Physics-Guided Multi-Fidelity Surrogate and Regime-Aware Evolutionary Optimization for Climate-Adaptive Eco-Hydrological Wetland Restoration

Overview This record contains the de-identified eco-hydrological restoration benchmark used to evaluate PG-RAMEO (Physics-Guided Regime-Aware Multi-objective Evolutionary Optimization), a surrogate-assisted many-objective framework for climate-adaptive wetland restoration. The benchmark couples two-dimensional Saint–Venant shallow-water hydrodynamics with multi-decadal vegetation, salinity and soil-organic-carbon succession across 240 wetland restoration units spanning three geomorphological archetypes (Floodplain Marsh, n = 96; Coastal Estuarine, n = 84; Depressional Peatland, n = 60) under three hydro-climatic forcing regimes (Historical, CMIP6 SSP2-4.5, CMIP6 SSP5-8.5). Source data and de-identification. Hydrological boundary conditions, bathymetric depths, porewater salinity gradients and archetype proportions are anchored to three public monitoring repositories: the USGS Everglades Depth Estimation Network (EDEN) and NWIS (sofia.usgs.gov/eden, waterdata.usgs.gov/nwis), the CAMELS catchment dataset (ral.ucar.edu/solutions/products/camels), and the Global Lakes and Wetlands Database GLWD v2 / HydroSHEDS (hydrosheds.org/products/glwd). All station identifiers and geographic quasi-identifiers have been replaced by randomized hash codes (WRU_xxxx), calendar dates are omitted, and the spatial partition is encoded in the split column (train = 144, val = 36, test = 60). Contents data/processed/ — granular evaluation records (8 files, 8,315 data rows): wetland_basins_metadata.csv — 240 rows. Physical, hydrological and biogeochemical attributes of every restoration unit: area, drainage area, cell count, control structures, inflow, mean depth, hydroperiod, reference evapotranspiration, salinity, initial biomass, soil organic carbon, Manning roughness, bathymetric discontinuity, critical salinity (22.5 ppt) and anoxic (42.0 d) thresholds, and sensor missing rate. method_basin_regime_evaluations.csv — 5,760 rows (240 units × 3 regimes × 8 methods). Four-objective hypervolume (hv), igd_plus, candidate and feasible counts, ecological regime violation counts by mechanism (salinity_violations, anoxia_violations, desiccation_violations), surrogate state errors (rmse_h, rmse_biomass, mass_balance_err), individual objective scores (obj_hsi, obj_ncsr, obj_faee, obj_lcec), and computational speedup. seed_level_test_runs.csv — 720 rows (30 seeds × 3 regimes × 8 methods) on the held-out test split, the input to all paired significance tests. ablation_basin_evaluations.csv — 1,440 rows across the eight variants (Full, Abl-C1, Abl-C1b, Abl-C2, Abl-C2a, Abl-C2b, Abl-C3, Backbone). surrogate_operator_grid.csv — accuracy and mass-balance consistency of seven neural and statistical operator architectures. sensitivity_sweep_grid.csv — sweeps over flux weight λflux ∈ [0.0, 2.0], high-fidelity query ratio ρHF ∈ [0.05, 0.20], repair step ηrep ∈ [0.02, 0.40] and solver budget BHF ∈ [60, 480]. pareto_front_solutions.csv — representative non-dominated solutions per method. regime_shift_trajectories.csv — monthly coupled state trajectories (S(t), h(t), B(t)) over a 36-month SSP5-8.5 compound drought–flood sequence, comparing unrepaired offspring, random-mutation repair and CR-FEH directional physics-Jacobian repair. data/derived/ — the summary tables behind every reported result (7 files, 182 rows): main_table.csv (primary comparison on val and test), regime_archetype_table.csv (subgroup breakdown), ablation_table.csv, surrogate_consistency_table.csv, sensitivity_table.csv, dataset_summary_table.csv, and failure_mode_breakdown.csv (violation counts by physical failure mechanism under SSP5-8.5). Every file in data/derived/ is regenerated deterministically from data/processed/ by the companion code record. Reuse Released under CC BY 4.0. If you use these data, please cite this record together with the associated article. The columns obj_hsi, obj_ncsr, obj_faee and obj_lcec denote habitat suitability index, net carbon sequestration rate, flood attenuation efficiency and lifecycle engineering cost respectively; ehr_fr is the ecological regime feasibility ratio.