Jellyfish blooms can significantly impact marine food webs, disrupt industries like tourism and fisheries, and pose serious public health risks. To effectively manage these events, we must move beyond generalizations and pinpoint the precise regional forces driving the occurrence of hazardous species like the Portuguese man o’ war (Physalia physalis).

Knowledge gap

Despite being a cosmopolitan and highly hazardous species, knowledge about the ecology of P. physalis remains remarkably scarce. This lack of ecological understanding severely limits our ability to predict where and when blooms will occur, hindering the development of targeted, effective management strategies.

Main approach

We developed two independent Boosted Regression Trees (BRT) machine learning models to analyze two contrasting ocean regions: the temperate North Atlantic (Azores, Portugal) and the Southeast Pacific (Australian East Coast).

The models fitted long-term field observations of P. physalis occurrence against high-resolution environmental data, including:

  • Sea surface temperature
  • Primary productivity (food availability)
  • Wind speed and direction (passive transport)


Technological challenge - how we tackle the study

BRT models excel at handling complex, non-linear relationships and sparse occurrence data, which is common in monitoring marine life. This AI approach allowed us to move beyond simple correlations and use hindcasting to predict the probability of man o’ war occurrence over a 30-year period (1993–2021). This decadal-scale prediction revealed long-term trends and fluctuations that would have been invisible using traditional statistical methods or shorter datasets.

Main finding

The primary drivers of P. physalis occurrence are highly region-specific, challenging the notion of a single global trend for jellyfish blooms:

  • Shared Drivers: In both the Azores and the Australian East Coast, wind direction (passive transport to shorelines) and primary productivity (food availability) were the most important drivers.
  • Australian Driver: On the Australian East Coast, warming conditions (temperature) emerged as a critical additional driver, significantly increasing the probability of occurrence.
  • Contrasting Long-Term Trends: While the Azores showed a multi-decadal fluctuation in occurrence (sigmoidal trend), the Australian East Coast exhibited a significant and continuous increase in occurrence over the entire 30-year period.


Main implications

Our findings provide a clear baseline for data-driven coastal management strategies. The emphasis on regional dynamics—particularly the role of temperature in Australia—underscores that a one-size-fits-all approach is ineffective.

Moving forward, we must:

  • Develop Regional Strategies: Management policies must be tailored to region-specific environmental dynamics (e.g., wind patterns, current systems).
  • Mitigate Climate Risks: Given the positive link between warming and blooms in the Southeast Pacific, P. physalis emerges as an important candidate species to track the effects of climate change.
  • Implement Early Warning Systems: Predictive models like the ones developed in this study can be instrumental in creating real-time early warning systems to minimize the socio-economic and public health impacts on tourism and coastal communities.