Researchers used 1 million data points and a machine learning algorithm to estimate groundwater stores with higher resolution than ever before.