Authors: Rhys Pulling – Colorado School of Mines
Title: A working basin-scale evaluation of the pros and cons of a using heterogeneous multi-model mosaics for streamflow simulation
Presentation Type: Poster
Abstract: This research is part of a larger “multi model mosaic” (MMM) project designed to evaluate the implementation of a heterogeneous selection of catchment models across river basins, including diverse process-based and machine learning formulations. This model tapestry concept is a guiding principle of the Next Generation Water Resources Modeling Framework (NextGen). NextGen is designed to optimize streamflow prediction by enabling the selection of fit-for-purpose configurations of hydrologic models and their components at the catchment level, combining across a river basin into a unified system: that is, its structure allows for the training of different models on different sub-catchments within a single basin. We put this concept to the test in a medium sized river basin (100s to 1000s of catchments at the NextGen Hydro Fabric level), assessing whether heterogeneous selections of models can actually lead to better streamflow simulations within the current NextGen system. Experiments to date include using the Conceptual Functional Equivalent (CFE) model, Noah-OWP-Modular (NOM), and Long Short-Term Memory (LSTM) neural network models within a basin mosaic. This presentation gives a progress update on the work done for our chosen basin using the NextGen system. We report on the various challenges and insights gained from the effort to use NextGen (running on a Linux HPC system) to try to achieve its design goals.