To improve the scientific basis and accuracy of regional landslide risk identification, the uncertainty of geotechnical parameters in shallow landslides triggered by earthquakes is considered. An improved First Order Reliability Method (FORM) is used to develop a probabilistic framework for regional landslide susceptibility assessment. A physical model of an infinite slope under seismic loading is constructed using a pseudo-static approach, where key soil parameters are treated as random variables to quantify the impact of uncertainty on landslide failure probability. A QGIS plugin, QGIS-FORM, is developed in Python to automate the generation of regional susceptibility maps based on failure probability ( ) and factor of safety (FOS). Using the Maerkang earthquake-induced landslides as a case study, key parameters—including slope, peak ground acceleration, and geological lithology—are analyzed. A comparison is made between FORM and the Mean First Order Second Moment method (MFOSM). The predictive performance of both   and FOS is evaluated under different buffer zone sizes and levels of coefficient of variation (COV), using receiver operating characteristic (ROC) curves and balanced accuracy (BA) as evaluation metrics. Results indicate that FORM performs better than MFOSM in addressing the nonlinear behavior of complex slopes, showing an improvement of 5.5% in AUC. Under different buffer sizes, the AUC values for   are 82.9%, 84.1%, and 85.0%, all exceeding those of FOS. BA analysis shows that with increasing COV, the optimal   thresholds for FORM are 0.08, 0.2, and 0.27, each corresponding to a maximum BA of 0.704. These findings suggest that while COV influences the sensitivity of threshold selection, it does not compromise the model?s predictive performance. The FORM-based method accounts for input uncertainty and provides more stable and detailed landslide risk zoning. It offers a scientifically grounded visualization tool for landslide risk management.