Study Of Parameter Temporal Transferability Of Hydrological Models Under Changing Environments

Under the changing environments,the runoff mechanisms and rainfall-runoff relationship of many watersheds have witnessed great changes due to the effects of both climate change and human activities.As a result,the parameters of hydrological models may show a decreased temporal transferability between periods with distinct climate and watershed underlying conditions.Given numerous researches have attempted to improve the simulation performance of hydrological models during single period,few of them focus on improving the temporal transferability of model parameters under the changing environments.Thus,this study aims to explore the possibility of improving the temporal transferability of model parameters under the changing environments by applying parameter calibration schemes considering multiple hydrological variables and accounting for time-varying parameters.The lumped hydrological model GR4 J and the distributed hydrological model DDRM were chosen and the investigation was carried out in three watersheds(Weihe Basin,Hengjiang Basin and Tuojiang Basin)in western China.The main results and findings of this study are summarized as follows:(1)The main factors that results in the nonstationary rainfall-runoff relationship in each watershed were identified based on the Mann-Kendall trend analysis on the hydrometeorological data series.A sliding window of 5-year in length was applied to split the30-year historical data and 26 sub-periods were therefore defined.The GR4 J model and DDRM model were calibrated during each sub-period,and the corresponding simulation performance of streamflow were assessed.The parameter sets obtained from each sub-period were then validated during other sub-periods to evaluate their temporal transferability.The results show that both GR4 J model and DDRM model can achieve a satisfactory streamflow simulation performance during each calibration sub-period.Besides,the posterior distributions of model parameter for the GR4 J and DDRM model show distinct difference when they were calibrated during a wetter period and a drier period,respectively.As a result,a weak temporal transferability can be witnessed when model parameter set was transferred between sub-periods with different hydrometeorological conditions for both the GR4 J and DDRM model.In contrast,when parameter set was transferred between sub-period with close hydro-meteorological conditions,a satisfactory temporal transferability can be achieved.(2)A joint calibration schemes and a stepwise calibration schemes based on multiple hydrological variables were applied to calibrate model parameters for both GR4 J model and DDRM model during the 26 sub-periods.The corresponding simulation performance in terms of soil moisture,actual evapotranspiration and streamflow were compared with the case where the traditional calibration schemes considering only streamflow was applied.Besides,the difference in temporal transferability of model parameters under three cases were also assessed.The results show that the joint-and stepwise-calibration schemes can reduce parameter uncertainty and achieve a better simulation in terms of soil moisture and actual evapotranspiration,but a slightly worse simulation of streamflow when compared to the case of the traditional calibration schemes.In addition,the joint-and stepwise-calibration schemes proved their ability to improve temporal transferability of model parameters in most cases,especially between sub-periods with different hydro-meteorological conditions,using the case of traditional calibration schemes as reference.(3)The specific parameter within GR4 J model was treated as time-varying and as a function of multiple covariates representing hydro-meteorological conditions and watershed underlying conditions.The GR4 J model with time-varying parameter is denoted as GR4J-T model.Both the GR4 J model and GR4J-T model were calibrated during 26 sub-periods.Then their simulation performance of streamflow and the corresponding temporal transferability of model parameters were assessed.The results show that the GR4J-T model outperformed the GR4 J model in terms of streamflow simulation in most cases for all three watersheds.Besides,the GR4J-T model shows a better temporal transferability of model parameters in most calibration-validation cases when compared to the GR4 J model,especially between sub-periods with different hydro-meteorological conditions.(4)A comprehensive comparison of the impact of applying parameter calibration schemes considering multiple hydrological variables and time-varying parameters for GR4 J model on the temporal transferability of model parameters was applied.The results show that applying time-varying parameter can lead to a better temporal transferability of model parameters for GR4 J model.