Multitemporal time series analysis using machine learning models for ground deformation in the Erhai region, China

Area of Science:

• Geosciences
• Remote Sensing
• Machine Learning

Background:

• Ground deformation (GD) is a critical global issue endangering public safety.
• Monitoring GD is challenging due to its complex and dynamic nature.
• Interferometric Synthetic Aperture Radar (InSAR) is a key technology for large-scale GD assessment.

Purpose of the Study:

• To model the effects of ground deformation in the Erhai region.
• To investigate the influence of factors like building area, water level, precipitation, and temperature on GD.
• To compare the efficacy of various machine learning methods for predicting GD.

Main Methods:

• Utilized small baseline subset interferometric SAR (SBAS-InSAR) data.
• Applied machine learning (ML) techniques including Multiple Linear Regression (MLR), Multilayer Perceptron Backpropagation (MLP-BP), Least Squares Support Vector Machine (LSSVM), and PSO-LSSVM.
• Evaluated model performance using Root Mean Square Error (RMSE) and Mean Relative Error (MRE).

Main Results:

• The PSO-LSSVM model achieved the lowest RMSE (11.448) and MRE (0.112).
• This indicates superior performance in predicting ground deformation compared to other tested ML methods.
• The study successfully modeled the relationship between influencing factors and GD.

Conclusions:

• The PSO-LSSVM method demonstrates high efficiency and accuracy for analyzing ground deformation.
• Machine learning approaches, especially PSO-LSSVM, offer a powerful tool for GD monitoring and risk assessment.
• Accurate GD prediction is crucial for mitigating risks to public safety.