Performance evaluation of large-scale hydrological models for different hydro-climatic zones in Australia
Summary
Climate change will affect volumes and timings of river discharges and will therefore determine the distribution and amount of people experiencing floods and droughts in the future. Therefore, the need for reliable hydrological modeling increases in order to respond to these changes. Modeled output of these models have been compared with observations and it is seen that modeled output is not limited to input forcing only, which results in uncertainty of modeled output. This research aims to evaluate the performance of two large-scale land surface models (LSM) and two large-scale global hydrological models (GHM) for actual evapotranspiration, soil moisture and runoff across different climate zones of Australia. This contribution offers an insight into the spatial performance distributions for different large-scale hydrological models for those hydrological variables. Also this research evaluates the multi-model ensemble median for performance. Results obtained are compared with performances for the calibrated national model, AWRA-L, which serves as benchmark in this research. For the evaluation of the large scale models, the Tier-1 of the EU-funded EartH2Observe datasets have been used. In this research. the Kling-Gupta efficiency is selected as model performance metric and objective function for calibration of PCR-GLOBWB.
For actual evapotranspiration, global hydrological models perform on average better in arid/semi-arid climate (KGEGHMs= 0.34; KGELSMs = -0.412), whereas land surface models simulate actual evapotranspiration on average better in tropical climates (KGEGHMs = 0.47; KGELSMs = 0.74). For the other evaluated variables, no such spatial performance differentiation can be made between land surface and water balance models. Still, all models show different spatial patterns of good and bad performances, but those are not related to whether the model is a land surface model or a global hydrological model. The ensemble median does not necessarily lead to improved results for all climate zones for actual evapotrtanspiration and soil moisture. For the ensemble median of streamflow, improvements have been found for some climate zones.
The effects of calibrating PCR-GLOBWB using monthly streamflow observations from different climate zones has been evaluated in this study. The research found major improvements after validating the calibrated PCR-GLOBWB model compared with the reference scenario for climate zones used for calibration. However, performances for the other climate zones improved as well. Taking all climate zones into account, the PCR-GLOBWB run with the default parameter setting obtains for 20% of its simulations a KGE > 0.2. Calibration of the PCR-GLOBWB model increased this percentage with 15%. Also, calibration of PCR-GLOBWB leads to an increase of 10% of simulations with KGE > 0.5. Also, this research demonstrated that performance improvements differ in magnitude between climate zones (Calibrated PCR-GLOBWB: KGEtropics<0.2 = 55%, Reference PCR-GLOBWB: KGEtropics<0.2 = 90% ; Calibrated PCR-GLOBWB: KGEarid <0.2 = 80-85%, Reference PCR-GLOBWB: KGEarid<0.2 = 90% ; Calibrated PCR-GLOBWB: KGEtemperate< 0.2 = 65%, Reference PCR-GLOBWB: KGEtemperate<0.2 = 75%). Differences in magnitude of improvements for certain climate zones are related to in which climate zone the catchments used for calibration are situated.
Still, after calibration, performances changed from really bad KGE to bad KGE for some regions. This is mainly attributable to bad model structure or poor forcing dataset rather than the use of a sub-optimal combination of parameters in the PCR-GLOBWB model. In addition this research demonstrates that the calibration of PCR-GLOBWB using streamflow data negatively influences the performance of actual evapotranspiration of the calibrated PCR-GLOBWB model. Therefore, before calibrating a hydrological model and applying it to a certain region, the purpose of the model application needs to be fully known.
Apart from being able to allow for improved insights in the full extent of availability and the distribution of water resources, this research proved that global scale hydrological model could be a valuable source of knowledge for developing countries without a fine resolution hydrological model. However, further research needs to be carried out for both stepwise calibration to enhance applicability and improved characterization of rainfall amounts at 0.5° or preferably at smaller scales in order to increase reliability of large-scale hydrological models.