SAIS2S
Members
This team has chosen to keep its participants anonymous.
Model
Model name
SAI
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 SAI
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)
The real-time model runs with Spire’s proprietary initial condition, SOF-D (Spire Operational Forecast–Deterministic), which integrates open-source and proprietary observations through Spire’s in-house data assimilation system.
If any, what data does your model rely on for real-time forecasting purposes?
As mentioned earlier, it depends on SOF-D data.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
The model is trained on processed ERA5 data from 1979 to 2017, and validated on 2018. Following training
the model on ERA5 data, the model is finetuned on SOF-D from April 2022 to July 2023.
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 employs a computationally efficient autoregressive CNN-CVAE for probabilistic S2S forecasting, trained on 0.5° daily ERA5 data using four 80 GB A100 GPUs. It produces 201 ensemble members for 45-day forecasts, addressing key challenges such as incorporating slow-varying boundary forcings (SST, soil moisture), managing prime-meridian discontinuities, mitigating spherical artifacts, and introducing a novel ensemble generation method to reduce sampling noise.
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?
Yes, the model is validated against EAR5 data and benchmarking with ECMWF's S2S data.
Did you face any challenges during model development, and how did you address them?
We addressed two major challenges—data discontinuity at the prime meridian and artifacts from the spherical data structure—through the following solutions:
Prime Meridian Discontinuity:
Padding width: We introduced a geocyclic padding hyperparameter (padw) to handle East–West boundary discontinuities, finding that padw ≤ 2 is essential.
Convolution type: We used ‘valid’ padding for all convolution operations, as the commonly used ‘same’ padding introduces artificial continuity issues at data edges.
Spherical Artifacts:
Large receptive field: To mitigate artifacts while keeping padw ≤ 2, we used dilated convolutions, which expand the receptive field without increasing kernel size.
Controlled training: We avoided overfitting by monitoring the training–validation error gap and applying early stopping when divergence began.
Are there any limitations to your current model that you aim to address in future iterations?
We will incorporate "coupling" in the current model in the next version where two AI model (one for Atmosphere and one for Ocean) will train and couple it.
Are there any other AI/ML model components or innovations that you wish to highlight?
The solution for the Spherical Artifacts should be highlighted as lots of AI model is struggling with this.
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