Evaluating the predictability of PM₁₀ grades in Seoul, Korea using a neural network model based on synoptic patterns

Area of Science:

• Environmental Science
• Atmospheric Science
• Data Science

Background:

• The Korean Ministry of Environment (KME) forecasts particulate matter (PM10) into four grades.
• The performance of KME's PM10 forecasting system has not been fully evaluated.
• Accurate PM10 forecasting is crucial for public health and environmental management.

Purpose of the Study:

• To evaluate the performance of the KME's PM10 forecasting system.
• To develop a new neural network model for improved PM10 forecasting.
• To establish a statistical reference for air quality prediction.

Main Methods:

• Developed a neural network model using meteorological fields (geopotential height, temperature, humidity, wind) and prior-day PM10 concentrations.
• Quantified meteorological patterns as cosine similarities to train the model.
• Performed hindcast analysis for Seoul's cold seasons (2001-2014).

Main Results:

• Achieved an overall hit rate of 69% for PM10 grade forecasting.
• Specific hit rates: 33% (low), 83% (moderate), 45% (high), and 33% (very high).
• Identified meteorological synoptic patterns as reliable predictors for PM10 grades and transboundary transport.

Conclusions:

• The developed neural network model provides a reliable statistical reference for PM10 forecasting.
• Meteorological patterns are key indicators for predicting air quality and transboundary PM10.
• Findings can complement and enhance the existing KME air quality forecasting system.