LP
Member
First name (team leader)
Peng
Last name
Lu
Organisation name
Jiangsu Climate Center
Organisation type
Meteorological Institution
Organisation location
China
Model
Model name
LPM
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 LPM
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.
- Statistical model focused on generating quintile probabilities.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
We used ECMWF subseasonal forecast outputs as the main input to initialise the model. Two parallel approaches were applied: one combines the raw model forecasts with thresholds derived from hindcast data, and the other combines bias-corrected forecasts with thresholds derived from reanalysis data. The outputs from these two approaches were then merged through weighted averaging to produce quintile probability forecasts.
If any, what data does your model rely on for real-time forecasting purposes?
For real-time forecasting, the model relies primarily on operational ECMWF subseasonal forecast outputs. It also uses historical hindcast data and historical reanalysis data to calibrate and correct the forecasts, and to construct the thresholds required for probabilistic prediction.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
The method uses ECMWF hindcast datasets and reanalysis data. Hindcast data are used to derive model-based thresholds, while reanalysis data are used to derive observational thresholds. These datasets are used for statistical calibration and post-processing.
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 approach is a statistical post-processing framework. It consists of two branches: (1) combining raw forecasts with hindcast-based thresholds, and (2) combining bias-corrected forecasts with reanalysis-based thresholds. The probabilistic outputs from both branches are then fused using weighted averaging to generate the final quintile probability forecasts. The key components include threshold construction, forecast calibration, and probabilistic fusion.
Have you published or presented any work related to this forecasting model? If yes, could you share references or links?
https://aiweatherquest.ecmwf.int/wp-content/uploads/2026/03/DJF-Awards-presentation.pdf
Before submitting your forecasts to the AI Weather Quest, did you validate your model against observational or independent datasets? If so, how?
yes
I use AI_WQ_package to evaluate.
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?
We will further improve the machine learning model to surpass the climatological baseline and incorporating it into the later stages of the competition
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.
the model is developed by Lu Peng.
Which of the following descriptions best represent the overarching design of your forecasting model?
- Post-processing of numerical weather prediction (NWP) data.
- Statistical model focused on generating quintile probabilities.
What techniques did you use to initialise your model? (For example: data sources and processing of initial conditions)
For the post-processing of Numerical Weather Prediction (NWP), ensemble forecast data from the ECMWF sub-seasonal to seasonal (S2S) dynamical model are directly utilized.
If any, what data does your model rely on for real-time forecasting purposes?
For the post-processing of Numerical Weather Prediction (NWP), ensemble forecast data from the ECMWF sub-seasonal to seasonal (S2S) dynamical model are directly utilized.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
It is a training-free, computationally efficient, and non-parametric post-processing framework
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 Lightweight Probabilistic Model (LPM), a training-free, computationally efficient, and non-parametric post-processing framework that generates probabilistic forecasts by blending dural reference thresholds derived from both hindcast and reanalysis data.
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
I use AI_WQ_package to evaluate.
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?
We will further improve the machine learning model to surpass the climatological baseline and incorporating it into the later stages of the competition
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.
the model is developed by Lu Peng.
Which of the following descriptions best represent the overarching design of your forecasting model?
- Post-processing of numerical weather prediction (NWP) data.
- Machine learning-based weather prediction.
- 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)
For the Post-processing of Numerical Weather Prediction (NWP) method, the ensemble forecast data from the ECMWF sub-seasonal to seasonal (S2S) dynamical model is directly employed.
For the Machine Learning-based Weather Prediction method, the weekly observational data from the A1_WQ_package is utilized.
If any, what data does your model rely on for real-time forecasting purposes?
For the Post-processing of Numerical Weather Prediction (NWP) method, the ensemble forecast data from the ECMWF sub-seasonal to seasonal (S2S) dynamical model is directly employed.
For the Machine Learning-based Weather Prediction method, the weekly observational data from the A1_WQ_package is utilized.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
For the Post-processing of Numerical Weather Prediction (NWP) method, the historical reforecast data from the ECMWF sub-seasonal to seasonal (S2S) dynamical model is utilized.
For the Machine Learning-based Weather Prediction method, the historical training data from the A1_WQ_package is employed.
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 LPM model is a hybrid that combines the post-processing of the ECMWF sub-seasonal dynamical forecast with outputs from a machine learning model. However, due to the current underperformance of the machine learning component, the model primarily relies on results from the dynamical model 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?
yes
I use AI_WQ_package to evaluate.
Did you face any challenges during model development, and how did you address them?
When traning the deep learning model, the result is not good, so in the SON phase in the hybrid model the weight of deep learning model is 0.
Are there any limitations to your current model that you aim to address in future iterations?
We will further improve the machine learning model to surpass the climatological baseline and incorporating it into the later stages of the competition
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.
the model is developed by Lu Peng.
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