Estimating the Effect of Climate Internal Variability and Source of Uncertainty in Climate-Hydrological Projections in a Representative Watershed of Northeastern China

: 7 The decomposition and quantification of uncertainty sources in ensembles of 8 climate-hydrological simulation chains is a key issue in climate impact researches. The mainly 9 objectives of this study partitioning climate internal variability (CIV) and uncertainty sources in 10 the climate-hydrological projections simulation process, the climate simulation process formed by 11 six downscaled GCMs under two emission scenarios called GCMs-ES simulation chain, the 12 hydrological simulation process add one calibrate Soil and Water Assessment Tool (SWAT) model 13 called GCMs-ES-HM simulation chain. The CIV and external forcing of climate projections are 14 investigated in each GCMs-ES simulation chain. The CIV of precipitation and ET are large in 15 rainy season, and the single-to-noise ratio (SNR) are also relatively high in rainy season. 16 Furthermore, the uncertainty decomposed frameworks based on analysis of variance (ANOVA) 17 are established. The CIV and GCMs are the dominate contributors of runoff in rainy season. It 18 worth noting the CIV can propagate from precipitation and ET to runoff projections. In additional, 19 the hydrological model parameters are the third uncertainty contributor of runoff, which embody 20 the hydrological model simulate process play important role in hydrological projections in future. 21 The findings of this study advised that the uncertainty is complex in hydrological, hence, it is 22 meaning and necessary to estimate the uncertainty in climate simulation process, the uncertainty 23 analysis results can provide effectively efforts to reduce uncertainty and then give some positive 24 suggestions to stakeholders for adaption countermeasure under climate change. study also reveal that the internal variability is non-negligible in predicting climate-hydrological projections,


Introduction
The mainly aim of this study is:(1) to analyze the precipitation, temperature, ET and runoff 105 projections changing under climate change.
(2) to estimate the role of internal variability and 106 external forcing on the climate-hydrological projections.
(3) to quantify the source of uncertainty 107 contribution on the overall uncertainty. (4) to confirm the important influence factors and 108 uncertainty source of runoff. The uncertainty decomposition framework of this study shows in 109

Hydrological modeling and parameter uncertainty assessment 152
The SWAT 2012 is used to simulate runoff in this study. SWAT is a physically based water-scale 153 model which is widely used in investigating hydrological processes around the world (Wang et al. parameters have been generated. The initial iteration of LHS derived 1000 simulations, for all 166 initial parameter sets, the best 100 parameter sets were selected by the condition as ENS above 0.9, 167 R 2 above 0.9 and Re below 10.

The internal variability estimate method 181
The internal uncertainty is expected to present the natural viability of the regional climate at Where φi,j is the hydrological variability under the hydrological simulation chain; ηi,j is the residual variance of 203 the climate variability for the given hydrological simulation chain, it can also be express as internal variability.

204
The hydrological variability φi,j of any simulation chain can be defined as Eq. (2): 205 Where μ is the overall mean of hydrological variability under climate change; αh is the effect contributed by 207 hydrological model parameters; βk is the effect contribute by GCMs; γl is the effect contribute emission scenarios; 208 ξh,k,l is the interaction terms of the model. 209 On the base of the above expression of the raw output from simulate chains, the overall 210 variance of the runoff projections as flowing: 211

217
(2) The uncertainty quantified and decomposition 218 This manuscript constructs a three-way ANOVA framework to decomposition the different 219 uncertainties contribution, this technology has ability to partition the total observed variance into 220 different sources, and then quantify the contribution of different sources to total variance (Wang et In the ANOVA model, the total variance of the hydrological variable is expressed as 237 the total sum of squares (SST), and it can decompose into individual variance of each effect: 238 Where SSA, SSB, SSC is the uncertainty contribution of SWAT model parameters, GCMs, emission scenarios 241 respectively, SSIV is the internal variability and SSI is the contribution of the interaction effects between SWAT 242 model parameters, GCMs and emission scenarios.       303 The box chart of Fig.5a and Fig.5b shows the maximum and minimum temperature (Tmax and RCP4.5 scenarios. While the summer runoff projections showed increases from 7.93% to 85.76% 377 and decreases from -11.6% to -29.15%, the decrease trend is smaller than increase trend, thus, a 378 slight increase trend with the mean increase value as 11.95% can be found in 2080s under RCP8.5. 379

The temperature projections change under climate change
In addition, the runoff projections shown a slight increases trend in autumn and winter both 380 under RCP4.5 and RCP8.5 scenarios, and also shown a small various among different GCMs. respectively.

396
Furthermore, the box-and-whisker plots show in Fig.7a and Fig.7b, the upper and lower ends 397 represent the highest and lowest runoff, and the change range indicated the uncertainty bound.

Impacts of climate factors to runoff change 416
After analyzing the changes of precipitation, Tmax, Tmin, ET and runoff in future, it can be found 417 that the different climate factors may produce different contribution to runoff changing. Hence, it 418 is important to analyze the relationship between the change of runoff and change of climate factors. 419 In order to determine the relationships between them, the multiple linear regression was performed 420 for each model chain using changes of precipitation, Tmax, Tmin and ET as the independent 421 variables and the runoff as the dependent variables. 422 The regression coefficients for runoff are shown in Table 3 In general, the increase of 423 precipitation may cause a positive effect on runoff increasing, this trend can be found in all of the 424 models and scenarios and coefficients at the 0.001 significant level. In contrast, the increase of ET 425 projections was negatively related to runoff, and there are seven projections at the 0.001 426 significant level. In addition, the increase Tmax and Tmin may contribute the increase trend of runoff, 427 however, the coefficients did not pass the significant test even at 0.05 level. Above all, the 428 precipitation and ET has a larger influence in runoff projection in most model chain.   The SNR is defined as the absolute value of ensemble mean divided by the CIV, which can 468 measure the relative contribution of external forcing and internal variability. The SNR values of 469 precipitation, temperature, ET and runoff are showed in Fig.9. This metrics convey useful 470 information about the magnitudes of the forced and internally generated components of climate 471 projections under future climate change. It can be seen from the Fig.9 that the SNR values of 472 precipitation and runoff are relatively smaller than the other climate projections. 473

Evaluation of the uncertainty influence factors of runoff
The SNR values of Tmax and Tmin demonstrate a relatively higher values in May to October, it 474 worth noting that the temporal pattern of the SNR is mainly determined by the internal variability 475 pattern in November to March and by a mainly combination of forced response in April and 476 October. The SNR of ET is higher in June to October than the other month in 2050s period, and it 477 is relatively stable in 2080s period. Hence, the external forcing is the mainly components of ET 478 projections changing. In addition, the SNR of runoff is relatively small which like precipitation. 479 An important result is that the external forcing contributed a considerable higher component in 480 temperature and ET changing than precipitation and runoff, and the SNR exhibits higher values in 481 June to September than the other models in both two emission scenarios and future periods. 482 483 Fig.9 The SNR values of climate-hydrological projections 484 The contribution of uncertainty sources showed in Fig.10. It is noteworthy that the effect of 497 internal variability is non-negligible, which is exceeded the contribution due to the GCMs. It 498 contributes 29%-48% and 31.4% -47.4% of the total variance in 2050s and 2080s, respectively. 499

Contribution analysis of uncertainty sources
The biggest contribution embodies in September in two future periods, which is late flooding 500 season in watershed. The second significant uncertainty contributor is GCMs, which account for 501 21% -41% and 15% -33% in 2050s and 2080s, and the biggest uncertainty is in September (2050s) 502 and August (2080s) respectively. For the SWAT model parameter sets, the contribution accounts 503 for 4%-39% and 4.8% -32.4% in 2050s and 2080s, respectively. Compared with the previous two 504 uncertainty sources, the SWAT model parameters main effect the Spring (March to May) and 505 Winter (December to February) runoff projections. The interaction term contribution to the runoff 506 projection explaining approximately 8% -11% and 7% -12% throughout the 2050s and 2080s 507 periods, respectively. The contribution of emission scenarios is relatively small, which bellows 5% 508 and 10.5% in 2050s and 2080s, respectively. 509 Overall, the results of uncertainty decomposition in Fig. 10

Uncertainty assessment 549
The ANOVA framework was constructed to quantify the uncertainty sources contribute to the 550 overall uncertainty, furthermore, in considering the substantial effects of internal variability on the 551 uncertainty of runoff projections, the uncertainty contribution of internal variability has been 552 considered to ensure the comprehensive of uncertainty assessment. (1) Based this study analysis of future climate conditions for the Biliu River basin, it can be 596 found that an increase in seasonal mean temperature for both emission scenarios, with greatest 597 increase in summer and autumn. In term of precipitation, it indicates an increased trend in summer, 598 autumn and winter and a relatively larger uncertainty in summer and winter. Results based on the 599 SWAT modeling indicated that the ET shows a slight increase in summer and winter, and the 600 runoff projections trend a diversity changing trend in future, especially in summer and autumn. As the rain season in the study basin, some water resources adaptation measures need be planned 612 to alleviate the climate change influence, especially in high emission scenarios (RCP8.5) and far 613 future (2080s). 614 (3) This study found GCMs, internal variability and SWAT model parameters are the mainly 615 uncertainty contributor of runoff. In addition, the SWAT model parameters uncertainty 616 significantly effects runoff projections in spring and winter, thus the calibration of sown melt 617 parameters needs more attention. The influence of external forcing is smaller in GCMs-EM-HMs 618 than GCMs-EM, because the uncertainty sources increased and the hydrological simulation 619 process bring more uncertainty to runoff. 620 The findings of this study indicate that the uncertainty of climate-hydrological system is 621 noticeable in future, these kinds of uncertainties may extremely influence the stakeholders and 622 local water resources government to provide correct hydrological regulation and flood control 623 measures. This study also reveal that the internal variability is non-negligible in predicting 624 climate-hydrological projections, which is worth more research in future. 625

Funding Statement 627
This study was sponsored by the Natural Science Foundation of Shanxi Province, China. Grant