<p>Accurate sediment yield estimation is essential for watershed management, yet modeling uncertainties persist, particularly in the absence of continuous event-based observations. This study evaluates event-scale sediment yield predictions for 62 storm events over four years using RUSLE/SDR and MUSLE within a GIS-based framework for the Çarşamba Suyu Sub-Basin, Türkiye. Model parameters were derived from rainfall erosivity, soil data, topographic features, and satellite imagery. Calibration incorporated energy-based erosivity metrics rainfall kinetic energy (<i>R</i>-factor) and runoff energy (<i>Q.q</i>_<i>p</i>) to enhance predictive performance. Although official records have shifted from event-based to monthly discharge monitoring in recent years, available observations allowed model calibration and partial validation. Both MUSLE and RUSLE showed strong internal consistency (<i>R</i>² = 0.95), and significant correlations with observed data (MUSLE: <i>R</i>² = 0.70, RUSLE/SDR: <i>R</i>² = 0.56). Overestimation was noted in 2017 (MUSLE: 1.189 t ha⁻¹ yr⁻¹, RUSLE/SDR: 1.419 t ha⁻¹ yr⁻¹, observed: 0.328 t ha⁻¹ yr⁻¹), while MUSLE closely aligned with observations in 2018 (0.952 vs. 0.991 t ha⁻¹ yr⁻¹). By 2019, both models converged on observed values (MUSLE: 0.259, RUSLE/SDR: 0.288, observed: 0.238 t ha⁻¹ yr⁻¹). Integration of the Composite Weighted Connectivity Index (CWCI) and seasonal adjustment of the soil erodibility factor (<i>K</i>) further improved accuracy. The findings highlight RUSLE's suitability for long-term trend analysis, while MUSLE provides event-specific resolution. A hybrid framework combining both models alongside connectivity-based calibration and regression between <i>R</i> and <i>Q</i>·<i>qp</i> is recommended to improve sediment yield prediction, especially in data-limited basins.</p>