<p>Insufficient soil erosion and sediment yield data in numerous countries, with a growing demand for this information, have necessitated the utilization of erosion and sediment models. Also, accurate identification of the input factors of the models, and determining the contribution of each factor to the temporal changes of the target variable, is one of the essential steps in increasing the accuracy and efficiency of the model results. The aim of this study is to assess priorities of G2 dynamic factors using the GeoDetector method. To achieve the study purpose, the annual soil erosion and sediment yield were estimated using the G2 model which has two dynamic factors of R and V and three non-dynamic factors of S, T and L for a twenty years period (2001–2021) in the Kasilian watershed, Northern Iran. Data from MODIS, Landsat, and Sentinel-2 were used to create distribution maps for model input factors. The G2_Loss model provided soil erosion intensity maps, while the G2_sed model estimated the sediment delivery ratio and annual sediment yield. Results showed an average annual soil erosion rate of 1.69 t ha^− 1 and a sediment yield of 0.57 t ha^− 1. GeoDetector analysis revealed rainfall erosivity had the greatest impact on soil erosion, followed by vegetation retention. Higher erosion rates were found in rangelands due to slope and vegetation factors. The underestimation of observed data affected the accuracy of the model and highlighted the need for more reliable observational data. The results of this study emphasized that prioritizing dynamic factors affecting soil erosion risk is essential for better understanding the interaction of factors, focusing management and implementation measures, and achieving maximum efficiency. This management tool can be used by decision-makers in natural resource management, especially soil conservation.</p>