Study regions: The Koga watershed is in the Blue Nile River Basin in Ethiopia. Study focus: The study conducted in a hilly dominated tropical catchment in Ethiopia adopted a multi-model approach by comparing six hydrological models, including the semi-distributed Soil and Water Assessment Tool (SWAT) and five lumped conceptual models, namely: Australian Water Balance Model (AWBM), Sacramento, Simplified HYDROLOG (SimHYD), Soil Moisture Accounting and Routing (SMAR), and Tank models. The study aimed at evaluating model performance and sensitivity of parameters in predicting streamflow for tropical watersheds. Calibration and uncertainty analysis (UA) for SWAT was performed using four UA techniques available in the SWAT (SWAT-CUP) and the genetic algorithm was used for parameter estimation for conceptual models. The model comparative analysis was supplemented by multi-model averaging using Sum of Squared Errors (SSEW) and Performance-Based Weights (PBW-NSE) to improve prediction reliability. New hydrological insights for the region: Results showed that the semi-distributed scheme using SWAT-SUFI-2 provided practical applications in land use change modelling and provided better calibration efficiency with limited computational resources. Sacramento offers improved performance with cost-effective alternative to hydrological modelling in a data-scarce environment. For monthly flows, SWAT-SUFI-2 achieved NSE of 0.60 and PBIAS of 24 % for calibration (P-factor = 73, R-factor = 93) and 0.60 and 9 % for validation (P-factor = 70, R-factor = 88). Sacramento model yielded NSE of 0.55 and PBIAS of 16 % during calibration, and 0.72 and-4 % for validation, respectively. Compared to the individual models, multi-model averaging enhanced overall performance, with monthly NSE ranging from 0.60 to 0.67 and R2 from 0.71 to 0.74. While similar in terms of model performance, SWAT takes into account the spatial variability in soil, land use, and terrain, hence has the capacity for impact assessment of environment change in comparison to lumped conceptual models.