Predicting the cleaning time required to remove a thin layer of soil is a challenging task and subject of current research. One approach to tackle this problem is the decomposition into physical sub-problems which are modelled separately and the subsequent synthesis of these models. In this paper, an existing model for adhesive detachment is extended for the prediction of the cleaning time of cohesively separating soil layers. The extension is based on measurements of the pull-off forces and their correlation to the local water mass fraction. The resulting new model is validated using cleaning experiments with starch in a fully developed channel flow. Furthermore, an inhomogeneous soil distribution and its effect on cleaning results like cleaning time and removal rate is investigated. It is shown that accounting for the local soil distribution in the model leads to a significant improvement of the prediction of the cleaning behaviour.