[Objective] Accurately obtaining slope gradient data of the rolling hilly region in the Chinese Mollisol area is crucial for the quantitative evaluation of soil erosion. [Methods] This study aimed to address the issue of slope gradient reduction when extracting farmland slope gradient in the rolling hilly region in the Chinese Mollisol area using the currently available 30 m resolution Digital Elevation Model (DEM). To achieve this, a 5 cm resolution DEM was generated based on drone survey images and resampled to obtain 1 m, 5 m, and 12.5 m DEM-resolutions. Combined with the free downloaded 30 m resolution DEM, five groups of different DEM-resolutions were obtained. A comparative study was conducted to analyze the difference between the slope gradient information extracted by different DEM-resolutions and 5 cm resolution DEM to determine the best resolution of DEM for extracting slope gradient in the study area. Based on the histogram matching algorithm, a slope gradient conversion model between 30 m and the best resolution DEM was fitted for each slope gradient segment. [Results] (1) The slope gradient information extracted by the five groups of DEM resolutions showed that the frequency distribution of slopes gradient extracted by 1 m and 5 m DEMs have strong similarity with those extracted by 5 cm resolution DEM. Considering that the resolution of 5 m DEM corresponds to that of a 1:10000 scale topographic map, 5 m was determined as the best DEM resolution for building the slope gradient conversion model. (2) Based on the histogram matching method, a univariate linear model and a univariate quadratic non-linear model of slope gradient conversion between 30 m and 5 m resolution DEMs were constructed for each slope gradient segment. It was appropriate to select the linear slope gradient conversion model when the ground slope gradient was less than 7°, while it was appropriate to select the non-linear slope gradient conversion model when the ground slope gradient was greater than 7°. (3) After optimization by linear and non-linear slope gradient conversion models, the frequency distribution of slope gradients was basically similar to that of 5 m resolution, and the covariance and correlation coefficient were greatly improved. This indicated that the slope gradient information extracted by 30 m resolution DEM could truly reflect the ground undulation features after model conversion, and the optimization effect of the non-linear slope gradient conversion model was better. [Conclusion] The constructed low-high resolution slope gradient conversion model provides method support for obtaining the real slope gradient of the ground in the study area.