High-resolution machine learning processed satellite data reveal accelerated reef collapse

Abstract Coral reef area loss is accelerating globally, yet quantifying the mechanisms driving this loss remains challenging where monitoring is sparse. The Arabian Sea coast hosts distinctive coral communities exposed to extreme environmental gradients, including globally unique Montipora -dominated assemblages at Bar Al Hikman, Oman. Recent work documented substantial reef area loss over five decades, but temporal resolution precluded determining whether reefs lose area continuously under chronic stress or episodically following acute disturbances. Here we show, through analysis of 223 satellite images (average 5.8 observations per year) spanning 1985–2025 processed via machine learning, that reef area loss follows a sustained trajectory rather than episodic disturbance. Thermal stress metrics are strongly associated with area loss, and reef margin retreat aligns with persistent monsoon-driven wave directions rather than cyclone tracks. These findings suggest that reef area loss operates through a coupled thermal–mechanical process in which recurrent episodic thermal stress events cause coral mortality and framework weakening, and the resulting dead skeletal material is progressively removed by persistent monsoon-driven wave action, producing a continuous area loss trajectory rather than the episodic storm-driven loss previously assumed. Area loss averaged 0.10 km $$^{2}$$ yr $$^{-1}$$ before 2006, accelerating to 0.15 km $$^{2}$$ yr $$^{-1}$$ thereafter, projecting complete reef loss by 2060, some 40 years earlier than previous estimates. Our machine learning-assisted approach demonstrates that high-frequency satellite monitoring can reveal area loss dynamics invisible to conventional decadal-scale assessments, providing a framework for establishing quantitative baselines in data-sparse reef systems globally.