CMAandFDU
Members
First name (team leader)
Bo
Last name
Lu
Organisation name
China Meteorological Administration
Organisation type
Meteorological Institution
Organisation location
China
First name
Hao
Last name
Li
Organisation name
Fudan University
Organisation type
Research Organisation (Academic, Independent, etc.)
Organisation location
China
First name
Lei
Last name
Chen
Organisation name
Fudan University
Organisation type
Research Organisation (Academic, Independent, etc.)
Organisation location
China
First name
Xiaohui
Last name
Zhong
Organisation name
Fudan University
Organisation type
Research Organisation (Academic, Independent, etc.)
Organisation location
China
First name
Jie
Last name
Wu
Organisation name
China Meteorological Administration
Organisation type
Meteorological Institution
Organisation location
China
First name
Chunyan
Last name
Zhao
Organisation name
China Meteorological Administration
Organisation type
Meteorological Institution
Organisation location
China
First name
Chenguang
Last name
Zhou
Organisation name
China Meteorological Administration
Organisation type
Meteorological Institution
Organisation location
China
First name
Yang
Last name
Zhao
Organisation name
China Meteorological Administration
Organisation type
Meteorological Institution
Organisation location
China
First name
Jiahui
Last name
Hu
Organisation name
China Meteorological Administration
Organisation type
Meteorological Institution
Organisation location
China
First name
Yuhang
Last name
Xin
Organisation name
China Meteorological Administration
Organisation type
Meteorological Institution
Organisation location
China
Models
Model name
Fengshun
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
Model summary questionnaire for model Fengshun
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)
Daily ERA5 data with multiple variables is used to initialise Fengshun. The z-score normalization technique is employed to normalize all input and output variables.
If any, what data does your model rely on for real-time forecasting purposes?
Daily ERA5 was used for real-time forecasting, although with several days delay.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
Only daily ERA5 was used for training.
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)
Fengshun follows the architecture of previous published FuXi-S2S model, consists of three primary components: an encoder P, a perturbation module, and a decoder. The encoder, processing predicted weather parameters from two preceding time steps, with each time step representing l day. Specifically, it takes inputs into a two-dimensional convolution layer with a kernel size of two, which reduces the dimensions of the input data by half. Following this, the hidden feature is derived from 12 repeated transformer blocks. The input to the encoder is a data cube that combines both upper-air and surface variables. These dimensions represent two preceding time steps. To account for the accumulation of forecast error overtime, the forecast lead time is also included in the encoder's input. The encoder also generates a low-rank multivariate Gaussian distribution. Intermediate perturbation vectors are sampled from this Gaussian distribution. These vectors, after being weighted by a learned weight vector, yield the final perturbation vectors. The decoder then processes the perturbed hidden features through 24 transformer blocks and a fully connected layer, resulting in the final ensemble output. The number of ensemble members generated equals the number of samples drawn from the Gaussian distribution.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
https://doi.org/10.1038/s41467-024-50714-1
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 quintile is adopted in this competition, which is different from the usually-used MSE loss in AI models. We are still struggling to find a better way to address this challenge.
Are there any limitations to your current model that you aim to address in future iterations?
The detailed spatial information seems to be smoothed in Fengshun output.
Are there any other AI/ML model components or innovations that you wish to highlight?
To update a climate model, large historical hindcast is needed to generate quintile climatology in model space, which takes more time and storage resources. Here, the submitted Fengshun is trying to predict quintile directly.
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.
Overall technical roadmap: LI Hao, LU Bo.
Model architecture: CHEN Lei, DOU Zesheng.
Data preparation: ZHOU Chenguang, WANG Chenpeng, WU Jie
Model Validation: HU Jiahui, Zhong Xiaohui, ZHAO Yang, QIAN Qifeng
Computational Resource Optimization: ZHAO Chunyan, XIN Yuhang
Model name
FengshunAdjust
Number of individuals supporting model development:
6-10
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:
4-16
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 FengshunAdjust
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.
- 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)
Same as Fengshun, daily ERA5 data with multiple variables is used to initialise Fengshun_Adjust.
If any, what data does your model rely on for real-time forecasting purposes?
Daily ERA5 was used for real-time forecasting, although with several days delay.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
Only daily ERA5 was used for training.
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 architecture for Fengshun_Adjust is similar with Fengshun. Multiple versions of the Fengshun models with different parameter settings were utilized for Fengshun_Adjust. Then a Swin-Transformer is used to extract features from their outputs and generate an ensemble forecast by Fengshun_Adjust.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
https://doi.org/10.1038/s41467-024-50714-1
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 ERA5 data we were able to obtain was delayed by 6 days, which led to performance degradation on a real-time basis.
Are there any limitations to your current model that you aim to address in future iterations?
The forecast skill of precipitation is still limited for Fengshun_Adjust, and we are still trying to improve the algorithm of precipitation forecast.
Are there any other AI/ML model components or innovations that you wish to highlight?
The AI-based ensemble of various AL/ML models is useful for uncertainty analysis.
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.
Overall technical roadmap: LI Hao, LU Bo.
Model architecture: ZHAO Yang, DOU Zesheng, HU Jiahui, QIAN Qifeng.
Data preparation: ZHOU Chenguang, WANG Chenpeng, WU Jie
Model Validation: ZHAO Yang, HU Jiahui, Zhong Xiaohui, QIAN Qifeng
Computational Resource Optimization: ZHAO Chunyan, XIN Yuhang
Model name
FengshunHybrid
Number of individuals supporting model development:
6-10
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:
4-16
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 FengshunHybrid
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?
- Hybrid model that integrates physical simulations with machine learning or statistical techniques.
- 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)
Daily ERA5 data is used, with the z-score normalization technique employed. In addition, the hindcast of ECMWF S2S numerical model is also utilized.
If any, what data does your model rely on for real-time forecasting purposes?
Daily ERA5 data and ECMWF S2S operational model (CY49R1).
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
Daily ERA5 and hindcast of ECMWF S2S numerical model.
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 architecture of Fengshun_Hybrid is similar with Fengshun_Adjust. The difference is that Fengshun_Hybrid includes the ECMWF S2S operational model, while Fengshun_Adjust only considers the data-driven AI/ML models.
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 ERA5 data we were able to obtain was delayed by 6 days, and the dynamic model was delayed by almost 3 days. This time lag caused some trouble for training, but we ended up aligning all the data.
Are there any limitations to your current model that you aim to address in future iterations?
How to balance the outputs of AI model and dynamic model is a question to be considered in the future.
Are there any other AI/ML model components or innovations that you wish to highlight?
The hybrid use of AI/ML and NWP models.
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.
Overall technical roadmap: LI Hao, LU Bo.
Model architecture: ZHAO Yang, DOU Zesheng, HU Jiahui, QIAN Qifeng.
Data preparation: ZHOU Chenguang, WANG Chenpeng, WU Jie
Model Validation: ZHAO Yang, HU Jiahui, Zhong Xiaohui, QIAN Qifeng
Computational Resource Optimization: ZHAO Chunyan, XIN Yuhang
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