In this work we present Transformers as an innovative model for dealing with groundwater time series and compare it to other more common prediction adapted models. Herein lies our goal to explore the applicability of the models on multivariate data and predict farther steps for 18 stations scattered in France. In order to fulfil our purpose we followed a classic pipeline for the preprocessing of the data for the feature engineering also experimenting with different optimizers helped in partially avoiding the vanishing gradient problem and there is plenty of room to experiment with the model's other settings.