The CSI framework for the evaluation of binary classifications in rare events: applications to land use change models and environmental valuation

This dissertation examines the evaluation of binary classification models by analyzing current metrics and practices, proposing a novel validation framework, and exploring their practical implications for land change science and environmental evaluation. Motivated by structural issues in land change model (LCM) validation, the manuscript focuses on applications where the event of interest is rare. The first Chapter introduces the general problem of evaluating binary classifications and situates it within a broader interdisciplinary context, with particular reference to LCM. It provides a review of a literature that is dispersed across multiple disciplines, bringing together and systematizing different approaches to the evaluation of binary events. The chapter characterizes the main strands of this literature, summarizes their key critiques, and highlights the challenges arising from both the complexity of binary event evaluation and the fragmentation of existing research. The second Chapter applies concepts from classification evaluation literature to propose and explore the CSI framework as an alternative suitable for any application where the event of interest is rare. In addition, it identifies and corrects an inconsistency in the formulation of the Gilbert Skill Score (GSS). The third Chapter applies the findings and the interdisciplinary literature review to LCM validation practices. It examines how validation is currently conducted, highlighting key methodological limitations, particularly in the presence of spatial dependence and the rarity of land change. The chapter shows how the proposed CSI framework can address strongly rooted blunders in the field. The fourth Chapter presents an empirical application based on LCM, in which the proposed methodological framework is used to estimate the economic value of soil erosion. The approach combines LCM, the revised universal soil loss equation (RUSLE), and replacement cost methods to provide a dynamic and spatially explicit valuation of erosion. The chapter demonstrates how LCM can be effectively integrated into an economic valuation framework.