Rainfall Forecasting App

The project forecasts monthly rainfall for March to October 2026 and estimates which days in a selected month have the highest chance of rainfall based on historical daily behavior.

Highlights

• compares multiple forecasting models with a time-aware validation setup
• selects the final model using validation metrics instead of test leakage
• saves forecasts, evaluation tables, figures, and model metadata
• serves the results through a clean Streamlit app
• keeps training dependencies separate from lightweight app deployment

Best Saved Model

• Model: BiLSTM
• Validation RMSE: 88.26
• Test RMSE: 110.92
• Forecast output: outputs/forecasts/future_forecasts.csv

Repository Layout

rainfall-timeseries-forecasting/
|-- app.py
|-- streamlit_app.py
|-- train.py
|-- evaluate.py
|-- forecasting_pipeline.py
|-- data_module.py
|-- daily_rainfall_profiles.py
|-- lightning_module.py
|-- models.py
|-- project_paths.py
|-- data/
|   |-- raw/
|   |-- processed/
|   `-- study_area/
|-- outputs/
|   |-- forecasts/
|   |-- evaluation/
|   `-- figures/
|-- artifacts/
|   |-- models/
|   |-- checkpoints/
|   `-- metadata/
|-- docs/
|-- scripts/
|-- .streamlit/
|-- requirements.txt
|-- requirements-training.txt
`-- runtime.txt

Folder Guide

• data/raw/: original daily rainfall dataset
• data/processed/: monthly dataset and daily climatology tables used by the app
• data/study_area/: geospatial boundary files for the study area
• outputs/forecasts/: saved future rainfall forecasts
• outputs/evaluation/: model comparison tables and backtest predictions
• outputs/figures/: charts generated by the forecasting pipeline
• artifacts/models/: trained model files saved after training
• artifacts/checkpoints/: sequence-model checkpoints
• artifacts/metadata/: selected-model metadata for the app and reports
• docs/: deployment notes and social sharing copy
• scripts/: supporting scripts, including the Google Earth Engine export script

Main Entry Points

• Local app: app.py
• App implementation: streamlit_app.py
• Training pipeline: train.py
• Evaluation summary: evaluate.py

Quick Start

For the Streamlit app only:

pip install -r requirements.txt
python -m streamlit run app.py

If Streamlit is installed in a different Python environment:

C:\ProgramData\Anaconda3\python.exe -m streamlit run app.py

For the full training and evaluation workflow:

pip install -r requirements-training.txt
python train.py
python evaluate.py

Important Outputs

• Forecasts: outputs/forecasts/future_forecasts.csv
• Evaluation metrics: outputs/evaluation/evaluation_metrics.csv
• Horizon-wise metrics: outputs/evaluation/evaluation_metrics_by_horizon.csv
• Backtest predictions: outputs/evaluation/backtest_predictions.csv
• Monthly dataset: data/processed/monthly_rainfall_dataset.csv
• Daily climatology: data/processed/daily_rainfall_climatology.csv
• Best model metadata: artifacts/metadata/best_model.json
• Forecast figure: outputs/figures/future_forecast_march_october_2026.png

Supporting Files

• Deployment guide: docs/DEPLOY_STREAMLIT_CLOUD.md
• LinkedIn post draft: docs/LINKEDIN_POST.md
• GEE export script: scripts/gee_export.js
• License: LICENSE