Evaluating a prototype ensemble water quantity and quality forecasting system for the Fitzroy River Basin

The eReefs initiative is developing a series of marine hydrodynamic and biogeochemical models that will model and provide forecasts of rainfall and flooding impacts on the Great Barrier Reef. These models require real-time predictions and forecasts of riverine inflows and associated concentrations of fine sediments, speciated nutrients and carbon at each time step. This paper describes and evaluates one possible approach to the generation of water quantity and quality predictions and forecasts by linking ensemble streamflow forecasts and empirical Generalised Additive Models (GAMs). Forecasts of daily sediment and nutrient concentrations are generated by forcing GAMs with hourly streamflow forecasts that have been aggregated to daily totals. The streamflow and water quality forecasts are evaluated for over a 24-month period concluding in December 2013. The ensemble streamflow forecasts have considerably lower errors than simple persistence, which is used as input for the prototype marine models in forecasting mode. This suggests that marine modellers can potentially improve their simulations by using the streamflow forecasts in place of simple persistence. The ensemble forecasts of nutrient concentrations however display large errors, often significantly overestimating the observed values, which may limit their value for marine modelling. Errors in sediment and nutrient concentration forecasts, and the forecast uncertainties tend to be largest when the GAMS are extrapolating beyond the range of observations used to fit the GAMS model. Therefore improvements in the performance of sediment and nutrient concentration forecasts are most likely to be realised by fitting the GAMS to a larger set of either modelled or observed data.