Blazers
Members
First name (team leader)
Collins
Last name
Asega
Organisation name
African Center of Meteorological Applications for Development
Organisation type
Meteorological Institution
Organisation location
Niger
First name
Pierre
Last name
Kamsu
Organisation name
African Center of Meteorological Applications for Development
Organisation type
Meteorological Institution
Organisation location
Niger
First name
Godefroy
Last name
Nshimirimana
Organisation name
African Center of Meteorological Applications for Development
Organisation type
Meteorological Institution
Organisation location
Niger
First name
Leon
Last name
Guy
Organisation name
African Center of Meteorological Applications for Development
Organisation type
Meteorological Institution
Organisation location
Niger
First name
Ousmane
Last name
Ndiaye
Organisation name
African Center of Meteorological Applications for Development
Organisation type
Meteorological Institution
Organisation location
Niger
First name
Patrick
Last name
Kinyua
Organisation name
ACMAD
Organisation type
Meteorological Institution
Organisation location
Niger
Model
Model name
umojaForecast
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 umojaForecast
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.
- Machine learning-based weather prediction.
- Statistical model focused on generating quintile probabilities.
- 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)
he model is initialised using ERA5 reanalysis data provided through the AI Quest package. Weekly aggregated variables: tas, mslp, and pr are retrieved via the official data-access API. Each variable is normalised against its multi-year climatology (2005–2020) to compute anomalies, which serve as input features for the regression models.
Forecast initialisation corresponds to the current competition forecast start date (Thursday), where recent ERA5 conditions (lags of up to four weeks) are used to generate feature fields for each variable. These are then standardised using pre-computed climatological mean and standard-deviation fields before prediction.
If any, what data does your model rely on for real-time forecasting purposes?
For real-time forecasting, the model relies solely on ERA5 reanalysis data accessed through the ECMWF AI Quest data interface. The most recent available weekly ERA5 fields for tas, mslp, and pr are automatically downloaded and used to compute the most recent anomalies and lag features relative to their climatologies.
No external or proprietary real-time datasets are used all inputs originate from the official ERA5 archive provided by ECMWF within the AI Quest environment.
What types of datasets were used for model training? (For example: observational datasets, reanalysis data, NWP outputs or satellite data)
The model was trained exclusively on ERA5 reanalysis datasets provided through the ECMWF AI Quest data portal. Weekly aggregated fields of tas, mslp, and pr were used from the period 2005–2020.
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)
To start with, we have used a simple regression model and this is the overview of Umojaforecast.
Model type: Lightweight statistical / machine-learning model (per variable: tas, mslp, pr)
Algorithm: Ridge Regression (scikit-learn)
Data source: ERA5 reanalysis (2005–2020) via AI Quest package
Preprocessing: Compute weekly climatology (mean & std) per week-of-year. Standardize ERA5 data to anomalies (z-scores). Build lag features (own-variable and cross-variable, up to 4 weeks back)
Features: 8 per grid cell (4 own lags + 4 cross-variable lags)
Training target: Weekly anomaly at lead week (3–4 weeks ahead)
Prediction output: Mean anomaly field
Uncertainty estimation: Residual standard deviation from training
Post-processing: Convert Mean anomaly and residual standard deviation to quintile probabilities using Normal CDF and then normalize probabilities per grid to sum to 1
Frameworks used: Python, NumPy, xarray, scikit-learn, SciPy, AI_WQ_package
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 external observational or independent datasets were used for validation.
Did you face any challenges during model development, and how did you address them?
Yes. Several practical challenges were encountered: Package compatibility and forecast inizialization (i.e sticking to current start date).
Are there any limitations to your current model that you aim to address in future iterations?
Yes, we intend to use a deep neural network to compare the results
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.
Pierre Kamsu
Nshimirimana Godefroid
Leon Guy
Ousmane Ndiaye
Patrick Kinyua
Collins Asega