Discontinuity Predictions of Porosity and Hydraulic Conductivity Based on Electrical Resistivity in Slopes through...

Area of Science:

• Geophysics
• Soil Science
• Data Science

Background:

• Electrical resistivity surveys provide critical data on soil strata.
• Predicting soil electrical resistivity aids in forecasting soil behavior.

Purpose of the Study:

• To apply and evaluate deep learning algorithms (DNN, LSTM, GRU, LSTM-DNN, GRU-DNN) for electrical resistivity prediction.
• To determine the reliability of these predictions for identifying porosity and hydraulic conductivity discontinuities.
• To estimate discontinuity depth in soil strata.

Main Methods:

• Utilized 101 electrodes with a Wenner array for electrical resistivity measurements over 15 months.
• Applied deep learning models including DNN, LSTM, GRU, and hybrid DNN-based algorithms (LSTM-DNN, GRU-DNN).
• Trained models using a 6:2:2 ratio of training, validation, and test data, with hyperparameter tuning for reliability.

Main Results:

• Deep learning models demonstrated high reliability in predicting electrical resistivity.
• The models successfully deduced distributions of porosity and hydraulic conductivity.
• An average discontinuity depth of 25 m was estimated.

Conclusions:

• Deep learning techniques are effective for predicting soil electrical resistivity, porosity, and hydraulic conductivity.
• These methods provide a reliable approach for estimating discontinuity depth.
• The study validates the utility of advanced machine learning in geotechnical and hydrological assessments.