Analysis and prediction of stream water quality in the Great Barrier Reef catchments using Bayesian hierarchical model

  • The Great Barrier Reef (GBR) is the world's largest coral ecosystem and it has been experiencing significant water quality deterioration due in part to agricultural intensification and urban settlement in adjacent catchments. The land-derived pollutants are responsible for the degradation of instream water quality in the GBR catchment. The spatial and temporal variations in water quality hinder the interpretation of the water quality monitoring data. The water quality monitoring program provides a potential opportunity to develop a data-driven understanding of water quality at catchment scale; including both natural and anthropogenic influences on water quality. In this project, we investigated water quality monitoring records from 32 site across the GBR catchments (Figure 1). We adopted Bayesian hierarchical modelling: decomposes the complex interactions in the observed data into a series of conditional models. We incorporated the spatial variability in stream water quality within a two-level modelling structure to capture the variability between sites (Figure 2). The modelling results (Figure 3) indicated large positive deviation: Burdekin and Fitzroy sites, especially 3 upland sites in Fitzroy.  The inference of the model parameters (Figure 4) showed that: 1) land use - grazing and dry land agriculture have significant impacts on source of TSS; 2)   geology: soil erodibility - impact on pollutant mobilisation (erosion effect); 3) topography: slope - impact on pollutant delivery.
  • Shuci Liu (PhD scholar), Prof. Andrew W. Western (Co-supervisor), Dr. Anna Lintern (Co-supervisor, Monash), Dr. Angus Webb (Co-supervisor), Dr. David Waters (Collaborator)

Hydrological model calibration in the ungauged basins using ground-based and satellite observations of river stages

  • Jie Jian (PhD Scholar), Dr. Justin Costelloe (Co-supervisor), Prof. QJ Wang (Co-supervisor)


  • Hydrological modelling is used as a tool to understand and quantify hydrological processes and is applied in predictions and decision-making processes. Model parameters that cannot be measured directly should be calibrated to make models accurate. Thus, the observed discharge data that can be used to constrain the model parameters are essential. However, the majority of the streams in the world are ungauged or sparsely gauged. In these catchments, the uncertainty issue is quite obvious and significant due to the lack of continuous discharge data. Thus new calibration methods that could use water level data instead are crucial in streamflow prediction in ungauged and poorly gauged areas. 

The potential to use water level data

  • There is an obvious positively monotonic relationship between discharge (Q) and water level (h) in most natural rivers;
  • Water level data is available in a significant number of catchments that lack rated discharge;
  • The availability of altimetry data is likely to provide h to many ungauged catchments.


  • Two calibration schemes, which are not reliant on extensive observed discharge data, are examined: 
  • Scheme 1: Spearman Rank correlation based scheme (SRC);
  • Scheme 2: Inverse Rating Curve based scheme (IRC).
  • A small number of discharge measurements (50th, 75th and 95th percentiles of Q_obs) and some regionalised runoff ratio values are used to constrain the model parameters.

Results and discussions

  • SRC reproduced the dynamics of flow events but contained large biases, because it did not contain the information on the dynamic range of the discharge observations (Figure 1b);
  • IRC performed better because it provided stronger relationship between Q and h (Figure 1c);
  • SRC/IRC with a small number of high flow Q_obs could effectively improve the calibration performances (Figure 2).

Ongoing works

  • The study is extended to 200+ Hydrological Reference Stations in Australia;
  • More hydrological models are tested, such as SIMHYD, GR4J, AWBM, IHACRES and SACSMA;
  • The influences of different hydro-climatic and catchment variables are examined;
  • Instead of high flow data, other effective factors are explored to constrain the model performances.
 Figure 1. Hydrographs of the control case, SRC case and IRC case

Figure 1. Hydrographs of the control case, SRC case and IRC case

 Figure 2. Difference between Q_est and Q_obs in FDCs

Figure 2. Difference between Q_est and Q_obs in FDCs