PM10 air pollutant prediction using deep learning LSTM Model: A case study of Istanbul, Türkiye


  • Omar Wisam Alqaysi
  • Dursun Zafer Şeker


Deep learning, LSTM, Air pollution, PM10, GRU


Accurate forecasting of PM10 concentrations is crucial for air quality management and public health protection. This study proposes a deep learning-based model for predicting PM10 concentrations in Istanbul, Türkiye, utilizing a combination of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models. Historical air pollution data from the Ministry of Environment, Urbanization, and Climate Change of Turkey and meteorological data from NASA for the period of January 2018 to August 2023 were employed for model development. Ümraniye district was selected as the study area due to its comprehensive air quality data availability. An extensive model development process involved identifying the optimal input sliding window, input features, and model architecture through parameter tuning. The LSTM+GRU model resulted in the best metrics, achieving an RMSE of 6.71, R2 of 0.86, and MAPE of 15.9%. The model demonstrated strong generalization capabilities when tested on data from eight different stations in Istanbul. While the proposed model exhibited promising performance, certain limitations warrant further investigation. The effectiveness of the model for air pollutants other than PM10 remains unexplored. Additionally, an evaluation of feature importance ranking for the input parameters is necessary to identify the most influential factors contributing to PM10 concentrations. Future research endeavors will address these limitations and refine the model for broader applicability.