Spatial-Temporal Forecasting of Air Pollution in Saudi Arabian Cities Based on a Deep Learning Framework Enabled by...

Area of Science:

• Environmental Science
• Data Science
• Public Health

Background:

• Industrialization and economic activities are increasing air pollution globally and in Saudi Arabia, posing health risks.
• Mathematical modeling of air pollution is essential for effective environmental and health decision-making in Saudi Arabia.
• Existing air quality monitoring networks require complementary modeling approaches for comprehensive assessment.

Purpose of the Study:

• To develop and evaluate a data-driven deep learning model for temporal and spatial air pollution modeling in Saudi Arabia.
• To assess the performance of Long Short-Term Memory (LSTM) algorithm against ensemble methods like Random Forest and XGBoost.
• To investigate the influence of meteorological factors on pollutant concentrations and their impact on model accuracy.

Main Methods:

• A data-driven approach utilizing historical pollutant records (PM10, PM2.5, CO, O3) and time series analysis.
• Implementation of a Deep Learning (DL) Long Short-Term Memory (LSTM) algorithm for temporal modeling.
• Spatial modeling focused on major Saudi cities and comparison with Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) ensemble methods.

Main Results:

• LSTM demonstrated superior performance in forecasting air pollutant concentrations, especially with larger datasets, achieving R² values up to 0.8073.
• LSTM effectively captured complex, hidden relationships in the data, outperforming RF and XGBoost in predictive accuracy.
• Meteorological factors, including ambient temperature, showed a weak to moderate association with pollutant levels, and their inclusion did not significantly enhance LSTM's predictive accuracy.

Conclusions:

• The developed LSTM-based approach provides accurate temporal and spatial modeling of air pollution in Saudi Arabia.
• This data-driven methodology offers a valuable tool for supporting environmental and health policy decisions.
• The study highlights the potential of deep learning for air quality monitoring and management using historical time series data.