Large sample hydrology datasets such as Caravan provides the community with hydrometeorological information and catchment attributes for many catchments in the world and offers the opportunity for hydrological research. However, there are considerable differences between the forcing data of Caravan compared to the CAMELS datasets, especially with potential evaporation. This can lead to wrong conclusions on catchment hydrological drivers and affect regionalization. This papers shows the important of robustness of large sample datasets and the need to keep assessing that.