Amberland
Members
This team has chosen to keep its participants anonymous.
Models
Model name
Amber1
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 Amber1
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?
- Machine learning-based weather prediction.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
We do not initialise.
If any, what data does your model rely on for real-time forecasting purposes?
None
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
We train the models using ERA5 data from 2000 till 2025. We arndomly sample 1000 locations each year.
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 train the model using [d-7; d-7-150] days, where d is the day of the year of the submission openning date plus longitude and latitude of locations. We use Random Forest.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
We have not published or presented any work related to this forecast.
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
To test model performance, we trained the model on years 2000-2020 and tested on data from years 2021-2024.
Did you face any challenges during model development, and how did you address them?
No
Are there any limitations to your current model that you aim to address in future iterations?
No
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.
This team has chosen to keep its participants anonymous.
Model name
Amber2
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 Amber2
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?
- Machine learning-based weather prediction.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
We do not initialise.
If any, what data does your model rely on for real-time forecasting purposes?
None
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
We train the models using ERA5 data from 2000 till 2025. We randomly sample 1000 locations each year.
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 train the model using [d-7; d-7-150] days, where d is the day of the year of the submission openning date plus longitude and latitude of locations. We use Random Forest. Finally, we rescale probabilities to align to climatological data.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
We have not published or presented any work related to this forecast.
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?
No
Are there any limitations to your current model that you aim to address in future iterations?
No
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.
This team has chosen to keep its participants anonymous.
Model name
Amber3
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 Amber3
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?
- Machine learning-based weather prediction.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
Climatological data is obtained from the ARCO-ERA5 full 37 level 1h dataset (https://github.com/google-research/arco-era5), which is then normalised to approximate zero mean and unit variance via random subsampling (for each data variable). The model is trained on week-long reanalysis observations, randomly chosen across the dataset, per batch.
If any, what data does your model rely on for real-time forecasting purposes?
The same ARCO-ERA5 data source for the most recent week-long observations.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
Reanalysis data, as outlined above.
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)
Main frameworks used: PyTorch, Prodigy plus Schedule Free (https://github.com/LoganBooker/prodigy-plus-schedule-free), Zuko (https://github.com/probabilists/zuko).
Latitude and longitude are mapped to spherical coordinates, which are then converted a cartesian 3-vector, and are concatenated with time (proportion of the week elapsed, normalised to the range -1,+1), and pressure level (log transformed and normalised to -1,+1).
A SIREN architecture is used to generate week-long observations (from Time x Level x Location-3-vector), which is modulated by a bias-only adaptation vector to fit the SIREN generator to a given week (L1 loss function). Additionally, an autoencoder is used to reduce the dimensionality of the adaptation vector.
Finally, a normalising flow network utilises a given (low-dimensional encoded) adaptation vector, as well nine additional metadata inputs (time between observation and prediction, divided by 6 weeks; sine and cosine of day/year offsets for the observation and prediction start datetimes) to predict future adaptation vectors in a probalistic manner. These predicted adaptation vectors are used to generate the final variable predictions.
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 initially used a direct low-lank adaptation vector, but this struggled to learn effectively, resulting in low-frequency outputs. By utilising a full-rank adaptation vector, compressed by a concurrently-trained autoencoder, we overcame this issue.
Are there any limitations to your current model that you aim to address in future iterations?
We plan to increase the number of iterations per adaptation cycle.
Are there any other AI/ML model components or innovations that you wish to highlight?
N/A
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.
This team has chosen to keep its participants anonymous.
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