Focal areas: (1) Data assimilation enabled by machine learning, unsupervised learning (including deep learning), and (2) Predictive modeling through the use of AI techniques and AI-derived model components, comprising a hierarchy of models. What is the 10-year vision? The application of retrospective-predictive modeling will provide the information of what constitutes good governance for water, and what is needed for advancing water priorities and to transform water governance for the upcoming

more »

... ecades (Water Affordability..., 2020). The retrospective-predictive modeling approach is well suited for accurate forecasting of extreme meteorological and hydrological events, droughts and floods. The retrospective-predictive approach meets the FAIR principles of findability, accessibility, interoperability, and reusability. The approach will help increase the accuracy of simulations of extreme hydrological events at multiple spatial scales --from a local to a sub-watershed and to the watershed scale, and multiple temporal scales --from short-term to long-term scales. Solving the input and training data challenges through the help of machine learning algorithms for navigating software, predicting the many defects of datasets, advanced data correlation, enabling automation and providing graphical analysis of data coverage. This approach is suitable to resolve the following typical challenges of the AI/ML predictions: compatibility of data and models, portability, i.e., porting a model from one environment to another, computing power (advanced models may require large capacities of computer power), scalability (an ML environment should be able to scale up over time to meet performance and accuracy requirements), and model size (the model hosting environment should have sufficient storage and processing capabilities). The 10-year vision includes the development of a scalable open-source software for advanced analysis techniques such as data mining, machine learning, pattern scaling, and visualization will enable scientific discovery for advancing water priorities. Advanced workflows will enable the reproducibility of research results and simulations and the ability to easily apply new techniques to existing datasets. A modern and effective cyberinfrastructure for archiving, managing, analyzing, and visualizing experimental, observational, and model-generated data is critical for supporting scientific investigation of Earth system processes. Solving the AI/ML challenges of modeling and risk assessment of "extreme" water cycles through the help of self-learning algorithms for navigating software, enabling automation and providing graphical analysis of data coverage.