Hydro-biogeochemical models are used to foresee the impact of mitigationmeasures on water quality. Usually, scenario-based studies rely on singlemodel applications. This is done in spite of the widely acknowledgedadvantage of ensemble approaches to cope with structural model uncertaintyissues. As an attempt to demonstrate the reliability of such multi-modelefforts in the hydro-biogeochemical context, this methodologicalcontribution proposes an adaptation of the reliability ensemble averaging(REA) philosophy to nitrogen losses predictions. A total of 4 models areused to predict the total nitrogen (TN) losses from the well-monitored EllenBrook catchment in Western Australia. Simulations include re-predictions ofcurrent conditions and a set of straightforward management changes targetingfertilisation scenarios. Results show that, in spite of good calibrationmetrics, one of the models provides a very different response to managementchanges. This behaviour leads the simple average of the ensemble members toalso predict reductions in TN export that are not in agreement with theother models. However, considering the convergence of model predictions inthe more sophisticated REA approach assigns more weight to previously lesswell-calibrated models that are more in agreement with each other. Thismethod also avoids having to disqualify any of the ensemble members.
展开▼
