Investigations into ground subsidence in Tianjin coastal area based on random forest

Objective The spatial distribution of ground subsidence in the coastal area of Tianjin was predicted using a random forest machine learning model, in which the performance and significance of the variables were evaluated. Methods The random forest model was trained and validated in this study using datasets of ground subsidence in 2020, aquifer lithology, water level differences in aquifers in 2020, and hydrogeological parameters. Results The results demonstrate the effectiveness of the random forest model for fitting and predicting ground subsidence (R2=0.98, RMSE=0.52 mm). Moreover, it is found that water level difference emerges as the most influential factor affecting ground subsidence, followed by lithology and hydrogeological parameters. Conclusion The present study introduces several novel contributions: ① utilization of spatial distribution data for training ground subsidence models; ② identification of significant controlling factors based on physical mechanisms; ③ assessment of the relative importance of these controlling factors. Additionally, this paper highlights the limitations and future directions in ground subsidence research, offering valuable insights for the rapid and accurate prediction of ground subsidence using the random forest model.