Predicting jellyfish risk for tomorrow: A forecasting model for beachgoers
Is it possible to anticipate the arrival of jellyfish on our coasts 24 hours in advance? This has always been one of Meduseo's main goals.
Until now, information relied solely on immediate observations by beachgoers once on the beach. Meduseo takes a major step forward by deploying its prediction engine. By combining community reports with marine weather and physical oceanography data, our models are now able to assess the risk of jellyfish arrival for the next day (D+1) with sufficient precision to make it available to you.
What exactly does the model predict?
-
Discomfort Prediction: The model predicts jellyfish presence at a level likely to disrupt swimming (appearance of multiple individuals or a swarm situation).
-
"No Jellyfish" Situation: When the model indicates a "No jellyfish" situation, it means general swimming conditions are favorable. However, as in any natural marine environment, the presence of an isolated and rare individual remains possible without altering overall swimming.
A unique model per city: The coastline is not a straight line
Modeling research shows that there is no generic algorithm blindly applicable to the entire coastline, which is completely natural. Each coastal sector has its own geometry, wind exposure, and unique maritime characteristics. That is why each city has its own dedicated forecasting model, developed on the specific history of its coastal basin.
The algorithm relies daily on:
Marine Weather & Physical Oceanography
Sea surface temperatures, ocean current vectors, wind direction and speed, wave height and period, as well as atmospheric pressure.
Adjacent Spatial Dynamics
Analysis of observed presence in neighboring maritime sectors.
Why are "absence" reports crucial?
To learn how to recognize risk situations, the algorithm needs equally precise knowledge of days when there is a total absence of jellyfish or strandings.
Reports of no jellyfish (zero individuals observed) are just as crucial for model learning as presence reports.
Without this absence information, the algorithm cannot balance its decision thresholds. For a new city to benefit from its own predictive model, the community must accumulate a rich and balanced history of daily observations.
Validated statistical performance: Understanding our results
To measure the effectiveness of our models on new seasons (notably 2025 and mid-season 2026), we use two key indicators:
Measures the percentage of correct predictions. An accuracy of 80% to 83% means that in more than 8 out of 10 cases, the model perfectly anticipates the actual situation on the beach the next day (whether an alert or a calm day).
Evaluated on a scale from 50% (pure chance) to 100% (perfect model), this index measures the model's ability to distinguish a risk day from a jellyfish-free day without hesitation. A score above 80% confirms strong operational maturity.
Results achieved in our pilot sectors
For now, the forecasting tool is available in 4 pilot Mediterranean cities:
Marseille
Bouches-du-RhôneSix-Fours-les-Plages
VarBandol
VarSaint-Cyr-sur-Mer
VarRollout and future prospects
Meduseo's predictive modeling aims to be progressively extended to other coastal cities as soon as report density and balance allow.
By continuing to share your daily observations on the website and app, whether there are jellyfish or none at all, you directly contribute to developing this forecasting system!