Cross-year reconstruction of harmful algal blooms in the East China Sea with a physics-informed neural network: what a learnable cyst-germination source can and cannot identify (code)
Cross-year reconstruction of harmful algal blooms in the East China Sea with a physics-informed neural network: what a learnable cyst-germination source can and cannot identify (code)
Code archive (v1.6.0) accompanying the manuscript 'Cross-year reconstruction of harmful algal blooms in the East China Sea with a physics-informed neural network: what a learnable cyst-germination source can and cannot identify'. It reproduces the leave-one-year-out (LOYO) physics-informed neural network reconstruction of chlorophyll-a for 2021-2024, the independent GOCI-II cross-sensor evaluation without retraining, deep-ensemble uncertainty quantification with split-conformal calibration, and the source-parameter identifiability diagnostics (freeze-and-sweep affine verification and quantitative observing-system requirements). Every table and figure can be regenerated on a single consumer GPU via run_full.sh. Released under the MIT Licence.