Accurate estimates of poplar plantations area and their distribution are not available at the French national scale due to their short rotation cycle (15 years on average) and the low update rate of existing forest databases. The availability of high spatial, spectral and temporal resolution Sentinel-2 images has opened up new opportunities for classifying poplar plantations over large areas. When supervised classification models, trained on a specific area or over a given time period, are used to predict a different region or different periods of time, the predictive performance drops off due to data shift issues. Domain adaptation techniques and in particular optimal transport (OT) are proposed in this study to overcome the transfer problem aforementioned. OT was performed between two different Sentinel-2 tiles located in the north-east and southwest of France and according to two transfer modes: spatial (i.e. same year, different sites) and spatio-temporal (i.e. different years, different sites). The main objective is to assess the potential of OT to adapt the data distribution of a source tile to better identify poplar plantations on a target tile where little knowledge is available. The results show the ability of spatial OT to improve the producer's accuracy of the poplar class up to 27% as well as an increase in the classification's overall accuracy up to 35%. The improvement is particularly important when source and target labels are considered. OT proved to be of little relevance for classes that were initially very well identified including poplar plantations, but it showed all its potential for classes that were harder to classify, which consequently improved the global classification performance.