PINN-ECS-HAB: Physics-Informed Neural Network with Strong PDE Residuals and a Learnable Cyst-Germination Source for Forecasting Harmful Algal Blooms in the East China Sea (code )

v1.3.0 (2026-08-13): resubmission to Environmental Modelling & Software. Adds scripts/interpretability_analysis.py, a post-hoc interpretability diagnostic that (i) quantifies the consistency of the prescribed cyst beds with the benthic-survey envelope and the SST shelf front (2.48x front gradient over ocean background; beds within 78 km of the front), and (ii) evaluates the transport-consistency residual of the released PINN-step3 vs PINN-nophys mean fields with a single finite-difference operator (PINN-step3 residual RMS 0.83 vs 1.29 mg m-3 d-1, 1.56x smaller, with the gap largest where MODIS coverage is thinnest), outputs processed/pinn_step3/interpretability_{metrics.json,pde_residual.png,cyst_vs_benthic.png}. README, CITATION.cff and ZENODO_GUIDE updated to the EMS resubmission. All headline numbers and data are unchanged from v1.2.1; this release only adds the interpretability artifacts required by the new manuscript figures.