AI-Driven Autocalibration of a Heterogeneous Sensor Network

The "AI-Driven Autocalibration of a Heterogeneous Sensor Network" project addresses a critical challenge in environmental monitoring: the need to calibrate low-cost sensors, which are widely used in Citizen Science due to their affordability but often suffer from inaccuracies caused by environmental and structural factors. The project builds on the SmartAQnet initiative, led by KIT TECO, which established a unique, open dataset of air quality measurements collected over nearly three years in Augsburg. This dataset includes data from a dense, heterogeneous network of sensors—spanning low-cost devices to official government stations—providing an unprecedented resource for research.

Leveraging this dataset, the project investigates AI-supported remote calibration techniques to automatically calibrate low-cost sensors. By using a underlying, sparser network of high-quality reference stations, the goal is to enable cities (Smart Cities) to deploy affordable sensors for continuous, real-time air quality monitoring. The project focuses on machine learning interpolation methods, comparing various spatial-temporal machine learning techniques to validate their effectiveness in producing reliable calibration results.

The project aims to demonstrate that automated, mobile calibration of low-cost sensors is feasible when supported by a dense network of reference stations. Its primary goal is to prove that AI-driven interpolation methods can produce qualitatively valid results without requiring manual calibration using expensive reference equipment. By validating this approach, the project seeks to enable Smart Cities to adopt low-cost sensor networks for comprehensive air quality monitoring, with the assurance that these sensors remain accurate through continuous, automated calibration. The ultimate vision is to bridge the gap between cost-effective Citizen Science applications and the need for reliable, real-time environmental data.