Spatio-temporal prediction of soil hydro-thermal response in embankment to the varying climatic conditions using a Two-Step LSTM-ML approach

Soil hydro-thermal response is vital in the assessment of embankment stability, especially considering the increasing extreme climate events along with global climate change. Machine learning has emerged as a promising approach to estimate soil hydro-thermal dynamics. Current spatio-temporal prediction methods of soil hydro-thermal response faces significant challenges when applied to embankments, including the employment of remote sensing data with large-scale resolution, extensive datasets and high computational costs requested by layered modeling approaches. To address these limitations, this study develops a Two-Step LSTM-ML approach to predict the spatio-temporal variations of soil temperature and volumetric water content in embankments. The method is conducted in two steps: a Long Short-Term Memory (LSTM) model for the prediction of surface soil hydro-thermal response in step 1 and then combined with machine learning algorithms, e.g., Support Vector Regression (SVR), Random Forest (RF), and Artificial Neural Networks (ANN), to estimate the soil hydro-thermal response at various depths in step 2. The developed approach is trained and validated using measured data from July 6, 2011 to October 7, 2011 at the Héricourt embankment in France. The results demonstrate that the LSTM model effectively captures temporal variations in surface soil temperature and volumetric water content, while spatial mapping using RF and ANN models provides reliable predictions of soil variables at different depths. Additionally, this study examines the effects of different time-step and an iterative computation approach for surface soil variable prediction, further refining the model’s performance. The proposed method offers a robust and efficient framework for predicting soil temperature and volumetric water content in embankments, with potential applications for transportation infrastructure management and climate change adaptation.