[Paper Review] A Bayesian Non-linear State Space Copula Model to Predict Air Pollution in Beijing

Alexander Kreuzer, Luciana Dalla Valle+1 more|arXiv (Cornell University)|Mar 20, 2019

Air Quality and Health Impacts29 references3 citations

TL;DR

This paper proposes a novel Bayesian non-linear state space copula model to predict PM2.5 levels in Beijing using hourly meteorological and pollution data from 2014. By modeling both observation and state equations via copulas, it captures non-Gaussian, non-linear dependencies and enables accurate forecasting and scenario-based climate impact analysis, outperforming standard Gaussian state space models in capturing extreme pollution events.

ABSTRACT

Air pollution is a serious issue that currently affects many industrial cities in the world and can cause severe illness to the population. In particular, it has been proven that extreme high levels of airborne contaminants have dangerous short-term effects on human health, in terms of increased hospital admissions for cardiovascular and respiratory diseases and increased mortality risk. For these reasons, accurate estimation and prediction of airborne pollutant concentration is crucial. In this paper, we propose a flexible novel approach to model hourly measurements of fine particulate matter and meteorological data collected in Beijing in 2014. We show that the standard state space model, based on Gaussian assumptions, does not correctly capture the time dynamics of the observations. Therefore, we propose a non-linear non-Gaussian state space model where both the observation and the state equations are defined by copula specifications, and we perform Bayesian inference using the Hamiltonian Monte Carlo method. The proposed copula state space approach is very flexible, since it allows us to separately model the marginals and to accommodate a wide variety of dependence structures in the data dynamics. We show that the proposed approach allows us not only to predict particulate matter measurements, but also to investigate the effects of user specified climate scenarios.

Motivation & Objective

• To address the limitations of standard Gaussian state space models in capturing non-linear and non-Gaussian dynamics in hourly PM2.5 and meteorological data from Beijing.
• To develop a flexible statistical framework that models marginal distributions and dependence structures separately, enabling robust prediction of extreme pollution events.
• To enable scenario-based analysis of climate impacts on air pollution by simulating future PM2.5 levels under user-specified meteorological conditions.
• To provide a foundation for health risk assessment by accurately modeling time-varying latent pollution states linked to health outcomes.

Proposed method

• The model uses bivariate copulas to specify the dependence structure in both the observation equation (PM2.5 and covariates) and the state equation (latent pollution states).
• Latent states are modeled as a first-order Markov process with a copula-based transition distribution, allowing flexible non-linear dynamics.
• Bayesian inference is performed using Hamiltonian Monte Carlo (HMC), with the posterior distribution defined via copula densities and uniform priors on dependence parameters.
• The latent variables of the state equation are treated as parameters in the HMC algorithm, enabling joint estimation of states and model parameters.
• The approach uses probability integral transforms to convert observed data into uniform variates for copula modeling, ensuring valid copula inference.
• The No-U-Turn sampler is employed to adaptively tune HMC parameters (step size, trajectory length, mass matrix), improving sampling efficiency.

Experimental results

Research questions

• RQ1 Can a non-linear, non-Gaussian state space model with copula-based dependence structures better capture the dynamics of hourly PM2.5 and meteorological data in Beijing than standard Gaussian models?
• RQ2 How do latent pollution states identified by the model relate to extreme pollution events not fully explained by covariates like weather and seasonality?
• RQ3 To what extent can the model simulate future PM2.5 levels under hypothetical climate scenarios, providing actionable insights for public health planning?
• RQ4 How does the flexibility of copula-based modeling improve the estimation of marginal distributions and tail dependencies in air pollution time series?

Key findings

• The standard Gaussian state space model fails to capture the true time dynamics of PM2.5 data, particularly extreme values and non-linear dependencies.
• The proposed copula-based state space model successfully models both marginal distributions and complex dependence structures, enabling accurate representation of non-Gaussian behavior.
• The model identifies time-points where latent states significantly impact PM2.5 levels, corresponding to unexplained pollution spikes beyond covariate effects.
• Hamiltonian Monte Carlo with adaptive tuning (No-U-Turn sampler) enables efficient and reliable posterior sampling for the high-dimensional, non-linear copula model.
• The model supports scenario-based forecasting, allowing stakeholders to predict PM2.5 levels under different meteorological conditions.
• The approach is extendable to multivariate responses and higher-dimensional dependence via vine copulas, enabling future applications in multi-pollutant modeling.