Bias Nonstationarity Of Climate Model Outputs And Its Propagation And Influence In Runoff Impacts

Global warming changes the atmospheric circulation process at different spatial and temporal scales,which in turn changes the global and regional water cycle process including runoff in the basin.Studying the impact of climate change on watershed runoff has become an important topic of wide concern in the international community.The bias correction method of climate model outputs is widely used as an important link between climate model output and hydrological model input,and is widely used to evaluate the impact of climate change on watershed runoff.The traditional bias correction method is generally based on the assumption that the climate model outputs(such as precipitation and temperature)have stationary biases,that is,the biases of climate model outputs in the historical and future time periods are the same.In the bias correction method,the same amount of bias estimated from the historical period is removed from the climate model outputs of the future period.However,with the deepening of climate change research,the assumption of bias stationarity has been widely questioned.Therefore,this study selects the Weihe,Hanjiang,Xiangjiang and Dongjiang river basins as study areas,to investigate the bias nonstationarity of climate model outputs and its attributions,that is,internal climate variability and climate model sensitivity.The study established an evaluation method used internal climate variability as a baseline to evaluate the performance of tradition bias correction method,and a method of bias correction considering bias nonstationarity is also proposed.Based on this,the propagation of bias nonstationarity in runoff response is further studied.The performance of the bias correction method considering bias nonstationarity in runoff response is evaluated.The main contents and conclusions of this paper are as follows:(1)The bias nonstationarity of global climate model(GCM)simulated precipitation,maximum and minimum temperature is estimated based on historical observations and GCM simulations.The results show that GCMs are considerably biased in simulating precipitation,maximum and minimum temperature.Their biases are nonstationary in historical and future periods.The changes in precipitation bias are similar in both historical and future periods relative to the baseline period.However,changes of bias in maximum and minimum temperature over future period are larger than twice those over historical period.(2)The roles of internal climate variability(referring to unpredictable,internally generated climate variability)and climate model sensitivity(referring to the differences in a climate models' response to the same external forcing)in bias nonstationarity of GCM simulated precipitation and temperature are investigated,and their impacts on the performance of a traditional bias correction method are further evaluated.The results show that both internal climate variability and climate model sensitivity can lead to bias nonstationarity,and further affect the performance of bias correction.In general,the performance of bias correction in the validation period is worse than that in the calibration period.For precipitation,bias nonstationarity attributed by internal climate variability makes the bias of bias-corrected precipitation worse than that of raw GCM-simulated precipitation over Hanjiang and Xiangjiang watersheds.Climate model sensitivity also has the similar impacts.For maximum and minimum temperature,internal climate variability has a little impact on the performance of bias correction.However,under the impacts of climate model sensitivity,the bias correction cannot reduce the GCM-simulated temperature bias effectively.(3)To take into account the uncertainty linked to internal climate variability,the performance of a traditional bias correction method is evaluated using internal climate variability as a baseline.If the remaining bias between bias-corrected simulations and observations is outside the range of internal climate variability,it is considered that the bias correction method has achieved the expected effect.On the contrary,it is considered that the bias correction cannot eliminate the bias of the climate model outputs well.Furthermore,the impacts of uncertainty of internal climate sensitivity estimated by 4 GCMs multi-member ensembles are explored in evaluating the performance of bias correction.The results show that the traditional bias correction method can reduce the bias of raw GCM simulations in the historical period,indicated by the remaining bias is within internal climate variability.However,in the future period,the traditional bias correction method can only reduce the bias of raw GCM simulations to some extent.The remaining bias can be outside the range of internal climate variability,even though the uncertainty of internal climate variability is large.(4)A bias correction method considering the nonstationarity of bias caused by climate model sensitivity is proposed.The method predicts the bias of future time periods by establishing a functional relationship(both first-order linear function and second-order nonlinear function)between GCM simulated precipitation and its bias in the historical period in the four basins.The results show that both the first-order linear equation and the secondorder nonlinear equation can fit the relationship between the GCMs simulated monthly precipitation and its bias well,even though the latter is slightly better simulated than the former.In the future period,the nonstationarity-based bias correction method effectively weakens the influence of bias nonstationarity on the precipitation change signal,and also effectively reduces the uncertainty of GCM prediction of precipitation.(5)Firstly,the propagation of bias nonstationarity of GCM-simulated precipitation and temperature in hydrological process is investigated by the simulated runoff using the precipitation and temperature as inputs of the GR4J-6 hydrological model.The results show that the bias of simulated runoff is also nonstationary through the hydrological process.For the runoff simulated by the bias-corrected precipitation and temperature,the impacts of bias nonstationarity attributed by internal climate variability are small,while the bias nonstationarity attributed by climate model sensitivity can increase the remaining bias of simulated runoff.Secondly,based on the hydrological variability attributed by the internal climate variability,the performance of a traditional bias correction method in assessing hydrological impacts is evaluated in the historical and future periods.The results show that the remaining bias of the runoff simulated using the bias-corrected precipitation and temperature is within the range of hydrological variability in the calibration period.In the historical validation period,the traditional bias correction method can reduce the bias of the runoff simulated using GCM outputs effectively.In the future validation period,the remaining bias of the runoff is still significant,even though the traditional bias correction method can reduce the runoff bias to some extent.Finally,the corrected precipitation using the developed bias correction method is used to drive the hydrological model for assessing the impacts of climate change on watershed hydrology and its uncertainty.The results show that runoff simulated using the developed bias correction method will decrease in the future.In addition,the developed bias correction method can reduce the uncertainty of the runoff simulated using GCM outputs.