To address this challenge, NOAA’s Office of Water Prediction has been developing Python-based software to support regionalization of calibrated formulations and parameters within the NextGen Framework from calibrated basins to uncalibrated areas. Leveraging and expanding on the regionalization framework of the National Water Model (NWM), the NextGen regionalization capability relies on physical similarity and spatial proximity to pair donors with receivers. Physical similarity is computed from a set of hydroclimatic and physiographic characteristics defined by a conceptual framework, such as the Hydrologic Landscape Region (HLR) and the Catchment Attributes and Meteorology for Large-Sample studies (CAMELS). Several methods have been developed for donor-receiver pairing, including methods based on similarity metrics (e.g., Gower’s distance and the distance computed with unsupervised Random Forest), as well as those based on clustering, such as K-means clustering, K-medoids clustering, Hierarchical Density-Based Spatial Clustering of Applications with noise (HDBSCAN) and Balanced Iterative Reducing & Clustering using Hierarchy (BIRCH).
We hypothesize that different regionalization methods will perform differently in different regions. In this presentation, we discuss the relative merits of different regionalization methods, compare the application of these methods to several different hydroclimatic regions, analyze the donor-receiver pairing by each method, and examine the performance of streamflow simulations from applying these regionalization methods.

