Meteorological Normalization or Deweathering for Predicting Air Pollutant Concentration: Pitfalls and Limitations

Valentino Petrić,Mario Lovrić,Bernhard C. Geiger

DOI: https://doi.org/10.26434/chemrxiv-2024-sw2cm

2024-10-14

Abstract:Meteorological normalization is a key concept in studying anthropogenic effects on air pollutant concentrations and its temporal trends. While apparently successful in revealing anthropogenic effects and often used, there are downsides to the methods and limitations which should be taken into account when using it. This study examines the concept of meteorological normalization in the context of predicting air pollutant concentrations, particularly focusing on the method's theoretical background and implicit assumptions. We provide a rigorous analysis that outlines the conditions under which meteorological normalization can effectively estimate expected values of response variables, which are commonly time series of air pollutant concentrations such as PM$_{10}$ . Through our analysis, we identify critical assumptions, such as the independence of meteorological and non-meteorological variables, which can significantly impact the validity of the method's outcomes. We also explore potential failure modes of meteorological normalization when these assumptions are violated, offering practical examples and visualizations that highlight the limitations of this technique. Our findings suggest that while meteorological normalization can be a useful tool, careful consideration of its applicability is necessary to avoid misleading conclusions in air quality assessments.

Chemistry

What problem does this paper attempt to address?

The paper attempts to address the issues and limitations of meteorological normalization in predicting air pollutant concentrations. Specifically, the paper mainly explores the following points:

1. **Theoretical Background and Assumptions**:
- Study the mathematical principles of meteorological normalization and derive its applicable conditions. These conditions include successful machine learning modeling and the marginal independence between meteorological and non-meteorological variables.
- When these conditions are met, meteorological normalization can effectively estimate the expected value of the response variable (e.g., PM10 concentration) based solely on non-meteorological variables (e.g., seasonal changes).

2. **Limitations in Practical Applications**:
- The paper demonstrates through experiments the failure modes that meteorological normalization may exhibit under different conditions. For example, when meteorological variables are correlated with non-meteorological variables, meteorological normalization cannot correctly estimate the expected value of the response variable.
- In forecasting settings, when using meteorological normalization to predict pollutant concentrations at future time points, the model may fail to extrapolate well due to data outside the range of the training data.
- In extrapolation settings, if there is a correlation between meteorological and non-meteorological variables, meteorological normalization will sample combinations not covered by the training data, leading to model prediction bias.

3. **Conclusions and Recommendations**:
- The paper points out that in practical applications, there is usually a significant correlation between meteorological variables and non-meteorological variables (such as time), making meteorological normalization inapplicable in many cases.
- It advocates for caution when using meteorological normalization and considers other alternative methods or improvements to existing techniques to overcome its limitations.

In summary, this paper aims to reveal the potential issues of the meteorological normalization method in practical applications and proposes corresponding analyses and solutions.