Project Summary
Project 2.5i.1: Hydrodynamics at the Whole-of-GBR Scale
This project aims to develop three-dimensional models that encompass the whole of the GBR at high resolution. The project is a joint CSIRO / AIMS initiative, with Richard Brinkman and Craig Steinberg contributing from AIMS. Co-funding is provided by the Reef and Rainforest Research Centre.
Project Description
Hydrodynamic models simulate the advection and mixing of water; with respect to the Great Barrier Reef (GBR), these processes are fundamental in controlling the fate and impact of freshwater, sediment, nutrients and pesticides delivered from catchments in to the receiving waters of the GBR lagoon. The modeling community is currently in a position where it is feasible to develop a whole of GBR three-dimensional hydrodynamic model that includes all of the important factors affecting currents, mixing, temperature and salinity within the GBR Lagoon and exchanges with the adjacent Coral Sea.
We propose to develop a whole-of-GBR three-dimensional baroclinic hydrodynamic model (GBR model) at a spatial resolution of approximately 1 km, with accurate boundary forcing for offshore ocean boundaries provided by a global, data-assimilating, eddy-resolving model. Such a model will underpin the future development of other essential components of a Large-Scale Water Quality (LSWQ) model, primarily sediment dynamics and biogeochemical models, and provide a capability to support the prediction and analysis of connectivity and exchange of material, including larvae, throughout the GBR.
At this spatial resolution of about 1 km, a critical challenge will be to deal with the range of spatial scales encountered within the model domain, in particular the need to resolve effects of reefs and reef passages at scales of less than the model resolution. In order to meet this challenge, an important part of this project will involve an investigation and assessment of the necessity and suitability of sub-grid parameterization schemes within a model of this spatial resolution.
As a proof of concept for a full, 3D hydrodynamic model of the GBR, we will use the model developed in this project to hindcast the circulation within the GBR lagoon during the 2009 wet season, including prediction of the trajectories and spatial distribution of major freshwater inflows during this period.
Project Methodology
In order to develop a 1 km model of the whole GBR a nested modelling approach is required, where a lower resolution model of 4 km is to be developed so as to supply the open boundary conditions for the 1 km model. This 4 km model is in turn nested within a global eddy resolving model. The 1 km and 4 km model are to use the finite difference model SHOC. The development of these models will build on knowledge and experience generated from the MTSRF climate downscaling project (MTSRF Project 2.5i.1) which has implemented SHOC on the GBR at a grid resolution of between 2 and 5 km from Fraser Island to the Daintree.
The 4 km regional grid covers the GBR section of the continental shelf, slope and deep ocean from Moreton Bay to the mainland of Papua New Guinea, extending eastwards into the Coral Sea Territories a sufficient distance to avoid the topographical complexities of the Queensland and Marion plateaus (see Fig. 1). Within this regional grid, a finer 1 km resolution model will be nested to cover the continental shelf areas (Fig. 2). Accurate forcing data for the offshore ocean boundaries of the regional grid will be provided by the data-assimilating eddy-resolving models of the BlueLink initiative. The vertical resolution of the model will vary with depth, but will have ~50 vertical layers, with appoximately 1 metre resolution near the surface.
Limited model validation will be undertaken utilizing ocean current and temperature data from the GBR Ocean Observing System (GBROOS) and available remote sensing imagery. The nested suite of models will be used to perform a preliminary hindcast of the 2009 wet season to simulate the circulation within the GBR lagoon, including prediction of the trajectories and spatial distribution of major freshwater inflows during this period. Additionally, these models will be run in near real-time, so that a current description of sea surface height, flow patterns, temperature and salinity is available.
Once established, the models will be run using contemporary ocean and atmospheric forcing and all available river inflow data for catchments draining into the GBR. Atmospheric forcing products (wind, pressure, heatflux) will be supplied by the MesoLAPS atmospheric model at 1/8 degree resolution. This modelled product is preferable to spatially interpolated meteorological data supplied by BoM (Bureau of Meteorology) weather stations for surface forcing, since its spatial detail is superior. MesoLAPS provides wind, mean sea level pressure, cloud amount, air temperature and dew point temperature. From these variables, bulk schemes (Kondo, 1975) are used to compute sensible and latent heat fluxes, and black body radiation may be used to compute long wave radiation (Zillman, 1972). The sum of these and computed short wave input provides a net heat budget.
The regional model was forced with global ocean and atmospheric model products. The regional model was nested in BRAN2.3 (Schiller et. al, 2008); these outputs considered the best global product to date suitable for forcing the nested suite. BRAN2.3 is a data assimilating global model with 10 km resolution in the Australasian region, using MOM4p1 as the code base. The near real-time implementation of the models will use the operational version of BRAN operated by BoM; OceanMAPS. These data were used as initial conditions for temperature, salinity and sea level, and boundary forcing on the 4 km grid for temperature, salinity and velocity. An upstream advection open boundary condition (OBC) was used for temperature and salinity. The velocity open boundary condition follows the methodology of Herzfeld (2009), where a local flux adjustment is used to prevent long term basin-wide divergence, which may lead to filling or emptying of the domain. A tidal signal using the OTIS tidal model was superimposed on the low frequency sea level oscillation provided by BRAN2.3 on the regional grid open boundary. This tidal signal was introduced via the local flux adjustment.
A critical challenge for this project will be to deal with the range of spatial scales encountered within the model domain, in particular the need to resolve effects of reefs and reef passages at scales of less than the model resolution. Correctly incorporating the contribution of sub-grid scale processes is crucial for accurate simulation of the hydrodynamics, for example providing correct exchanges through dense outer-shelf reef matrices, and also for subsequent components of the LSWQ model, such as the biogeochemical consequences of enhanced residence times for water trapped within a reef boundary layer. This challenge will be addressed through the application of high-resolution implementations numerical models to resolve these effects, applied to both simplified test cases for which analytical solutions exist and ‘real-world’ examples. This approach will allow an assessment of suitability of sub-grid parameterization schemes for future incorporation into the whole of reef model.
At a spatial resolution of approximately 1 km, a whole of GBR model represents a major step forward in both spatial coverage and resolution, but also presents a significant increase in the complexity of the modeling problem. At the proposed resolution, a computational domain of similar extent to that shown in Fig. 1 will have around 1 million surface elements and a total of approximately 19 million active computational cells. This size of problem is approaching the size of many global ocean modeling problems (for example, the BlueLink OFAM model has approximately 1.2 million surface elements). Managing a problem of this complexity and utilizing new supercomputing infrastructure will require the development of new tools for tasks such as grid preparation, diagnosis of model performance, analysis of model output and optimising the computational efficiency for the chosen computing platform.