The process of land cover change prediction is generally subjected to several types of imperfections which affect the reliability of decision about these changes. Several works in literature are carried out in an attempt to mitigate the issues caused by data related imperfections. Although new prediction models are created and improvement is done for the existing ones, both the imperfection related to the input of models and its propagation through models are disregarded. To bridge this research gap, we propose a methodology that propagates imperfection throughout a model of land cover change prediction. The proposed approach incorporates three steps: 1) computing membership functions for input variables of the model of land cover change prediction, 2) applying a sensitivity analysis technique to determine which input variables are the most influential in the overall imperfection model, and 3) propagating distributions of the most influential input variables throughout the model of land cover change prediction. Experiments are made on images representing the Saint-Denis region, capital of Reunion Island. Results show the effectiveness of the proposed methodology in improving both computation time and prediction of the land cover change.