Building a composite indicator for biodiversity through supervised learning and linked indicator sets

Abstract: Understanding and predicting fate of global biodiversity amidst an increasingly complex and changing world is a major challenge facing the Earth-system science community.  Among the core research objectives within this challenge lies the ability to construct a comprehensive metric that not only faithfully quantifies the current and observed state of biodiversity, but also captures future trends that are driven by a variety of stressors across environmental, social, and economic systems. In order to give a better overview of our impact on biodiversity despite the obvious complexity inherent to the multi-sectoral nature of the problem, we have chosen to group together the indicators currently assessed and used internationally in a linked indicator set categorized according to the “Pressure-State-Response” framework. This approach stems from a desire to highlight and quantify the links between these different indicators in a logical and objective manner and allows us to construct a systematic synthesis of the key drivers of biodiversity. We develop a new methodology using predictive supervised learning to propose a statistical weighting of the linked indicator metric.