Exploring the drivers of reef island shoreline change using machine learning models.

Empirical records of reef island shoreline change show magnitude and styles of island change are highly variable over various spatio-temporal scales. However, the attribution of processes as drivers of observed change is poorly resolved. In this study, we develop machine-learning models to explore the drivers of shoreline and positional change of island footprints using multi-decadal records spanning the western-central Pacific. Our models identify a set of 'important' predictors, notably a combination of oceanographic, climatic, and local-scale morphological properties of islands and reef platforms. Additionally, we use the models to examine the interactions between these predictors. Results offer the first machine-learning models for reef island physical change, and highlight the complex relationships between a range of controls. While sea-level rise is considered a uniform threat across all islands, our results illustrate that the direct erosional response to high sea-level rise rates was attenuated in settings of 'positive' local-scale properties, such as broader reef platforms, and/or high vegetation density; underscoring the necessity for nuanced adaptation strategies that acknowledge local-scale variabilities. Results have implications for understanding attribution, developing vulnerability indices for small islands, and lay the groundwork for projections of island change as effects of climate change intensify over the coming decades.