<p>The inversion of self-potential (SP) data for estimating parameters of subsurface sources such as two-dimensional inclined sheets is a nonlinear and inherently non-unique problem that often leads to instability and inaccurate parameter recovery. In this study, a hybrid inversion strategy that combines the global Luus–Jaakola (LJ) algorithm with the local Levenberg–Marquardt (LM) method is proposed to enhance both the stability and accuracy of SP data interpretation. The approach is evaluated through three different parameterizations of the inclined sheet model, each representing alternative formulations of the source geometry and location. A numerical sensitivity analysis reveals that the SP anomaly is most sensitive to depth-related parameters and dip angles, while parameters controlling horizontal position, sheet length, and polarization coefficient have considerably less influence. Synthetic data contaminated with 20% Gaussian noise confirm that the hybrid LJ–LM algorithm produces substantially lower normalized root mean square (RMS) errors than the LJ-only inversion. The results also indicate that the proposed hybrid approach compares favorably with previously reported optimization approaches such as the very fast simulated annealing (VFSA) method. Across the three parameterizations, the hybrid solution achieves RMS values in the range of approximately 10⁻³–10⁻⁴, with relative errors of key parameters generally below 0.2%. Application to the Surda field dataset further demonstrates the method’s robustness, yielding a misfit of about 0.028—representing roughly a 45% improvement compared with previously reported values near 0.052–0.053. These results indicate that the proposed hybrid LJ–LM inversion significantly improves both the reliability of parameter estimates and the overall consistency between observed and modeled SP anomalies for inclined sheet sources.</p>