Physics-inspired Machine Learning on Acoustic Data for Global Coral Reef Monitoring

Description

Coral reefs are critical ecosystems that support a third of all marine species. Yet they are rapidly degrading due to climate change, nutrient overload, and overfishing—mass bleaching and ocean acidification are weakening reef resilience. There is a pressing need for scalable, non-invasive monitoring tools that enhance our understanding of these complex ecosystems and their degradation drivers, enable early detection of reef stress, guide targeted interventions, and prove the intervention’s effectiveness.

However, current monitoring is fragmented, laborsome, lacks standardization, and thus struggles to scale. Vision-based monitoring (cameras) fails as soon as the waters are turbid. Underwater acoustic sensing offers a promising solution, as reef soundscapes can reflect ecological activity and can serve as real-time indicators of reef health. While existing research strongly points in this direction, currently used acoustic indicators are not consistent and highly depend on parameter choices.

We propose a targeted pilot study across a gradient of reef health states, combined with advanced AI analysis techniques, to generate a replicable method and first prototype of a ready-to-use soundscape analysis platform that accurately predicts the current state of a reef.
The social enterprise rrreefs AG (www.rrreefs.com) is regenerating coral reefs worldwide and offers a rrreef-care service to monitor and future-proof coral reefs for paying stakeholders. To scale this offering, rrreefs seeks to develop an acoustic sensing platform that allows continuous monitoring, can deliver early warnings and track regeneration efforts globally. To support this, a multidisciplinary team at ZHAW is building an AI-powered framework that combines ecological knowledge, physical acoustics, and advanced machine learning—including physics-informed models.