An Interactive, Self-Organizing Map-based Framework for Decision-Making Under Deep Uncertainty: Insights from the Colorado River Basin

Abstract Submission: Decision-making for water and environmental systems under uncertainty often requires evaluating alternatives across large scenario ensembles that capture the range of plausible future conditions. These scenarios are defined by features (i.e., uncertainties), which can differ both in number (e.g., 3 vs. 30 features) and type (e.g., scalars vs. timeseries). As feature complexity grows, visualizing and conceptualizing the similarities and differences between scenarios becomes more challenging, complicating both the generation and evaluation of alternatives. Self-Organizing Maps (SOMs), a type of unsupervised neural network, help by organizing scenarios onto a 2D grid that preserves topological relationships, enabling intuitive, visual navigation of the uncertainty space. This work presents an interactive SOM-based decision-making framework that organizes problem-specific scenarios into a “scenario map.” Users can: a) input beliefs or projections about future uncertainties, while the SOM highlights map regions that align with those views; b) compare alternatives across specific map regions, while evaluation tools like parallel-coordinates plots update based on the selected map region; and c) conduct comprehensive analyses of alternatives by overlaying performance metrics onto the map to visually explore how scenario features relate to performance outcomes. We demonstrate the framework in a case study inspired by the development of new reservoir operating policies for the Colorado River Basin post-2026, a complex decision problem due to deep uncertainties about the basin’s future and multiple conflicting objectives. The framework helps visualize and communicate the range of potential future scenarios, while also facilitating the evaluation of alternatives across these scenarios.

Learning Objectives/Expected Outcome (Optional) :