Abstract : Climate change and anthropogenic activities create uncertainty with respect to future hydrological conditions, and thus pose challenges in predicting streamflow, particularly the magnitude of extreme events. Several studies have focused on understanding future flood risk under climate and land use/land cover (LULC) changes using hydrological models. In addition to biases from climate data, biases from hydrological models, especially on peak flow simulations were reported to be large (usually underestimations). This could limit the dependability of flood risk projections and their applicability for future decision making. This research study investigates techniques and approaches for improved simulation of streamflows with focus on peak flows using the Soil and Water Assessment Tool (SWAT) for three case study watersheds. In particular, evaluations include choice of criteria for sensitivity analysis and parameter identification, choice of objective function for calibration, and impact assessment when calibrated models are applied for periods with alternate climate and physical characteristics. For ease of calibration, sensitivity analysis is crucial to identify relevant parameters; however, it can provide different parameter sets based upon the implemented sensitivity criteria. Herein, four sensitivity criteria, namely the Nash-Sutcliffe Efficiency (NSE), coefficient of determination (R2), modified R2 (bR2), and percent bias (PBIAS) were compared in watersheds of contrasting climate, hydrology, and land cover. For rainfall-runoff dominated agricultural watersheds, NSE, bR2, and R2 produced relatively similar parameter sets, and thus these criteria can be used individually or together for the purposes of sensitivity analysis, especially if peak flows are the target. For a snowmelt dominated forested watershed, R2 was found to be the best sensitivity criterion to identify parameters affecting peak flows. Moreover, for this watershed, sensitivity analysis and light calibration of snowmelt related parameters separately followed by calibration of the hydrological parameters resulted in improved flow simulations compared to the default approach where all parameters were analyzed together. The ability of models calibrated to a given set of climate and LULC data to simulate flood risk under altered conditions was assessed in each watershed by applying parameters calibrated for 2002-2005 to 1970-1999. Simulated annual maximum daily flows for the latter period were used to estimate the instantaneous annual maximum flow (AMF) series, and the impact of altered parameter values on the resulting flood distribution was assessed via a one-at-a-time sensitivity analysis. As anticipated, AMFs in the agricultural rainfall-runoff dominated watersheds were sensitive to changes in runoff related parameters, whereas AMFs in the forested snowmelt and dominated watershed were sensitive to changes in snowmelt related parameters. Alteration of the bank storage recession constant was found to significantly affect AMFs in all three watersheds. It was observed that simulation of the flood risk distribution under altered climate can be improved by modifying snow related parameters based upon the observed change in temperature from the calibration period. In flood risk studies with projected urbanization and expansion of agricultural areas, the curve number parameter should be adjusted by the proportion of change relative to the baseline (or calibration) period.