🧠 Methodology & System Overview

This dashboard presents an integrated AI-driven system for the detection, quantification, and forecasting of methane emissions across Canada, using a hybrid, multi-source data fusion approach. The system combines satellite imagery, geospatial overlays, and temporal modeling to identify current hotspots and predict future methane concentrations.

📌 Data Sources

• Satellite Data (Sentinel-5P TROPOMI): Monthly gridded CH₄ (methane) concentration data collected between 2019 and 2024, processed at tile level resolution.
• Industrial Proximity & Land Use: Features derived from OpenStreetMap and provincial industrial registries, including distances to petroleum, mining, and agricultural zones.
• Environmental & Geographic Features: Data on elevation, land cover, and population density integrated using Google Earth Engine to enhance contextual learning.

Note: These datasets are not streamed in real time. They are periodically retrieved and processed in batch mode.

🧠 Modeling Approach

• Hotspot Detection: Anomaly detection using quantile-based thresholds and spatial clustering to flag high CH₄ zones.
• Forecasting: Supervised regression models (e.g., XGBoost, LSTM variants) trained to predict monthly methane concentration up to 3 years ahead.
• Feature Fusion: Spatial and temporal features are fused into structured tile-level datasets for localized and global prediction.

⚙️ System Architecture

• Preprocessing: Conducted in Google Earth Engine, with data exported for downstream modeling.
• Model Training: Performed offline using Python (scikit-learn, TensorFlow), with temporal validation splits.
• Dashboard Integration: Visualizations created using Plotly, Folium, and Leaflet.
• Deployment: This dashboard is a static web application; real-time data integration is not implemented.