Evaluating Machine Learning for BMKG Rainfall Prediction
- Client / Institution
- BMKG, Deputy for Climatology
- Year
- 2024
- Service Domain
- Corporate Training & AI Analytics
A rainfall modelling study and time series training programme for BMKG, comparing predictor scenarios and identifying directions for further method development.
BMKG wanted to evaluate the use of machine learning to predict rainfall from historical data. During requirements discussions, its team also asked that model evaluation reflect the two-year forecasting pattern used in its work.
Sadasa conducted a modelling study using historical rainfall data and the ENSO and IOD climate indices. The team compared three predictor scenarios: rainfall history combined with climate indices, climate indices alone, and rainfall history alone. Models were evaluated by comparing predictions with observed values. On 23–24 September 2024, Sadasa also delivered time series and machine learning training using rainfall data as a practical example.
The final report compared the models and identified limitations in the tested variables for detailed rainfall prediction. It recommended considering additional weather variables and further time series methods. The documented outputs comprise an evaluation report, modelling results, and training materials to support subsequent analysis.
Institution names are referenced as a factual record of completed engagements. Client proprietary materials, logos, and sensitive data remain strictly protected.