Accurate seismic site response prediction is vital for resilient earthquake-resistant infrastructure. Nonlinear site response analysis, widely recognized for its accuracy across a broad range of shaking levels and soil stiffnesses, lacks clear guidelines for model selection and calibration. This study investigates the impact of DEEPSOIL's constitutive models on surface response spectra, shear strains, and peak accelerations in South Carolina's Blue Ridge and Piedmont regions. Utilizing the modified Kodner- Zelasko (MKZ), general quadratic/hyperbolic (GQ/H), and Yee et al. models, calibrated using DEEPSOIL's MRDF technique, it was observed that the GQ/H model aligned best with low shear strain data. However, shear strength was underestimated across models, with the Yee et al. model exhibiting unrealistic oscillations. The GQ/H model consistently predicted lower peak accelerations and larger shear strains when using Kobe, Parkfield, Coyote, and Val-des-Bois motions. The characteristics of ground motion significantly influenced computed responses. Further studies are needed for model validation and optimal response prediction.
