How udithkumar-tech/meghova works
Meghova is a regime-based post-processor for numerical weather prediction (NWP) rainfall. Its principal workflow trains a weather-regime classifier and regime-specific rainfall-correction experts, then applies them to future NWP inputs to produce corrected rainfall predictions. Separate evaluation scripts compare predictions with observations using error and rainfall-event metrics. The sampled code supports these training, inference, and evaluation links; the README describes the product only briefly, so unsampled details are not inferred.
Main entry points: future_prediction.py Future prediction · meghova_training_data.csv Training observations · time_based_evaluation.py Time-based evaluation · regime_classifier.pkl Regime classifier and encoder
Components 18
Model training
- Train regime classifier
ML/regime_classifier.py - Train regime experts
ML/train_experts.py
Prediction workflow
- Future prediction
ML/future_prediction.py
Forecast verification
- Match and evaluate 2026
ML/evaluate_2026.py - 2026 rainfall metrics
ML/evaluate_2026_all_metrics.py - Historical evaluation
ML/evaluate_meghova.py - Time-based evaluation
ML/time_based_evaluation.py - Time-based event metrics
ML/time_based_rainfall_metrics.py - Final forecast assessment
ML/final_evaluation.py
Data and models
- Training observations
ML/data/meghova_training_data.csv - Regime classifier and encoder
ML/regime_classifier.pkl - Regime expert models
ML/experts - Future NWP input
ML/data/future_nwp_input.csv - Corrected forecasts
ML/meghova_future_predictions.csv - 2026 observations
ML/data/karnataka_nwp_forecast_2026 (1).csv - 2026 comparison data
ML/results/meghova_2026_evaluation.csv - 2026 event metrics
ML/results/meghova_2026_all_metrics.csv
Other components
- Forecast operator
Connections 20
- Forecast operator → Future prediction: runs forecast
- Training observations → Train regime classifier: trains from
- Train regime classifier → Regime classifier and encoder: saves classifier
- Training observations → Train regime experts: trains from
- Train regime experts → Regime expert models: saves experts
- Future NWP input → Future prediction: supplies NWP
- Regime classifier and encoder → Future prediction: classifies regimes
- Regime expert models → Future prediction: correct rainfall
- Future prediction → Corrected forecasts: writes forecasts
- Corrected forecasts → Match and evaluate 2026: provides predictions
- 2026 observations → Match and evaluate 2026: provides observations
- Match and evaluate 2026 → 2026 comparison data: saves comparison
- 2026 comparison data → 2026 rainfall metrics: provides comparison
- 2026 rainfall metrics → 2026 event metrics: saves metrics
- Training observations → Historical evaluation: provides test data
- Regime classifier and encoder → Historical evaluation: loads classifier
- Regime expert models → Historical evaluation: loads experts
- Training observations → Time-based evaluation: provides time series
- Time-based evaluation → Time-based event metrics: writes test results
- Time-based evaluation → Final forecast assessment: writes test results