AQI prediction using layer recurrent neural network model: a new approach.

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Chemistry

Background:

• Airborne pollutants pose a significant threat to human health in Delhi.
• Accurate air quality index (AQI) assessment is vital for evaluating pollutant impacts.
• Existing methods require enhancement for precise, long-term AQI prediction.

Purpose of the Study:

• To predict the daily air quality index (AQI) one year in advance for Delhi.
• To compare the predictive performance of three neural network models: FF-NN, CF-NN, and LR-NN.
• To identify key air pollutants influencing daily AQI levels.

Main Methods:

• Developed and trained three neural network models (FF-NN, CF-NN, LR-NN) using 2019 AQI data.
• Incorporated primary pollutants (PM2.5/PM10, O3, SO2, NOx, CO, NH3), non-criteria pollutants, and meteorological data as input.
• Assessed model performance using statistical analysis, focusing on Root Mean Square Error (RMSE) and R-squared (R²).

Main Results:

• The LR-NN model achieved the lowest RMSE (26.79), indicating superior prediction accuracy.
• PM2.5/PM10, CO, and NO2 were identified as the primary contributors to high daily AQI.
• LR-NN demonstrated strong performance across all seasons, particularly in the post-monsoon period (R² = 0.94).

Conclusions:

• The LR-NN model offers a precise method for one-year-ahead AQI prediction.
• Findings can assist air pollution control authorities in implementing timely and effective mitigation strategies.
• Further research comparing LR-NN with other modeling approaches is recommended for comprehensive air quality assessment.