MAQMS — Air Quality Prediction Platform, Mexico City

B.Eng. capstone project at Instituto Politécnico Nacional, published December 2021. Four real stations — one bike-mounted mobile unit, three fixed — stream PM2.5/PM10 readings over WiFi/GSM and MQTT into Azure IoT Hub, through Stream Analytics into CosmosDB, feeding an LSTM model that forecasts short-term pollution trends per station.

The live demo above replays the project’s actual historical sensor data and lets you click through the real pipeline, not a reconstruction — see the source repo for the full developer guide, firmware, and data.

Overview

Street-level air quality isn’t something Mexico City’s official monitoring network measures — its stations sample at ~3m height. MAQMS adds a street-level layer: fixed stations at public/ institutional sites plus a bike-mounted mobile station, each publishing PM2.5/PM10 readings in real time.

Architecture

• Sensors: PM2.5/PM10 particulate sensor + GPS on an Arduino MKR 1010 (WiFi, fixed stations) or MKR GSM 1400 (cellular, the mobile station).
• Ingestion: MQTT over TLS to Azure IoT Hub, authenticated via an on-device ECCX08 self-signed certificate.
• Routing: Stream Analytics filters and routes each station’s messages to its own CosmosDB container.
• Prediction: A per-station LSTM model (TensorFlow/Keras), trained on a 60-reading lookback window, forecasts near-term PM2.5/PM10.

Publication

This work was later extended, with IPN’s Urban Data Lab and MIT’s Senseable City Lab, into a peer-reviewed chapter: Chávez Galván, C. I., et al. (2023). Enhancing Air Quality Monitoring in Mexico City: A Hybrid Sensor-Machine Learning System. In Telematics and Computing (CCIS), Springer Nature Switzerland, pp. 270–288. https://doi.org/10.1007/978-3-031-45316-8_18