nd forecast quarter.
Are there any limitations to your current model that you aim to address in future iterations?
We are working on a v2 model with sea ice and upper ocean variables as well as more atmospheric levels (at present 30 atmospheric and 22 ocean variables, compared to 9 and 1 in the current model).
We would like to incorporate the 10 ERA5 EDA members as IC for our ensemble but have not done so yet.
Are there any other AI/ML model components or innovations that you wish to highlight?
It's not at all new to the astronomers, but I always like to highlight the HEALPix mesh.
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.
Nathaniel Cresswell-Clay: data preparation, model architecture, model validation
Peter Harrington: data preparation, model architecture, model validation
Zachary Espinosa: data preparation, model architecture, model validation
David Pruitt: data preparation, model architecture, model validation
Paul Moreno: data preparation, model architecture, model validation
Akshay Subramaniam: model architecture, model validation
William Yik: data preparation, model architecture, model validation
Bowen Liu: model architecture, model validation
Matthias Karlbauer: model architecture
Mike Pritchard: project management
Dale Durra: project management, model architecture
Which of the following descriptions best represent the overarching design of your forecasting model?
- Machine learning-based weather prediction.
- An empirical model that utilises historical weather patterns.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
The model was trained on ERA5 reanalysis data. The model uses a HealPix grid, the ERA5 reanalysis data was preprocess to the HPX64 grid
If any, what data does your model rely on for real-time forecasting purposes?
ECMWF OpenData IFS, GHRSST via CopernicusMarine open data,
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
ERA5 Reanalysis. Original model was also trained on ISSCP OLR satellite data, but we switched to the ERA5 TTR field because the later is available in near realtime.
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)
CNNs in a 3-level U-Net with all variables on a HEALPix mesh with 64x64 cells on each face. Model is extensively documented in the reference below. Aside from switching to TTR as a synthetic real-time substitute for satellite observed OLR, the only difference in our current model is the training loss function. We switched from RMSE to fair CRPS via an implementation similar to that in Lang et al., 2024 treatment of the AFIS
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
Cresswell‐Clay, N., Liu, B., Durran, D. R., Liu, Z., Espinosa, Z. I., Moreno, R. A., & Karlbauer, M. (2025). A deep learning Earth system model for efficient simulation of the observed climate. AGU Advances, 6, e2025AV001706.
https://doi. org/10.1029/2025AV001706
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
We performed a variety of tests as documented in our paper above, but we did not actually try any S2S forecasts.
Did you face any challenges during model development, and how did you address them?
Our biggest challenge for this forecasting exercise was correctly evaluating the rpss categories. our submissions were seriously in error for all forecasts submitted prior to 9 October.
But this problem has been eliminated before the start of the 2nd forecast quarter.
Are there any limitations to your current model that you aim to address in future iterations?
We are working on a v2 model with sea ice and upper ocean variables as well as more atmospheric levels (at present 30 atmospheric and 22 ocean variables, compared to 9 and 1 in the current model).
Are there any other AI/ML model components or innovations that you wish to highlight?
It's not at all new to the astronomers, but I always like to highlight the HEALPix mesh.
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.
Nathaniel Cresswell-Clay: data preparation, model architecture, model validation
Peter Harrington: data preparation, model architecture, model validation
Zachary Espinosa: data preparation, model architecture, model validation
David Pruitt: data preparation, model architecture, model validation
Paul Moreno: data preparation, model architecture, model validation
Akshay Subramaniam: model architecture, model validation
William Yik: data preparation, model architecture, model validation
Bowen Liu: data preparation, model architecture, model validation
Matthias Karlbauer: model architecture
Mike Pritchard: project management
Dale Durra: project management, model architecture
Which of the following descriptions best represent the overarching design of your forecasting model?
- Machine learning-based weather prediction.
- An empirical model that utilises historical weather patterns.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
The model was trained on ERA5 reanalysis data. The model uses a HealPix grid, the ERA5 reanalysis data was preprocess to the HPX64 grid
If any, what data does your model rely on for real-time forecasting purposes?
ECMWF OpenData IFS, GHRSST via CopernicusMarine open data,
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
ERA5 Reanalysis. Original model was also trained on ISSCP OLR satellite data, but we switched to the ERA5 TTR field because the later is available in near realtime.
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)
CNNs in a 3-level U-Net with all variables on a HEALPix mesh with 64x64 cells on each face. Model is extensively documented in the reference below. Aside from switching to TTR as a synthetic real-time substitute for satellite observed OLR, the only difference in our current model is the training loss function. We switched from RMSE to fair CRPS via an implementation similar to that in Lang et al., 2024 treatment of the AFIS
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
Cresswell‐Clay, N., Liu, B., Durran, D. R., Liu, Z., Espinosa, Z. I., Moreno, R. A., & Karlbauer, M. (2025). A deep learning Earth system model for efficient simulation of the observed climate. AGU Advances, 6, e2025AV001706.
https://doi. org/10.1029/2025AV001706
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?
OUR BIGGEST CHALLENGE FOR THIS FORECASTING EXERCISE WAS CORRECTLY EVALUATING THE RPSS CATEGORIES. OUR SUBMISSIONS WERE SERIOUSLY IN ERROR FOR ALL FORECASTS SUBMITTED PRIOR TO 9 OCTOBER. (Who would have thought this would be the biggest challenge...)
Are there any limitations to your current model that you aim to address in future iterations?
We are working on a v2 model with sea ice and upper ocean variables as well as more atmospheric levels (at present 30 atmospheric and 22 ocean variables, compared to 9 and 1 in the current model).
Are there any other AI/ML model components or innovations that you wish to highlight?
It's not at all new to the astronomers, but I always like to highlight the HEALPix mesh.
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.
Nathaniel Cresswell-Clay: data preparation, model architecture, model validation
Peter Harrington: data preparation, model architecture, model validation
Zachary Espinosa: data preparation, model architecture, model validation
David Pruitt: data preparation, model architecture, model validation
Paul Moreno: data preparation, model architecture, model validation
Akshay Subramaniam: model architecture, model validation
William Yik: data preparation, model architecture, model validation
Bowen Liu: data preparation, model architecture, model validation
Matthias Karlbauer: model architecture
Mike Pritchard: project management
Dale Durra: project management, model architecture
Model name
DLESyMS2Sv2
Model type
Number of individuals supporting model development:
6-10
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:
16-64
How would you best classify the IT system used for model development or forecast production:
High-Performance Computing (HPC) Cluster
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