Prediction of Pollen Concentration in Six Urban Districts of Beijing Using an LSTM-Based Model

Abstract

Airborne pollen can cause various allergic symptoms and impact the quality of life of a considerable part of the global population. Therefore, it is crucial to forecast airborne pollen concentrations and disseminate this information to protect public health and improve the quality of life of the overall population. In this study, a data-oriented approach was adopted to develop a daily pollen concentration prediction model using Long Short-Term Memory (LSTM) neural network. The prediction model is based on the pollen concentration time-series dataset of six urban districts of Beijing (Chaoyang, Haidian, Fengtai, Shijingshan, Changping, and Shunyi) recorded from 2021 to 2025. The pollen concentration dataset shows that the data characteristics of the six districts are different, even though they are all located in Beijing and close to each other. Therefore, different look_back values were set for the six districts in the LSTM-based prediction model to achieve better prediction performance. In the pollen concentration prediction tests for six districts, the model successfully predicted pollen concentrations in 2025 and achieved R2 values between 0.4 and 0.7. Moreover, this work also studied the seasonal variation of detailed pollen species concentrations (including nine woody and four herbaceous plants) in six urban districts. The daily airborne pollen concentration prediction model can provide a reference for local pollen allergy prevention.