The eReefs suite of marine models support monitoring and management of water quality in the Great Barrier Reef (GBR). The suite includes a biogeochemical model that relies on an underlying hydrodynamic model to provide information on river inflows, currents, upwelling and mixing processes. The models are computationally intensive and water quality predictions are arguably most accurate when driven by underlying physical, rather than biogeochemical processes. Here, I present an eXtreme Gradient Boost (XGB) model to predict chlorophyll concentrations in the nearshore GBR using features derived from the hydrodynamic model (CSIRO Coastal Environmental Modelling, 2025) and trained on in situ chlorophyll data (Moran et al., 2025). Spatial and temporal blocking and leave-one-year-out hindcasting are used to avoid pseudo-replication and ensure robustness.
The XGB model can be used to predict time-series at a monitoring site or to spatially map predicted chlorophyll. Performance metrics compare favourably with equivalent metrics for the biogeochemical model, however the two models differ in their predictions. The ML approach can complement the biogeochemical model by providing faster and potentially more accurate water quality predictions, while the biogeochemical model is better able to predict likely responses to changes in catchment management (Baird et al., 2021) and other long-term change.