Diagnosing the impacts of permafrost on catchment hydrology: field 1 measurements and model experiments in a mountainous catchment in 2 western China 3

: 18 Increased attention directed at permafrost hydrology has been prompted by climate 19 change. In spite of an increasing number of field measurements and modeling studies, the 20 impacts of permafrost on hydrological processes at the catchment scale are still unclear. 21 Permafrost hydrology models at the catchment scale were mostly developed based on a 22 “bottom - up” approach, hence by aggregating prior knowledge at the spot/field scales. In 23 this study, we followed a “top - down” approach to learn from field measurement data to 24 understand permafrost hydrology at the catchment scale. In particular, we used a stepwise 25 model development approach to examine the impact of permafrost on streamflow response 26 in the Hulu catchment in western China. We started from a simple lumped model (FLEX-L), 27 and step-wisely included additional complexity by accounting for topography (i.e. FLEX-D) 28 and landscape heterogeneity (i.e. FLEX-Topo). The final FLEX-Topo model, was then 29 analyzed using a dynamic identifiability analysis (DYNIA) to investigate parameters ’ 30 temporal variation. By enabling temporal dynamics on several parameters, we diagnosed 31 the physical relationships between parameter variation and permafrost impacts. We found 32 that in the Hulu catchment: 1) the improvement associated to the model modifications suggest that topography and landscape heterogeneity are dominant controls on catchment 34 response; 2) baseflow recession in permafrost regions is the result of a linear reservoir, and 35 slower than non-permafrost regions; 3) parameters variation infers seasonally non- 36 stationary precipitation-runoff relationships in permafrost catchment; 4) permafrost impacts 37 on streamflow response mostly at the beginning of the melting season; 5) allowing the 38 temporal variations of frozen soil related parameters, i.e. the unsaturated storage capacity 39 and the splitter of fast and slow streamflow, improved model performance. Our findings 40 provide new insights on the impact of permafrost on catchment hydrology in vast mountain 41 regions of western China. More generally, they help to understand the effect of climate 42 change on permafrost hydrology. 43 and reduces 531 the moisture deficit. The deep groundwater, increasing soil moisture and temperature, and 532 deepening frost boundary, form strong evidence to support this concept. Gradually, with 533 the increase of the thawing process, soil becomes wet from top soil to deeper soil, soil 534 moisture and temperature increases from top to bottom, the frost boundary deepens 535 downward. Groundwater level variation observed by 4 wells in WW01 and 2 wells in WW03, 536 show a similar phenomenon. These observations indicate the increasing of the active layer 537 depth, and the increase of soil moisture in the beginning of the melting season. In the 538 middle of the melting season, the active layer is thoroughly thawed, and the unsaturated 539 soil layer becomes saturated, and the value of S umax_V and S umax_D is reduced, resulting in larger 540 runoff generation. based to simulate the impact of the thawing on the variation unsaturated umax_V and umax_D ), impact of processes on hydrological connectivity between and

Increased attention directed at permafrost hydrology has been prompted by climate 19 change. In spite of an increasing number of field measurements and modeling studies, the 20 impacts of permafrost on hydrological processes at the catchment scale are still unclear. 21 Permafrost hydrology models at the catchment scale were mostly developed based on a 22 "bottom-up" approach, hence by aggregating prior knowledge at the spot/field scales. In 23 this study, we followed a "top-down" approach to learn from field measurement data to 24 understand permafrost hydrology at the catchment scale. In particular, we used a stepwise 25 model development approach to examine the impact of permafrost on streamflow response 26 in the Hulu catchment in western China. We started from a simple lumped model (FLEX-L), 27 and step-wisely included additional complexity by accounting for topography (i.e. FLEX-D) 28 and landscape heterogeneity (i.e. FLEX-Topo). The final FLEX-Topo model, was then 29 analyzed using a dynamic identifiability analysis (DYNIA) to investigate parameters' 30 temporal variation. By enabling temporal dynamics on several parameters, we diagnosed 31 the physical relationships between parameter variation and permafrost impacts. We found 32 that in the Hulu catchment: 1) the improvement associated to the model modifications 33 Permafrost is the ground that is at or below 0°C for at least two consecutive years. 46 Permafrost covers 24% of the exposed land surface of the Northern Hemisphere (Zhang et  47 al., 2005; Woo, 2012; Walwoord and . The high Asia region is largely covered 48 by permafrost and is characterized by a fragile cold and arid ecosystem (Immerzeel et al.,49 2010; Ding et al., 2020). As this region serves as the "water tower" for nearly 1.4 billion 50 people, understanding the permafrost hydrology is important for regional and downstream 51 water resources management and ecosystem conservation. Permafrost prevents vertical 52 water flow which often leads to saturated soil conditions in continuous permafrost, while 53 confining subsurface flow through perennially unfrozen zones in discontinuous permafrost 54 (Walvoord and Kurylyk, 2016). As an aquiclude layer, permafrost substantially controls 55 surface runoff and its hydraulic connection with groundwater. The freeze-thaw cycle in the 56 active layer significantly impacts soil water movement direction, velocity, storage capacity, 57 and hydraulic conductivity (Bui et al., 2020;Gao et al., 2021). 58 Permafrost hydrology attracts increasing attention, as the cold regions, e.g. Arctic and high 59 mountain Asia, are undergoing rapid changes (Tananaev et al., 2020). Permafrost 60 degradation and its impact on hydrology is one of the research frontiers (Zhao et al., 2020;61 Ding et al., 2020). The question "How will cold region runoff and groundwater change in a 62 warmer climate (e.g. permafrost thaw)?" was identified by the International Association of 63 Hydrological Sciences (IAHS), as one of the 23 major unsolved scientific problems (Blöschl et  64 al., 2019), which requires stronger harmonisation of community efforts. Permafrost thawing 65 also poses great threats to the release of frozen carbon in both high altitude and latitude 66 regions, which is likely to create substantial impacts on the climate system (Wang et al., 67 2020). Attention is also growing on the impacts of permafrost hydrology on nutrient 68 transport and organic matter, and permafrost-climate feedback (Tananaev et al., 2020 Arctic river basins, and found that most models project a long-term drying of surface soil, 110 but the projection vary strongly in magnitude and spatial pattern. Except for hydrological 111 models, many land surface models explicitly consider the freeze-thaw process, in order to 112 improve land surface water and energy budget estimation and weather forecasting accuracy Although there are many permafrost hydrological models, most models have strong prior 120 assumptions on permafrost hydrological behavior and therefore on its impact on catchment 121 hydrology (Walvoord and Kurylyk, 2016;Gao et al., 2021). Such models follow a "bottom-122 up" modeling approach, which presents an "upward" or "reductionist" philosophy, based on 123 the aggregation of small-scale processes and a priori perceptions (Jarvis, 1993 In this study, we utilized a top-down approach (Sivapalan et al., 2003), to understand the 179 impacts of permafrost on catchment hydrology. We used a series of hydrological models as 180 tools to diagnose which components play dominant roles controlling permafrost hydrology. 181 We asked the following scientific questions:

258
For vegetation hillslope, we constrained a larger prior range for the unsaturated storage 259 capacity (Sumax_V) (10~200mm), which means more water is needed to fill in its storage 260 capacity to meet its water deficit, which is evidenced by previous studies in this region (Gao 261 et al., 2014). For alpine desert, due to its sparse vegetation cover, we constrained a 262 shallower unsaturated storage capacity (Sumax_D) (1~150mm). For riparian area, due to its 263 location which is prone to be saturated, we also constrained a shallower unsaturated 264 storage capacity (Sumax_R) (1~150mm). 265 although not measurable quantities, are associated to specific process representations 295 (Fenicia et al., 2009). Hence the variability of parameters can represent the catchment 296

Model uncertainty analysis and evaluation functions
properties change over time. In the case of this study, we focus our attention to permafrost 297 related parameters and associated processes. 298 The chosen window size allows for tailoring across influence scales of parameters. In this 299 study, a moving window with non-overlapping 10-days length was applied (Osuch, 2019). 300 The same threshold as with the application of the GLUE approach was applied. Hence, we 301 selected the top 1% parameter as behavioral parameter sets for each time window. ( Figure 4). In Exp1, merely from analysis of the observed data, we found that the total runoff 369 is 499mm/a, which is even larger than observed precipitation 433mm/a. This means that the 370 runoff coefficient is larger than 1, and the water balance cannot be closed with the current 371 setup. This explains the relatively low value of the performance indicators.   (Figure 7) processes, but mostly in the beginning of melting season. 426 In order to test the key parameters on model performance, we allow three key parameters 427 permafrost impacts. We argue that clarifying which processes are more important than 441 others, and the selection of involving certain processes, and neglecting certain processes 442 may be more important than accurately solving differential equations ). the FLEX-Topo model was improved, likely due to the inclusion of the following processes: 455 in the beginning of the melting season, snow starts to melt in low elevations, which has 456 good vegetation cover and large rootzone storage capacity. The melt water firstly fills in the 457 water deficit in the large rootzone storage capacity on the vegetated hillslope, without 458 much runoff generation. Hence, considering landscape heterogeneity reduced model 459 discrepancy. However, the overestimation of runoff in the beginning of the melting season 460 still exists, which we hypothesize to be impacted by permafrost and will be discussed in 461 Section 6.3. 462 From stepwise model comparison, we learned that involving topography and landscape is 463 important to reproduce streamflow. Moreover, we found that even without the explicit

469
Only from the shape of the curves in Figure 6 and 7, we did not find significant differences

497
We found that considering the temporal variation of these parameters, model performance 498 was improved, especially for the beginning of melting season in all four years simulation. 499 We also plot parameters dynamics with multi physical variables in Figure 10, including 500 groundwater depth, soil moisture of different layers, soil temperature in different layers, 501 frozen depth and hydrograph simulation, highlighting the beginning of the melting season. 502 infiltrated water in these two periods is mostly stored in supra-permafrost soil, due to the 507 frozen bottom of the active layer. In the middle of the melting season, D is large, indicating 508 the connectivity between surface and groundwater systems, and that hydrological response 509 from rainfall/snowmelt to runoff was fast. These parameter behaviors match well with the 510 deep groundwater level measurements in 4 wells in WW01, and 2 wells in WW03, the 511 gradual increasing of soil moisture from 20cm depth to 300cm depth, the increase of soil 512 temperature from 0cm to 160cm, and the thawing of top frozen soil. All these phenomena 513 occur simultaneously with very limited river runoff generation (red dash box), likely due to 514 the initial dry soil and the increasing unsaturated soil storage capacity with the increase of 515 soil temperature and deepening of unfrozen top soil. The existence of supra-permafrost 516 water and the impermeability of bottom active layer, resulted in the vertical disconnection 517 between surface and groundwater. And different landscapes caused the lateral 518 disconnection between hillslope and river channel. With the increase of soil moisture and 519 thawing of the active layer, the vertical and lateral connections resulted in the increase of 520 runoff generation in the middle of the melting season. After considering parameter 521 dynamics, Exp10 improved model performance, especially in the beginning of the melting 522 season (Figure 8, 9, 10). 523 These results motivated the following perceptual model. In winter, although top soil was 524 frozen, groundwater recession did not stop, which increased soil moisture deficit in the 525 beginning of the melting season. Due to thin snow cover in the Hulu catchment, soil 526 evaporation in winter also continued. After a long winter groundwater recession and soil 527 evaporation, in the beginning of the melting season, soil became dry and deficit of 528 moisture, and the groundwater level was deep. Thus, the unsaturated reservoir storage 529 capacity (Sumax_V and Sumax_D) was large. That is why the precipitation during this period does 530 not generate much runoff, since this amount of precipitation firstly infiltrates and reduces 531 the moisture deficit. The deep groundwater, increasing soil moisture and temperature, and 532 deepening frost boundary, form strong evidence to support this concept. Gradually, with 533 the increase of the thawing process, soil becomes wet from top soil to deeper soil, soil 534 moisture and temperature increases from top to bottom, the frost boundary deepens 535 downward. Groundwater level variation observed by 4 wells in WW01 and 2 wells in WW03, 536 show a similar phenomenon. These observations indicate the increasing of the active layer 537 depth, and the increase of soil moisture in the beginning of the melting season. In the 538 middle of the melting season, the active layer is thoroughly thawed, and the unsaturated 539 soil layer becomes saturated, and the value of Sumax_V and Sumax_D is reduced, resulting in larger 540 runoff generation. 541 For future studies, we recommend to pay particular attention to the hydrological processes 542 in the beginning of the melting season. Moreover, we recommend to develop process-543 based models to simulate the impact of the thawing soil on the temporal variation of the 544 unsaturated reservoir storage capacity (Sumax_V and Sumax_D), and the impact of freeze/thaw 545 processes on hydrological connectivity between surface and groundwater (D

547
Our knowledge on permafrost hydrology in mountainous regions is still incomplete. We 548 have collected numerous heterogeneities and complexities in permafrost regions, but most 549 of these observations are still not well considered in catchment scale hydrological modeling. 550 More importantly, we still largely lack knowledge on which variables play a more dominant 551 role at certain spatial-temporal scales, and should be included in hydrological models as 552 priority. 553 By conducting this study with field measurements and model experiments, we reached the 554 following conclusions: 1) correct meteorological forcing input is essential in mountainous 555 hydrological modeling; 2) distributed modeling based on topography and landscape is 556 important in cold regions with complex terrain; 3) baseflow recession in permafrost region is 557 well approximated by a linear reservoir, but the recession parameter Ks is much larger than 558 in other regions; 4) even without explicitly involving the freeze-thaw process, the 559 hydrological model can mimic and reproduce most parts of the hydrograph; 5) allowing 560 parameter dynamics improved model performance, especially in the beginning of the 561 melting season. Particular attention needs to be paid to understand and model the thawing 562 process at the beginning of the melting season, and its impacts on hydrological connectivity 563 at the catchment-scale. This diagnostic study benefits our understanding on permafrost 564 hydrology from measured data rather than arbitrary prior assumptions. We believe this 565 study is able to give us new insights into further implications to understand the impact of 566 permafrost on hydrology, and projecting climate change on permafrost hydrology.