Environmental Analysis Methodology

NetCDF Spatial Slicing

Unlike traditional models that average environmental data across massive regional polygons, Meduseo utilizes dynamically sliced raw NetCDF files accessed natively through xarray.

For every valid location, the pipeline maps the exact GPS coordinates and forms a dynamic bounding box using a 0.15° spatial radius. This strictly limits the environmental assessment to approximately a 15km radius around the specific coastline, effectively capturing bay-specific micro-climates.

Data Sources

We harness two main pillars of the Copernicus scientific suite:

• ERA5 Atmospheric Reanalysis: Sourced via the CDS API at 0.25° grid resolution, covering Wind, Atmospheric Pressure, Temperature, and Precipitation.
• Copernicus Marine Environment (CMEMS): Sourced natively at 0.083° resolution, extracting physical indicators like SST, Salinity, Current Surface Vector Velocities, and Oceanic Waves.

Mann-Whitney U

Because weather variables notoriously violate normal distribution assumptions, we utilize the non-parametric Mann-Whitney U test. This algorithm compares the summed rank of parameter values observed on clear days against jellyfish days.

Cohen's d (Effect Size)

While a p-value strictly specifies if an effect exists, Cohen's d establishes the magnitude of the trigger. Calculated using the difference of means relative to the pooled standard deviation, allowing us to rank occurrences drivers.

FDR Correction

Running unconstrained tests inflates the risk of false positives. We handle this by passing all base p-values through a Benjamini-Hochberg FDR correction. Only metrics satisfying q < 0.05 are tagged as significant drivers.