Abstract
sWe evaluated the diagnostic and predictive performance of the coupled General Ocean Turbulence Model–Framework for Aquatic Biogeochemical Models–Water Ecosystems Tool (GOTM–FABM–WET) in shallow Kun Cheng Lake, China. Monthly near-surface observations supported model calibration during 2021–2023 and out-of-sample ecological validation during 2024–2025. Model skill was assessed through one-at-a-time parameter screening, moment-independent Delta sensitivity analysis with sample-size convergence diagnostics, sequential multi-parameter calibration, and comparisons against statistical and nutrient mass-balance benchmarks. Water temperature calibration was strong (Nash–Sutcliffe efficiency, NSE = 0.933), although temperature observations were available only for 2021–2023. Dissolved oxygen (DO) exhibited modest validation skill (NSE = 0.290). Nitrogen parameter refinement improved correlation direction and simulated variability, yet the simulated seasonal TN peak remained five months earlier than observed. Total nitrogen (TN), total phosphorus (TP), and chlorophyll a (Chl-a) all retained negative validation NSE values. In particular, Chl-a calibration remained inadequate (calibration NSE = −0.405; validation NSE = −0.481), and its dedicated calibration stage retained the incoming seed parameters. Only 0.50–0.78% of evaluated candidates satisfied the cross-variable calibration guardrails. Temporally constant nutrient boundaries, coarse TP reporting resolution, and frequent numerical state repairs highlight key priorities for model improvement. Overall, the current evidence supports retrospective near-surface thermal diagnosis and conditional interpretation of broad seasonal DO patterns, whereas reliable nutrient and algal prediction will require time-varying inflow and nutrient chemistry, process-level field observations, and independent verification of seasonal timing and spatial representativeness.