Distilling the Pareto optimal front into actionable insights

Abstract

Multi-objective optimization (MOO) is becoming increasingly important in environmental decision making, but interpreting highly-dimensional Pareto optimal data often constitutes a cognitive overload for both scientists and stakeholders. To address this challenge, we present PyretoClustR, a modular framework for post-processing Pareto optimal solutions. This tool aims to increase accessibility and applicability of MOO results by introducing a low-lift, iterative method to reduce the Pareto front. PyretoClustR is adaptable to various environmental datasets and decision-making scenarios, automatically selecting effective parameters for principal component analysis, clustering, and outlier handling. It produces digestible visualizations of the pruned dataset for decision-makers. We demonstrate its effectiveness using MOO results from a multifunctional landscape, highlighting trade-offs between agricultural productivity, biodiversity, water quality, and ecological flow. PyretoClustR successfully reduced the Pareto front (2419 points) to 18 representative solutions with a silhouette score of 0.33 based on decision space variables, facilitating understanding of MOO for informed decision making.