Spatio-temporal machine learning for continental scale terrestrial hydrology