Abstract: Knowledge of land cover dynamics and driving forces is a fundamental tool for landscape planning and management. Nevertheless, this understanding is often limited by the paucity of accurate land cover data. In this sense, remote-sensing offers the possibility of acquiring detailed land cover inventories by applying different methods of image classification. However, in heterogeneous and changing landscapes, these data may be insufficient to detect temporal changes (and their causes) because of the uncertainty associated with misclassification and the spatio-temporal variability of change patterns. In this work, we present a multi-temporal uncertainty-based method that incorporates regression models to establish the risk (probability) of land cover change (RLCC), as a function of a set of environmental and socioeconomic driving factors. After filtering out uncertainty for dependent variables (land cover changes), the accuracy of the models increased and regression yielded more parsimonious models that identified the relevant predictors more efficiently. Considering all land cover changes as a whole, drivers relating to the physical environment (i.e., soil properties, accessibility, altitude, slope, solar radiation and rainfall) were more frequently selected than those related to agriculture, society or economy, which may be due to the poor quality of the available socioeconomic data at the municipality level. When analysing changes separately, several differences appeared (e.g. woody vegetation cover was related with fire events and water availability or human management with forest expansion). Our methodological approach has demonstrated that uncertainty plays an important role in model characterisation and identification of potential drivers of change. [Copyright &y& Elsevier]