Three Distributional Approaches for PM10 Assessment in Northern Italy

2025-09-17

AI Summary

This study addresses the limitation of conventional mean-based metrics in characterizing spatial heterogeneity and exceedance risk in PM₁₀ concentration assessment across Northern Italy. We propose a fine-scale spatial modeling framework grounded in the full probability distribution of PM₁₀. Methodologically, we innovatively integrate compositional data prediction, non-crossing quantile regression, and simplex principal component analysis (SPCA), coupled with fixed-rank kriging and differential regularization spatial regression to enable multi-path joint estimation of PM₁₀ probability density functions. Applied to monitoring data from 2018–2022, the approach generates high-resolution, continuous spatial distribution maps. All three methodological pathways yield highly consistent macro-level pollution patterns, demonstrating the robustness and interpretability of distributional modeling for identifying exceedance hotspots and informing differentiated environmental policy interventions.

Technology Category

• Data Mining & Knowledge Management: Mining of Spatial, Temporal or Spatio-Temporal Data
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

We propose three spatial methods for estimating the full probability distribution of PM10 concentrations, with the ultimate goal of assessing air quality in Northern Italy. Moving beyond spatial averages and simple indicators, we adopt a distributional perspective to capture the complex variability of pollutant concentrations across space. The first proposed approach predicts class-based compositions via Fixed Rank Kriging; the second estimates multiple, non-crossing quantiles through a spatial regression with differential regularization; the third directly reconstructs full probability densities leveraging on both Fixed Rank Kriging and multiple quantiles spatial regression within a Simplicial Principal Component Analysis framework. These approaches are applied to daily PM10 measurements, collected from 2018 to 2022 in Northern Italy, to estimate spatially continuous distributions and to identify regions at risk of regulatory exceedance. The three approaches exhibit localized differences, revealing how modeling assumptions may influence the prediction of fine-scale pollutant concentration patterns. Nevertheless, they consistently agree on the broader spatial patterns of pollution. This general agreement supports the robustness of a distributional approach, which offers a comprehensive and policy-relevant framework for assessing air quality and regulatory exceedance risks.

Problem

• Estimating full PM10 probability distributions spatially
• Assessing air quality and regulatory exceedance risks
• Comparing three distributional methods for Northern Italy

Innovation

• Fixed Rank Kriging for class compositions
• Spatial regression with differential regularization
• Simplicial Principal Component Analysis framework