This paper proposes to use the spatial information to augment the training corpus of BERT-based text classification models on crisis related corpora. After having shown the importance of this kind of information thanks to a neural network explicability method, we propose GeoNLPlify, a set of three data augmentation techniques based on spatial information.