Ground-penetrating radar (GPR) is a convenient geophysical technique for active-layer soil moisture detection in permafrost regions, which is theoretically based on the petrophysical relationship between soil moisture (θ) and the soil dielectric constant (ε). The θ–ε relationship varies with soil type and thus must be calibrated for a specific region or soil type. At present, there is lack of such a relationship for active-layer soil moisture estimation for the Qinghai–Tibet plateau permafrost regions. In this paper, we utilize the Complex Refractive Index Model to establish such a calibration equation that is suitable for active-layer soil moisture estimation with GPR velocity. Based on the relationship between liquid water, temperature, and salinity, the soil water dielectric constant was determined, which varied from 84 to 88, with an average value of 86 within the active layer for our research regions. Based on the calculated soil-water dielectric constant variation range, and the exponent value range within the Complex Refractive Index Model, the exponent value was determined as 0.26 with our field-investigated active-layer soil moisture and dielectric data set. By neglecting the influence of the soil matrix dielectric constant and soil porosity variations on soil moisture estimation at the regional scale, a simple active-layer soil moisture calibration curve, named CRIM, which is suitable for the Qinghai–Tibet plateau permafrost regions, was established. The main shortage of the CRIM calibration equation is that its calculated soil-moisture error will gradually increase with a decreasing GPR velocity and an increasing GPR velocity interpretation error. To avoid this shortage, a direct linear fitting calibration equation, named as υ-fitting, was acquired based on the statistical relationship between the active-layer soil moisture and GPR velocity with our field-investigated data set. When the GPR velocity interpretation error is within ±0.004 m/ns, the maximum moisture error calculated by CRIM is within 0.08 m³/m³. While when the GPR velocity interpretation error is larger than ±0.004 m/ns, a piecewise formula calculation method, combined with the υ-fitting equation when the GPR velocity is lower than 0.07 m/ns and the CRIM equation when the GPR velocity is larger than 0.07 m/ns, was recommended for the active-layer moisture estimation with GPR detection in the Qinghai–Tibet plateau permafrost regions.

Accurate permafrost mapping in mountainous regions is hindered by sparse in situ observations and heterogeneous terrain. This study develops a GPR-augmented machine learning framework to map mountain permafrost in the northeastern Qinghai–Tibet Plateau. A total of 1037 presence–absence samples were compiled from boreholes, soil pits, 128 GPR transects collected in 2009, and 22 additional empirical points above 4700 m, covering diverse topographic and thermal conditions. Thirteen classification algorithms were evaluated using 5-fold cross-validation repeated 40 times, with LightGBM, CatBoost, XGBoost, and RF achieving top performance (F1 > 0.98). Elevation-based spatial comparisons revealed that LightGBM and CatBoost produced more terrain-adaptive predictions at high altitudes and slope transitions. Aspect-controlled permafrost boundaries were captured, with modeled lower elevation limits varying by >200 m across slope directions. SHAP analysis showed that climate and soil variables contributed nearly 80% to model outputs, with LST, FDD, BD, and TDD being dominant. Several predictors exhibited threshold or nonlinear responses, reinforcing their physical relevance. Additional experiments confirmed that integration of GPR and high-elevation constraint samples significantly improved model generalization, especially in underrepresented terrain zones. This study demonstrates that a GPR-augmented machine learning framework can support cost-effective, physically informed mapping of frozen ground in complex alpine environments.