Validation and calibration of Source water quality models in the Great Barrier Reef catchments

Abstract

The Paddock to Reef (P2R) Integrated Monitoring, Modelling and Reporting Program uses modelled catchment loads to estimate annual load reductions as a result of improved management practices in the Great Barrier Reef (GBR) catchments. Calibrating and validating catchment models across a large spatial scale is challenging, and requires a range of approaches to continue to refine and improve modelled load estimates. Catchment models have been built for the six Natural Resource Management (NRM) regions draining to the GBR in the Source framework. The NRM regions range in area from 9,000 km2 to 155,000 km2, with the total modelled area 423,000 km2. Load estimates are modelled for 10 water quality constituents. Many improvements have been made over the past seven years to improve confidence in modelled loads. An example includes hydrology recalibration and baseflow optimisation to align with estimates derived from gauged flow. Typically, the larger, drier regions required more gauges to be recalibrated than for the wetter regions. In Cape York (CY) following recalibration of the model, the baseflow proportion was improved from an average of 43% to 33%, which better aligns with the baseflow estimate from gauged flow of 31%. A second notable model update is the improved representation of Dissolved Inorganic Nitrogen (DIN) in sugarcane areas. The DIN runoff model was updated to incorporate a new algorithm which correlates fertiliser input to nitrogen in runoff, derived from the pool of P2R field monitoring data. This daily model of DIN concentration in runoff was based on >200 sugarcane field monitored DIN runoff events. These changes have resulted in modelled DIN loads now being within ± 30% of measured loads in the Wet Tropics (WT), compared to ± 50% in previous model iterations. The improved representation of fine sediment sources as a result of new datasets (for example, gully mapping in CY, Burdekin and Fitzroy) is another example of model calibration using the most up-to-date data. CY is typically a data poor region and early model estimates of the hillslope/gully/streambank contribution to fine sediment loads suggested that up to 90% of the fine sediment load was derived from hillslope sources. Recent field surveys combined with LIDAR data and a desktop gully mapping exercise in the Normanby basin conclude that gullies are the major source of sediment in CY. The updated gully map was incorporated into the Source model, and gullies now account for >70% of the fine sediment load in the Normanby basin. A major component of the P2R program is the catchment loads monitoring program critical for model validation. The number of monitoring sites has expanded from 26 in 2006 to 43 in 2017. This means that 85% of the sediment and nutrient loads exported are now monitored across the GBR. As the pool of water quality data increases both spatially and temporally, there has been a shift to manual model calibration to align with monitoring data as opposed to using the data for validation as was the case in the early years of the program. Whilst the average annual load estimates for the models have not changed significantly there is much greater confidence in the relative contribution of constituents, for example sediment sources and sinks, or DIN loss path ways (surface versus subsurface drainage) from different land uses. This leads to greater confidence in the model outputs, particularly from external stakeholders. At the outset of the P2R modelling program there was limited observed data available to calibrate and/or validate modelled loads. Over the course of the program, knowledge gaps were identified and prioritised, resulting in targeted research being undertaken. This has enabled the refinement of different aspects of the models in each region. When considering a long-term modelling exercise, that is data poor, the model improvement cycle should be flexible enough to for regular updates to algorithms in the models, input data sets, and modelled constituent loads. Model validation must draw on a range of data sources to constrain the models. The P2R modelling has been highly beneficial for identification of dominant sources of constituents, guided water quality monitoring and research, and just as importantly, generated debate about load predictions across the GBR.