Spatiotemporal prediction of O₃ concentration based on the KNN-Prophet-LSTM model

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Accurate prediction of air pollutant concentrations, specifically ozone (O3), is crucial for environmental monitoring and public health.
• Traditional time series models often struggle to capture complex spatio-temporal dependencies in air quality data.
• Existing hybrid models may not fully leverage both spatial proximity and temporal patterns for enhanced prediction.

Purpose of the Study:

• To develop and validate a novel hybrid model (KNN-Prophet-LSTM) for daily ozone (O3) concentration forecasting.
• To assess the contribution of Prophet decomposition, LSTM, and K-Nearest Neighbor (KNN) algorithm integration in improving prediction accuracy.
• To compare the performance of the proposed hybrid model against single models (ARIMA, Prophet, LSTM) and a simpler hybrid (Prophet-LSTM).

Main Methods:

• Utilized daily O3 concentration data from Wuhan (2014-2021).
• Employed Prophet decomposition to separate time series into trend, periodic, and error components.
• Applied Prophet for trend and periodic components, LSTM for error terms, and KNN to fuse spatio-temporal information for final O3 prediction.

Main Results:

• Daily O3 concentrations in Wuhan exhibit significant periodic variations influenced by surrounding environmental factors.
• Prophet decomposition effectively extracts time series information and reduces noise, enhancing prediction accuracy.
• Integrating spatial information via KNN further improves model accuracy, achieving substantial reductions in Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) compared to ARIMA.

Conclusions:

• The KNN-Prophet-LSTM hybrid model demonstrates superior performance in predicting daily O3 concentrations compared to individual models.
• The integration of spatial information through KNN significantly enhances the predictive power of the hybrid model.
• The findings highlight the effectiveness of combining decomposition techniques with advanced machine learning for accurate air quality forecasting.