[Objectives] The parameter calibration of the CASC2D hydrological model is mainly based on manual trial-and-error methods. It lacks a global sensitivity analysis of model parameters and the identification of relationships between parameters and simulation indices based on such analysis. Therefore, there remains considerable room for further exploration and discussion regarding parameter calibration methods and practices for the CASC2D hydrological model. [Methods] The CASC2D model is relatively suitable for flood forecasting in small semi-arid watersheds. This study selected the region upstream of the Suyukou hydrological station in the eastern foothills of the Helan Mountains as the study area. The Sobol index method, a representative global sensitivity analysis approach, was employed. Independent and global sensitivity analyses were conducted for eight key parameters of the CASC2D hydrological model, based on three performance indicators derived from simulation results: peak flow timing, peak discharge, and coefficient of determination. These analyses identified the correlations between model sensitive parameters and model-simulated peak discharge, peak flow timing difference, and coefficient of determination, providing references for model parameter calibration. [Results] The three parameters with the greatest global influence on the peak flow timing were saturated hydraulic conductivity (K_s), channel roughness coefficient (n_c), and soil water deficit (M_d). The peak flow timing of flood was mainly related to the infiltration calculation in the model. During the flow concentration process, channel routing played a controlling role, while overland flow routing played a supporting role. Additionally, the peak flow timing of flood was negatively correlated with river width (L) and vegetation interception (I), and positively correlated with n_c, overland flow roughness coefficient (n_s), K_s, capillary pressure head (H_c), and M_d. The parameters that had the greatest global influence on peak discharge and coefficient of determination were K_s, H_c, and n_c. Flood peak discharge was jointly influenced by infiltration characteristic parameters, overland flow routing parameters, and channel routing parameters. Among them, infiltration characteristics played the dominant role, while in the routing process, overland flow routing was relatively more influential. Coefficient of determination was mainly related to infiltration characteristics and channel routing parameters, with the former being dominant. Additionally, peak discharge showed positive correlations only with n_c, n_s, K_s, H_c, and M_d. Coefficient of determination was negatively correlated with I and K_s, but positively correlated with L, n_s, H_c, and M_d. [Conclusions] This study further explores and supplements research on global sensitivity analysis of CASC2D hydrological model parameters. It proposes the types and sequence for adjusting eight key parameters in response to errors in peak discharge, peak flow timing, and coefficient of determination during initial calibration. The study also suggests reasonable increase or decrease ranges for each parameter based on their sensitivity to different indicators. These findings provide references for properly selecting the directions and ranges of parameter adjustments during calibration. Due to the interactions among parameters, the selection of adjustment directions and ranges during calibration should be based on each parameter’s independent sensitivity and degree of interaction obtained from the first-order and total effect indices. The research findings can provide references for manual calibration efforts and improve the efficiency of parameter calibration. Furthermore, given the current lack of research on automated calibration for this model, the findings offer strategic guidance for the development of automated calibration algorithms.