Harmonic water-level baselines: a one-year fit window is exactly the wrong length
One year is not "a full cycle": it is the worst possible place to cut the fit window.
Why we will not publish a retro-routing gain percentage
+7.01 % to +4.57 % to +0.98 %: every methodological tightening ate into the gain. The "star vessel" was a grid artefact.
NetCDF or Zarr: chunking depends on the workload
The same ERA5 dataset, rechunked three times, produces three different winners depending on the read scenario.
Sargassum landfall alert at J+3: detection, advection, and confidence level
A high p_mean is worthless without knowing whether the ensemble members actually agree.
DNV-ST-N001 operability: what actually closes a Channel cable-lay window
An "Hs 2 m" proxy does not tell you what closes the window. Across 453 real transits, we named the limiting mode, season by season.
Validating a routing ETA: replaying 800 real AIS voyages
800 real voyages replayed, 2x2 ablation: the AIS polar pays off, currents alone do not.
Auditing a wave model: ERA5 and MFWAM against 11 altimetry missions and CANDHIS buoys
Before basing a weather window on a reanalysis, we checked it against eleven altimetry missions and the CANDHIS network — across three corridors.
Recalibrating Kwon on five million AIS positions
We check Kwon’s empirical formula against speed polars built from 5 million AIS positions, and publish the resulting three-tier production fallback.
Transatlantic weather routing: a case study across 28 winter departures
28 winter departures, two baselines, one honest negative result: optimized routing gains time on the Rotterdam–New York crossing, not additional safety.
Validating a seabed mobility atlas without ground truth
Seabed mobility cannot be measured: there is no ground truth. So we break the chain apart and attack each link with an independent source — SHOM currents, 908 observed dunes, a reference d50. Including when the test fails.
Submarine cable: seabed mobility atlas and optimal corridor
A route that looks perfect on the bathymetry map can cross mobile seabed. Here is how we map sediment mobility and extreme scour, then compute the least-risk cable corridor between two landing points.
AI coastal forecasting: combining WW3, CROCO and CMEMS
No single model fully describes the coastal zone. Here is how we combine WW3, CROCO and CMEMS, then refine the result with deep learning to gain resolution without resorting to a supercomputer.
WW3, CROCO, CMEMS: which ocean data for which need?
Waves, currents, water level, temperature: each variable has its source. A concrete guide to choosing between WaveWatch III, CROCO and Copernicus Marine by the resolution, extent and budget of your project.