mean absolute error
mean absolute error at World Data Ocean is a file of 3 stories. The newest of them: “Spectral confidence weighting sharpens trajectory estimation for ship magnetic fields”, “AI and oceanography converge to predict coastal cesium levels”, and “Validated vision model counts deformable shrimp larvae with measurable accuracy”. Spectral confidence weighting sharpens trajectory estimation for ship magnetic fields. The fastest cesium decay is in the Bohai Sea; the slowest is in the South China Sea. A Global Hub for Ocean Intelligence 4 World Data Ocean is a centralized digital platform where researchers, scientists, and ocean enthusiasts converge to… The list below is every mean absolute error story on World Data Ocean, newest first.
Spectral confidence weighting sharpens trajectory estimation for ship magnetic fields
Spectral confidence weighting sharpens trajectory estimation for ship magnetic fields. Shaft-rate electromagnetic fields, generated by the coupling of ship currents with propulsion shaft motion, offer stable line-spectrum characteristics for passive localization. Yet, real marine environments create pronounced quality differences among multi-station observations. The proposed method constructs a spectral-peak prominence index to adaptively weight bearing lines, suppressing degraded data. At 5 dB SNR, errors drop dramatically, mean absolute error falls from 23.48 to 11.22 meters. This is reliability-aware ocean intelligence.
AI and oceanography converge to predict coastal cesium levels
The fastest cesium decay is in the Bohai Sea; the slowest is in the South China Sea. A hybrid TRR, LSTM, TRR framework now reconstructs and predicts 137Cs activity in China's coastal waters, integrating a bound-constrained exponential baseline with an LSTM residual network. This is measurable progress: long-term projections carry 90% ensemble intervals, not point estimates. For deeper context on coastal modeling, see our related work on refining tidal models through calibrated bottom friction coefficients.
Validated vision model counts deformable shrimp larvae with measurable accuracy
Counting deformable shrimp larvae has always been a problem that resists automation. Their bodies flex into irregular shapes, and hundreds cluster together in a single frame, confusing conventional detectors. DeformDANet solves this with geometry-adaptive feature extraction and a density-aware assignment strategy, achieving 94.5% mAP@50 with just 1.38 million parameters. That efficiency matters for real hatchery deployment. The result is a validated, measurable path toward intelligent aquaculture management, complementing our coverage of sustainable practices in "Microalgae-Larvae Carbon Transfer."