Regionalization helps scientists run water models in places where there's little data by borrowing information from similar, well-studied areas. This is especially important for large-scale modeling efforts like NOAA’s NextGen system, which aims to provide flexible and accurate water predictions across the U.S.

We developed a Python tool that automates regionalization in two main steps. First, it chooses the best model setup for larger areas based on past calibration results. Then, for places with no direct data (called 'receivers'), it finds similar, calibrated places ('donors') to copy model setups and parameters from. To find the best donor-receiver matches, the tool uses machine learning methods that look at how similar places are—both physically and geographically.

We demonstrate how this tool works, explore the pros and cons of different methods. By looking at how donor and receiver areas are paired and how well the model predicts streamflow, we gain insight into which regionalization strategies work best.