Forecasting daily source air quality using multivariate statistical analysis and radial basis function networks.

Area of Science:

• Environmental Science
• Agricultural Engineering
• Data Science

Background:

• Accurate forecasting of gas and particle matter concentrations and emission rates (GPCER) from livestock facilities is essential for assessing environmental and health impacts.
• Modeling air quality from livestock operations is complex due to numerous nonlinear interactions.

Purpose of the Study:

• To introduce statistical methods and radial basis function (RBF) neural networks for predicting daily source air quality in swine deep-pit finishing buildings.
• To identify key variables influencing GPCER and simplify the modeling process.

Main Methods:

• Utilized statistical methods to identify important input variables for air quality modeling.
• Applied principal component analysis (PCA) to reduce variable dimensionality.
• Employed radial basis function (RBF) neural networks to predict daily GPCER.

Main Results:

• Four key variables (outdoor/indoor temperature, animal units, ventilation rates) were identified as significant predictors.
• PCA revealed two main factors (environmental and animal) explaining over 94% of variability.
• RBF network predictions showed strong agreement with actual measurements (correlation coefficient 0.741–0.995) with low systemic performance indexes.

Conclusions:

• RBF neural networks effectively model the nonlinear relationships in air pollutant emissions.
• Combining RBF networks with multivariate statistical methods offers a promising approach for livestock air quality modeling.