A predictive framework for land subsidence risk in Silakhor: integrating machine and deep learning

Area of Science:

• Geosciences
• Environmental Science
• Data Science

Background:

• Land subsidence poses significant environmental and economic challenges, particularly in the Silakhor Plain, Lorestan Province.
• Understanding the drivers of subsidence is crucial for effective land and water resource management.

Purpose of the Study:

• To investigate the key factors contributing to land subsidence in the Silakhor Plain.
• To evaluate the performance of machine learning and deep learning models in predicting and classifying subsidence.
• To identify high-risk areas for subsidence within the plain.

Main Methods:

• Utilized machine learning (Random Forest) and deep learning (RNN, LSTM, CNN) models.
• Analyzed factors including land use, drainage density, soil properties, and groundwater levels.
• Employed SHAP analysis to interpret model findings and identify significant drivers.

Main Results:

• Soil properties, excessive groundwater exploitation, and human infrastructure were identified as primary drivers of subsidence.
• The Random Forest model achieved the highest accuracy (R²=0.9880) in predicting subsidence intensity.
• Central and southeastern areas of the plain were identified as high-risk zones due to well density and fault proximity.

Conclusions:

• Machine learning and deep learning models are effective tools for subsidence prediction and risk assessment.
• Findings provide a basis for water and land resource management strategies in the Silakhor Plain.
• Effective mitigation strategies are needed to address subsidence risks in vulnerable areas.