AirQualityMap is an air quality prediction system that downscales regional pollution data to street-level resolution across Catalonia, developed for the "Fem visible l'invisible!" challenge by Barcelona Supercomputing Center.

AirQualityMap transforms low-resolution (1km×1km) air quality predictions from the CALIOPE system into highly detailed street-level pollution maps. By combining advanced spatial interpolation techniques with machine learning models that incorporate traffic patterns and local measurements, we provide citizens with accurate, hyper-local air quality forecasts.

• CALIOPE System Data: 1km×1km resolution NO₂ pollution predictions across Catalonia
• XVPCA Stations: Ground-truth air quality measurements from land monitoring stations
• Traffic Patterns: Road network data with traffic intensity information
• Temporal Factors: Historical patterns of pollution variation by hour and day

• Implemented Inverse Distance Weighting (IDW) to create a baseline high-resolution pollution map
• Interpolated values from CALIOPE grid points to target prediction areas
• Applied distance-decay functions to weight the influence of nearby measurement points

• Extracted features from road networks near air quality stations
• Trained a Random Forest Regressor to adjust pollution estimates based on:
  • Road width and type
  • Traffic intensity
  • Distance to major pollution sources
• Performed feature importance analysis to identify key pollution predictors

• Trained a LightGBM model to account for:
  • Hourly pollution variations (rush hour patterns)
  • Daily patterns (weekday vs. weekend differences)
  • Seasonal trends
• Created adjustment factors to modify base predictions according to specific forecast times

• Employed cross-validation techniques to ensure prediction accuracy
• Validated against held-out XVPCA station data
• Optimized hyperparameters for both spatial and temporal components

• Created an intuitive interface for accessing high-resolution pollution maps
• Implemented city selection for major Catalan urban areas
• Developed interactive visualization of pollution levels with color-coded indicators
• Enabled time-based forecasting capabilities

• Data Processing: Python, Pandas, NumPy, GeoPandas
• Machine Learning: Scikit-learn, Random Forest Regressors
• Geospatial Analysis: PostGIS, QGIS
• Visualization: Leaflet.js, D3.js
• Web Development: Flask, HTML/CSS, JavaScript

• Successfully downscaled 1km² resolution data to street-level detail
• Created an accurate predictive model despite limited training data
• Developed a system capable of real-time air quality forecasting
• Produced intuitive visualizations that make pollution patterns easily understandable for citizens

• Sergi Flores
• Clàudia Gallego
• Weihao Lin
• Jiahui Chen

This project directly supports La Marató de TV3's efforts in fighting respiratory diseases by:

• Raising awareness about air pollution and its health impacts
• Providing citizens with actionable information about local air quality
• Enabling better decision-making about outdoor activities and travel routes
• Supporting public health initiatives through improved environmental monitoring

Special thanks to:

• Barcelona Supercomputing Center for providing the CALIOPE system data and expert guidance
• BitsxLaMarató organization team for creating this meaningful hackathon experience
• La Marató de TV3 for their ongoing work in the fight against respiratory diseases