Model evaluation and spatial interpolation by Bayesian combination of observations with outputs from numerical models

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Spatial Statistics

Background:

• Mapping dry deposition pollution is crucial for air quality management but challenged by sparse monitoring networks.
• Existing data sources include the Clean Air Status and Trends Network (CASTNet) and regional air quality models like Models-3.
• Evaluating the accuracy of air quality models is essential for developing effective pollution control strategies.

Purpose of the Study:

• To develop formal methods for combining disparate data sources (sparse observations and model outputs) with different spatial resolutions.
• To create a Bayesian framework for estimating unobserved ground truth pollution levels.
• To validate and correct biases in air quality model predictions.

Main Methods:

• A Bayesian approach was used to model both Models-3 output and CASTNet observations in relation to the unobserved ground truth.
• The method allows for improved spatial prediction through the posterior distribution of the ground truth.
• Model validation was performed using the posterior predictive distribution of CASTNet observations.

Main Results:

• High-resolution SO2 (sulfur dioxide) distributions were generated by integrating CASTNet data with Models-3 output.
• The Bayesian method successfully improved spatial predictions and enabled bias correction of the Models-3 output.
• Model performance was found to be poorer near power plants, with SO2 values being overestimated.

Conclusions:

• The developed Bayesian framework effectively combines sparse monitoring data and model outputs for enhanced air quality mapping.
• The study provides a robust method for evaluating and improving the accuracy of regional air quality models.
• Findings highlight the need to address model biases, particularly in areas with significant pollution sources like power plants.