PEST hydrology calibration process for Source Catchments - Applied to the Great Barrier Reef, Queensland

Abstract

The rapid development in computer technology over the past decade has made automated calibration of models common practice. PEST, a model-independent parameter estimation tool, is one such tool used extensively in groundwater model calibration. PEST was linked to the eWater CRC Source Catchments water quantity and quality model to calibrate the parameter values for component SIMHYD rainfall runoff and Laurenson non-linear flow routing models. The calibration process was applied across 450,000 km 2 in the Great Barrier Reef (GBR) Queensland. Six models were built, with catchments ranging in area from 10,000 - 156,000 km 2. A consistent approach to model calibration was applied across all models. Unique SIMHYD parameter sets were generated for the three dominant Hydrologic Response Units (HRU): forest, grazing and agriculture. A three-part objective function was used, incorporating daily and monthly flows, and exceedence time at 9 - 92 gauging locations throughout each project; all three objective function components were given equal weighting. The contributions of each gauge to the initial value of the objective function were equalised to remove bias caused through magnitude of flow volumes. A regionalisation approach, based on contributing areas to each observation gauge, was used. The purpose of regionalisation was to rationalise the number of parameter sets that PEST needed to derive during the simultaneous calibration process, and enabled only those identifiable parameters, in regions that were underperforming, to be adjusted. For example, if a HRU did not occur in a region, or comprised a very small area of that region, it could be 'turned off', so that those parameters most likely to impact calibration could be adjusted more quickly by PEST. In projects representing very large areas, a supervised singular value decomposition approach was taken, whereby PEST determined a defined set of parameter combinations most significantly influencing calibration, and these parameter combinations were adjusted during the calibration process. In all six GBR models, calibration was deemed sufficient when modelled volume was calculated within 20% of measured volume, the daily Nash-Sutcliffe was greater than 0.5, and the monthly Nash-Sutcliffe greater than 0.8, at each of the utilised gauging locations. The results of the hydrology calibration process ensure that the six GBR models are set up with a platform that is considered suitable to estimate runoff rates from the identified HRUs, from which to simulate runoff and erosion processes, and ultimately to estimate loads to the GBR. The design of this hydrology calibration was intended to provide stream network flow and storage characteristics reasonable enough to route constituent loads in a reasonable manner, but not to provide water quantity estimates of an accuracy deemed necessary for water supply and regulation needs. Using PEST to parameterise the SIMHYD Rainfall Runoff and Laurenson non-linear flow routing models in Source Catchments was an efficient and repeatable methodology for simultaneous calibration across large catchments. Improvements to the process outlined above would include reconsidering the most appropriate objective functions, and using local and 'expert' knowledge to guide the parameter starting values and ranges.