Welcome to the Graph4Air wiki!

🌍 Graph4Air - Air Pollution Mapping with Graph Neural Networks

📘 Project Overview

Graph4Air is a collaborative research project aimed at constructing high-resolution air pollution maps using Graph Neural Networks (GNNs). The data consists of road-level air pollution measurements collected every 50 meters, enriched with contextual features such as traffic density and land use type.

Our focus is on graph design rather than neural architecture, meaning we investigate how different node and edge construction strategies affect the performance of the spatial map generation.

---

👥 Team Members

• David Labee – Graph Design A
• Pieter Noordam – Graph Design B
• Zhendong Yuan – Supervisor
• Jules Kerckhoffs – Supervisor

---

🎯 Project Goals

• Develop different graph construction methods to represent spatial data.
• Integrate external environmental and infrastructural features.
• Apply and compare GNN-based models to generate detailed pollution maps.
• Evaluate the effects of graph structure on final output.

---

🗂 Wiki Structure

This Wiki is divided into the following sections:

• Literature Overview : Papers and reviews relevant to air pollution modeling with GNNs.
• Data Exploration : Analysis and visualization of the raw data.
• Graph Construction Methods : Separate designs from David and Pieter.
• Model Architecture : Shared neural architectures used with each graph.
• Evaluation & Visualization : Metrics and map outputs from our experiments.
• Meeting notes : Notes of the weekly meetings with the team members