Abstract: The calibration processes of current distributed hydrological model has been a critical issue in data-scared or ungauged regions. In this study, we setup a hydrological model, in which, the variable saturated zone concept originated from the real-time interactive basin simulator is applied for runoff generation and the grid water droplet method for flow concentration. We also proposed a method for parameter estimation based on the characteristics of underlying surface. Based on field infiltration experiments and parameter sensitivity analysis, the quantitative statistical relationships were built between two sensitive parameters (surface saturated hydraulic conductivity K_0z, coefficient of attenuation of saturation hydraulic conductivity with depth f) and the topographic parameters and soil types. The overland confluence parameters were determined by field overland flow observation experiments. The proposed parameter estimation method was verified in selected basins. Our results showed that: ① The proposed method for K_0z estimation contributes to a better modeling performance for flood simulation in Jiangwan experimental watershed, the average Nash-Sutcliffe efficiency coefficient increased from 0.82 to 0.86, and the average absolute values of peak and flood volume errors decreased by 2.2% and 0.95%, respectively, but the average absolute value of the peak present time error increased by 4% (still controlled within 2 h). ② Using the measured flood data of 14 basins such as Jiangwan to calibrate the parameter f, we established the quantitative relationship between the calibrated parameter f and the soil type data of different depths was built in 14 basins including Jiangwan, and further tested in other six basins. The parameter f estimated by the soil type data of different depths was very close to that determined by traditional model calibration processes, the average absolute relative error is 2.8%, the average Nash-Sutcliffe efficiency coefficient of the flood simulation is 0.83, and the average absolute values of flood peak error and flood volume error were 10.07% and 6.86%, respectively, and the average absolute peak present time error was 2.61 h. Our results indicated that the proposed methods for determining sensitive runoff generation parameters are applicable in data-sparse areas and could provide a better or comparable parameter estimation and flood simulation than that determine by field measurements, model calibration, remote sensing data estimation, and other methods.