Dual-parameter inversion of snowpack GPR data using machine learning

Precise estimation of permittivity (ε) and conductivity (σ) is crucial for assessing the liquid water content and density in snowpack, which are the key factors for avalanche risk assessment, water resource management and seasonal hydrological forecasting. This study introduces a machine learning based framework to automate the inversion of dual paramter using ground penetrating radar (GPR) data. A high resolution synthetic training dataset was generated using finite difference time domain (FDTD) simulations in gprMax, representing diverse snowpack environments (dry, moist and wet snow) with variable layering and spatial heterogeneity. Principle component analysis (PCA) was applied to reduce the dimensionality of A-scans while preserving key signal features. Four ML models including random forest (RF), neural network (NN), extreme gradient boosting (XGB) and support vector Machines (SVM) were trained and evaluated on synthetic data. Gausian noise augmentation and model fine tuning were applied to enhance the generalization before testing on real field GPR data collected from Altay region of Xinjiang, China. The NN model achived the accuracy over 93% for permittivty and 85% for the resistivty across both moist and wet snowpacks. Model predictions were cross validated against the results from the frequency-domain attenaution analyis and in-situ snow fork measurments, confirming the reliablity of the ML-based inversion. This framework provides a robust and scalable approach for snowpack characterization using common offset GPR data.