Live Demo
Watch the platform in action — upload, pipeline run, KPI review, anomaly detection and human validation.
Project at a Glance
National precipitation series support hydrology, climate studies and operational forecasting. SIVEP automates detection while keeping climatologists in control.
27,679
Hourly observations (official DGM dataset)
97.5%
Temporal completeness
2
Pilot stations — Khemisset & Sidi Kacem
0.73
Quality score (hybrid_dgm mode)
Hybrid Architecture
A layered pipeline combining WMO/OMM regulatory QA/QC with machine learning — missing values and meteorological anomalies handled separately.
OMM RulesWMO regulatory foundation · 25%
MLQC-STSpatiotemporal ML · 25%
Isolation ForestOffline per station · 35%
Supervised RFWhen labels exist · 15%
hybrid_dgm0.65·s_pond + 0.35·s_vote · τ=0.52
Missing-value track
- NULL, gaps & PDF artifacts (0NULL)
- KNN imputation with spatial neighbours
- Separate from weather anomalies
Anomaly track
- Multi-detector fusion (hybrid_dgm)
- SHAP explainability for climatologists
- Human-in-the-loop — raw data never silently overwritten
Operational Dashboard
Seven Streamlit modules for climatologists — built with Python, PostgreSQL audit trail and institution-grade DGM theming.
Results on Official DGM Dataset
Evaluation on national precipitation data — hybrid_dgm mode selected for operational balance between detection coverage and institutional trust.
49.5%
Critical suspect rate (hybrid_dgm)
716
Temporal gaps detected
7
Missing values (Khemisset)
100%
Audit trail in PostgreSQL