A comparison of positive matrix factorization and the weighted multivariate curve resolution method. Application to...

Area of Science:

• Environmental science
• Chemometrics
• Air quality modeling

Background:

• Positive Matrix Factorization (PMF) is a widely adopted tool in environmental science for air quality control, recommended by the U.S. EPA.
• PMF effectively incorporates measurement uncertainty, handles missing data and values below reporting limits, and ensures physically meaningful, non-negative solutions.
• Multivariate Curve Resolution-Weighted Alternating Least Squares (MCR-WALS) is an alternative method that also utilizes measurement error information and non-negativity constraints.

Purpose of the Study:

• To compare the performance of PMF and MCR-WALS in analyzing environmental datasets.
• To evaluate how different correlation and error structures in simulated data affect the results of both methods.

Main Methods:

• Simulated datasets with varying correlation and error structures were generated.
• The performance of PMF and MCR-WALS was assessed using these simulated datasets.
• Both methods employ the same loss function but differ in profile extraction techniques.

Main Results:

• Both PMF and MCR-WALS demonstrated comparable performance in extracting source profiles.
• The profiles extracted by both methods were virtually identical across datasets with different error structures.
• The study highlights the robustness of both techniques in handling measurement uncertainty and constraints.

Conclusions:

• PMF and MCR-WALS are both suitable multivariate methods for environmental data analysis, particularly for air quality applications.
• The choice between PMF and MCR-WALS may depend on specific dataset characteristics and user preference, as their performance is similar.
• Both methods provide reliable and physically meaningful source composition and contribution profiles.