Sandtank-ML: An Educational Resource at the Interface of Hydrology and Machine Learning

Machine learning technology is dramatically transforming the world today. Within the realm of hydrology, machine learning (ML) is being mobilized to improve hydrological forecasting within the context of unprecedented climatic shifts. However, despite its increasing importance and utility, there are few educational tools and resources that engage with machine learning in general and even fewer that exist at the interface of hydrology and machine learning. Our team has developed a web-based educational application, Sandtank-ML, that allows users to gain an understanding of basic machine learning concepts, as well as the way in which machine learning can be used to understand and address hydrological processes and challenges. This application builds upon the ParFlow Sandtank (https://sandtank.hydroframe.org/), an interactive computer simulation of a physical aquifer model. The ParFlow Sandtank allows users to explore the subsurface, controlling various inputs, visualizing outputs in real-time, and using tools to evaluate factors that impact real hydrological systems. Using the output of the ParFlow Sandtank as the prediction goal, Santank-ML allows users to run different machine learning models and manipulate training sets and other variables to explore how particular decisions impact model accuracy. In doing so, the platform provides an easy to navigate tool for learning basic ML concepts. Sandtank-ML was developed as part of the HydroGEN project, a multi institutional and interdisciplinary effort to create an easily accessible and user-focused machine learning platform for use by water managers and other natural resource practitioners. Sandtank-ML will be used in a wide variety of educational settings in order to increase knowledge and engagement with ML and its hydrological applications.