Air quality forecasting: predictive modeling to enhance sustainability, health, and mobility

The project utilized affordable meteorological and air quality sensors, a computer vision-enabled camera system for vehicle counting and classification, and a web-based platform to monitor and analyze environmental and traffic data.

A comprehensive solution was required to monitor and forecast air pollution from mobile sources, provide accurate air quality predictions, and enable effective mobility management through predictive models and real-time environmental monitoring systems. The goal was to enhance sustainable mobility strategies, improve public health, and ensure compliance with national air quality standards in metropolitan areas facing significant challenges related to air pollution and traffic congestion.

The tasks involved planning, data collection, model development, and evaluation. We faced challenges such as integrating various data sources (e.g., air quality analyzers, meteorological stations, computer vision-enabled cameras), ensuring data accuracy, and developing a predictive model. The terms of reference required detailed phases for planning, development, and model calibration, as well as extensive data collection and technology integration requirements.

Two teams of experts were involved: one specializing in data management and the other in software development, and there were six phases:

• Planning (Conceptualizing the model; Defining variables and technologies; Designing architecture and the ML life cycle)
• Installation of Equipment and Sensors (Validating and testing sensors; Supporting field installations; Validating data transmission)
• Model Development (Reviewing existing algorithms/models; Collecting data and historical information; Conducting exploratory data analysis; Processing data; Training the model; Validating the model; Optimizing hyperparameters)
• Model Calibration and Evaluation (Developing APIs; Validating model predictions)
• Implementation and Deployment (Implementing and deploying the model; Deploying the forecasting web platform; Testing the platforms)
• Documentation (Documenting algorithms; Conducting training and workshop sessions for end users)

Our services included:

• Implementation of Traffic and Air Quality Stations: Configuring, integrating, and installing weather and air quality sensors and computer vision-enabled camera system for vehicle counting and classification to monitor road traffic.
• Development of Predictive Models Using Machine Learning: Data analysis and processing, along with the development, testing, and evaluation of air quality, meteorology, and vehicle traffic forecasting models using LSTM neural networks.
• Development of a Web Platform: Design, development, testing, and documentation of a web platform for sustainable mobility and emissions inventory.
• Cloud Infrastructure Management: Provisioning and support of AWS cloud services for data storage and hosting the web platform.
• Implementation of Self-Developed Technologies: Utilization of proprietary technologies like AmbiensQ and established data science standards for analyzing and processing sensor datasets. We employed machine learning to create predictive models for air quality and traffic forecasting. Additionally, we developed software to visualize and report data on the web platform, integrated various weather and air quality sensors with traffic cameras, and applied computer vision techniques to analyze traffic patterns.

Our differentiation was based on a comprehensive and integrated approach through the proprietary  AmbiensQ platform. This platform centralizes environmental data management and leverages technologies such as cloud computing, data analytics, map visualization, and machine learning. AmbiensQ Suite integrates data from a wide range of sensors and includes data validation features that ensure real-time monitoring and predictive capabilities.