Skill assessment of an integrated modeling system for shallow coastal and estuarine ecosystems

The predictive skills of an integrated physical–biogeochemical modeling system (CH3D-IMS) for shallow estuarine and coastal ecosystems are assessed using available field data in the Indian River Lagoon estuarine system, Florida during 1998–2000. The cornerstone of the modeling system is the circulation model CH3D (Curvilinear-grid Hydrodynamics in 3D), which is coupled to models of wave (SMB), sediment transport, water quality (nutrients: N, P, and Si, three phytoplankton species, zooplankton, and dissolved oxygen), light attenuation, and seagrass. To resolve the complex geometry and bathymetry of the estuarine system, the modeling system uses a boundary-fitted non-orthogonal curvilinear grid in the horizontal direction and a terrain-following sigma grid in the vertical direction. While water level and salinity data were collected continuously (at 15-min intervals) at 10 fixed stations, most water quality data were collected at much longer time scales (bi-weekly to quarterly) during ship surveys at more than 30 stations. Sediment-water quality data were collected at 24 stations once in 1998.Model skills for hydrodynamic and water quality simulations are assessed in terms of the absolute relative errors and the relative operating characteristic (ROC) scores. Both methods indicate that the modeling system has skills in simulating water level, salinity, dissolved oxygen, chlorophyll, and dissolved nutrients, with the ROC score between 0.6 and 0.862, indicating skills for most of the variables. Skills for simulating total suspended solids (TSS) and particulate nutrients are lacking, with ROC score and: between 0.5–0.6. Simulated diffuse attenuation coefficient, which depends on TSS, chlorophyll a, and dissolved organic matter, has an ROC of 0.55. Using high frequency time-varying field data collected during two episodic events in the study period, the skills of CH3D-IMS improved significantly for both TSS and particulate nutrients. Model skills for particulate nutrients are found to be very sensitive to nutrient concentrations in the sediment column which were sampled in 1998. Sensitivity tests were conducted to determine the sensitivity of model skill to bottom boundary condition, open boundary condition, and model parameters/coefficients. Results suggest that model skills can be improved with more detailed sediment-water quality data, addition of a coastal ocean domain, and improved knowledge of model parameters/coefficients. Parameter estimation method (e.g., modified Gauss–Newton method) can be used to reduce the effort required to select optimal model coefficients for the baseline simulation.