A Simple and Robust Multi-Fidelity Data Fusion Method for Effective Modeling of Citizen-Science Air Pollution Data

2025-11-19

AI Summary

This study addresses the challenge of fusing sparse, high-accuracy regulatory PM₂.₅ measurements with dense, low-quality citizen-science sensor data in urban air quality modeling. We propose a multi-fidelity Gaussian process framework that introduces a global Huber loss function to bound the influence of noisy observations, augmented by accuracy-weighted residuals, co-kriging coupling, and whitening for enhanced cross-fidelity robustness. We theoretically derive stability bounds for the estimator. Evaluated on real-world data from Hamburg, our method achieves significant improvements in prediction accuracy (reduced MAE and RMSE), produces physically consistent uncertainty maps, and maintains stable performance under increasing outlier prevalence. The framework is scalable, fully open-source, and reproducible.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Intelligent Robots: State Estimation

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Web Mining and Content Analysis: Web data integration and cleaning
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We propose a robust multi-fidelity Gaussian process for integrating sparse, high-quality reference monitors with dense but noisy citizen-science sensors. The approach replaces the Gaussian log-likelihood in the high-fidelity channel with a global Huber loss applied to precision-weighted residuals, yielding bounded influence on all parameters, including the cross-fidelity coupling, while retaining the flexibility of co-kriging. We establish attenuation and unbounded influence of the Gaussian maximum likelihood estimator under low-fidelity contamination and derive explicit finite bounds for the proposed estimator that clarify how whitening and mean-shift sensitivity determine robustness. Monte Carlo experiments with controlled contamination show that the robust estimator maintains stable MAE and RMSE as anomaly magnitude and frequency increase, whereas the Gaussian MLE deteriorates rapidly. In an empirical study of PM2.5 concentrations in Hamburg, combining UBA monitors with openSenseMap data, the method consistently improves cross-validated predictive accuracy and yields coherent uncertainty maps without relying on auxiliary covariates. The framework remains computationally scalable through diagonal or low-rank whitening and is fully reproducible with publicly available code.

Problem

• Integrating sparse high-quality monitors with dense noisy citizen-science sensors
• Addressing contamination sensitivity in Gaussian process data fusion
• Improving air pollution prediction accuracy using multi-fidelity robust modeling

Innovation

• Multi-fidelity Gaussian process with Huber loss
• Robust estimator with bounded influence parameters
• Computationally scalable whitening for data fusion