Quantification of Water Released by Thawing Permafrost in the Source Region of the Yangtze River on the Tibetan Plateau by InSAR Monitoring

The source region of the Yangtze River (SRYR, 1.4 × 105 km2 above Zhimenda station) on the Tibetan Plateau (TP) has 78% permafrost coverage. The streamflow depth increased at a rate of 2.5 mm/a since 2000. Quantification of the water contribution brought by permafrost thawing is a difficult task. In this study, we used Sentinel‐1 data and the SBAS‐InSAR technique to monitor terrain deformation from September 2016 to December 2021, and then utilized the long‐term deformation rate to assess ground ice meltwater release and the seasonal deformation to evaluate water storage in the active layer. Results reveal that 55.3% of the terrain in the SRYR has subsidence >2.5 mm/a, indicating widespread ground ice melting. The release rate of ground ice meltwater is 4.3 mm/a in the entire SRYR, above 6 mm/a at the Dangqu and Tuotuo River subbasins. The water release rate is relatively small (∼3%) in comparison to the streamflow depth of 151 mm per year during the investigation period of 2017–2021. We did not detect a strong increasing or decreasing trend among the 5‐year seasonal deformation, which reflects that the total soil water content in the active layer did not change significantly during the short investigation period. The results provide a data basis for ground ice richness and loss information in the SRYR and help to understand the impact of permafrost thawing on the regional water cycle in the permafrost environment.

. The large-scale permafrost in the Tibetan Plateau considerably weakens the exchange of groundwater and surface water on a certain temporal and spatial scale.
A substantial amount of ground ice is buried at the top layers of permafrost on the Tibetan Plateau (Cheng, 1983).According to estimates, the plateau has up to 12,700 km 3 of ground ice, twice that of plateau glacier ice, and the Yangtze River's source region has up to 935 km 3 of ground ice (Cheng et al., 2019).Over the past 50 years, air temperatures on the Qinghai-Tibet Plateau have increased by 0.3°C-0.4°Cevery decade, twice the rate of global warming (D.Chen et al., 2015).Consequently, permafrost is degrading, as evidenced by its areal extent shrinking, the ground temperature increasing, thickness decreasing, ground ice melting, and subsequent thermokarsting (S.L. Smith et al., 2022).The ice content buried below the permafrost table can exceed the natural pore volume of the soil (i.e., excess ground ice), which in case of thawing, will lead to settling and consolidation of the ground material (Wang et al., 2022).The surface subsides as a result of the ground ice meltwater being lost as surface or subsurface runoff (O'Donnell et al., 2016;Westermann et al., 2016).Jacques and Sauchyn speculated that permafrost thawing predominately causes increases in mean annual streamflow observed in Canada's Northwest Territories (St. Jacques & Sauchyn, 2009).The Golmud River Basin, which receives its water from the permafrost region in the Kunlun Mountains on the Tibetan Plateau, has experienced a sharp rise in groundwater levels since the 1980s and is therefore presumed that water released by permafrost thawing participates in the water cycle of the Golmud River (Wu et al., 2009).According to the monitoring in northwest Canada's lower Mackenzie and Peel River watersheds, the thawing of massive ground ice increases the stream water level (Kokelj et al., 2013).
Quantifying the amount of ground ice melting released water that contributes to surface or subsurface runoff is difficult due to the lack of direct measurement.By separating hydrographs using isotopes, Yang et al. (2019) estimated the contributions of ground ice meltwater to surface water runoff ranging from 13.2% to 16.7% in the Yellow River's source zone (Yang et al., 2019).The amount of ground ice meltwater could be alternatively estimated by multiplying permafrost's deepening/thawing rate with the average volumetric ice content.Based on this approximate calculation, Ma estimated that the potential release rate of ground ice meltwater is 2.2 mm per year in the Yellow River's source region, where the annual streamflow depth is 167 mm (Ma et al., 2020), and Zhang estimated that ground ice meltwater accounts for 12% of the lake's total water supply in the endorheic basin of the Tibetan Plateau (G.Zhang et al., 2017).
The variations in total soil water content (water storage) in the active layer during permafrost degradation processes, along with ground ice melting and water release, are less well studied.Most hydrological models have predicted that as the active layer deepens, the top layer of soil (0-20 cm) will eventually become drier (Andresen et al., 2020).Long-term monitoring along the Qinghai-Tibet engineering corridor also indicates that the soil moisture at the surface layer is stable or slightly declining despite increasing precipitation.Meanwhile, at the bottom of the active layer, the soil liquid water content has increased (L.Zhao et al., 2019Zhao et al., , 2021)).
Terrain deformation caused by permafrost activities (seasonal frost heave-thaw subsidence processes and long-term thawing settlements) can be captured and measured by the Synthetic Aperture Radar interferometry (InSAR) technique (Li et al., 2015;L. Liu et al., 2010;Reinosch et al., 2020;Short et al., 2014;Zwieback & Meyer, 2021).With millimeter-to-centimeter accuracy and a spatial resolution of tens to hundreds of meters, InSAR technology measures surface terrain deformation between SAR acquisitions.The multi-temporal InSAR technique (PS-InSAR and SBAS-InSAR) can, to some extent, reduce the decorrelation and atmospheric disturbances and obtain deformation time series (Berardino et al., 2002;Ferretti et al., 2001;Hooper, 2008;Lanari et al., 2004;Usai, 2003).SBAS-InSAR utilizes multi-master-image InSAR pairs of small temporal-spatial baselines to reduce the effects of decorrelation and is suitable for permafrost environments that undergo severe temporal decorrelation.Several studies have applied Sentinel-1 data SBAS-InSAR in monitoring deformation over permafrost terrain on the Tibetan Plateau (J.Chen et al., 2022;Daout et al., 2020;Daout et al., 2017;Lu et al., 2020;Reinosch et al., 2020;Wang et al., 2022aWang et al., , 2022bWang et al., , 2022;;X. Zhang et al., 2019).According to research, the seasonal deformation over the Tibetan plateau is normally less than 50 mm, and thaw subsidence rates are normally between 5 and 20 mm/a, with certain ice-rich permafrost regions having subsidence rates higher than 30 mm/year (J.Chen et al., 2022;Wang et al., 2022a).Recently, the implications of long-term subsidence trends and the seasonal deformation amplitudes have been elucidated.The seasonal deformation amplitude is shown to be proportional to the equivalent water depth in the active layer (Antonova et al., 2018;J. Chen et al., 2020;S. Liu et al., 2022), and the long-term subsidence rate directly reflects the melting of ground ice (Daout et al., 2020;Wang et al., 2022) and releasing of ground ice meltwater (Wang et al., 2022b).
Estimation of the water contribution of permafrost thawing is crucial to understanding the hydrological effect of permafrost degradation, and InSAR-measured deformation offers a chance to quantify the water released by permafrost thawing.In this study, we utilized the long-term deformation rate to measure the ground ice meltwater release and the seasonal deformation to assess the soil water content variation in the active layer.The deformation trend and seasonal deformation were first used together to estimate the water contribution of permafrost thawing.The result would also facilitate a deeper understanding of the status and changes in water resource structure in the Tibetan Plateau.

Study Area
The source region of the Yangtze River (SRYR) is in the interior Tibetan Plateau, between 90°33′ and 97°20′ East longitude and 32°26′ to 35°43′ North latitude.It is roughly 500 km long from east to west and has a total area of 1.4 × 10 5 km 2 , bounded by the Kunlun Mountains in the north, the Tanggula Mountains in the south, and the Hoh Xil, Ulan-Ula, and Zulken-Ula Mountains in the west.It is low-lying from the west to the east, and the Zhimenda hydrological station in the east is the outlet.Elevation ranges from 3,523 m to 6,569 m a.s.l., with a mean value of 4,764 m.At elevations above 5,500 m, glaciers are distributed (Figure 1b).There are 963 glaciers in the SRYR, based on the Second Chinese Glacier Inventory conducted in 2015, with a total area of 1,080 km 2 (∼0.8% of SRYR area) and the glacier volume of approximately 137 km 3 (Guo et al., 2015).The three main headwater rivers, the Tuotuo River in the west, the Dangqu River in the south, and the Chumaer River in the north, make up  (Zou et al., 2017) on the Qinghai-Tibet Plateau.Subfigure (b) depicts the study area's topography, major rivers, and subbasins.The three Sentinel-1 tracks that cover the study area are labeled in black frames, and red crosses mark the locations of our two InSAR reference points.Subfigure (c) shows the field photos of the land surface.
the source region.Accordingly, the study area is divided into five subbasins (Figure 1b and Table 1).Subbasin ④ Tongtian River I is located after the confluence of the Dangqu River and Tuotuo River, and subbasin ⑤ Tongtian River II is located after the confluence of Tongtian River and Chumaer River.
The study area has a high permafrost coverage of 78% (Table 1) (Zou et al., 2017).The upstream subbasins, ① Dangqu River subbasin, ② Tuotuo River subbasin, and ③ Chumaer River, all have permafrost coverage higher than 87%.Subbasin ⑤ Tongtian River II, in the lower latitude, has the smallest permafrost coverage of 65.6%.Subbasin ④ Tongtian River I has permafrost coverage of 70%.Taliks and seasonally frozen ground can be found close to the large rivers and south of the study area.Permafrost thickness varies from 10 to 312 m, and active layer thickness is typically 2-3 m (G.Liu et al., 2022;L. Zhao et al., 2020;Zhou, Liu, et al., 2022).The thermokarst ponds/lakes are widely distributed in flat plains with warm, ice-rich permafrost, such as the Beiluhe River plain, Chumaer River plain, and the Tongtian River plain (Figure 1b).Alpine steppe, meadow, swamp meadow, and alpine desert cover the land surface.Alpine meadows and alpine steppes make up the majority of the terrain surface.Alpine meadows are primarily found at elevations of 3,200-4,800 m, while alpine steppe is primarily found at elevations of 4,000-4,500 m.Above 4,900 m, there is little vegetation, showing an alpine desert environment.Swamp meadows develop on the banks of rivers and lakes as well as in low-lying areas with poor drainage.
The climate is in the transition zone between alpine semi-arid and semi-humid.Figures 2a and 2b depicts the mean annual air temperature and yearly precipitation at four meteorological stations (Tuotuohe, Zaduo, Qumalai, and Yushu) within or around the study area.Figure 2c depicts the runoff depth at the Zhimenda hydrological station from 1965 to 2021.The river discharge data is from the Yangtze River Water Resources Report (ChangjiangWater-ResourcesCommission, 2023;Jia et al., 2021).Based on the monitoring in this period, the mean annual air temperature is between −3.6°C and 3.5°C, rising at a rate higher than 0.3°C per decade.The yearly precipitation ranges from 260 to 496 mm, and there is a noticeable increasing trend of more than 22 mm/10a in 1965-2020.Most precipitation (80%) falls in summer and autumn, and there isn't much snowfall in winter.Over the past two decades, the streamflow depth has increased by 2.5 mm/a, from 87 mm before 2020  to 117 mm after 2020 (2020-2021).During 2017-2021, the annual streamflow depth was 151 mm, with an increasing rate of 11.5 mm/a.

Data
C-band Sentinel-1 (center frequency 5.4 GHz, or ∼5.55 cm in wavelength) SAR SLC data of IW mode and VV polarization acquired from September 2016 to December 2021 are used in this study.Three Sentinel-1 acquisition tracks (Orbits 150, 77, and 4) cover the study area.A total of 150 acquisition dates for orbit 150, 142 dates for orbit 77, and 136 dates for orbit 4 are used to derive the terrain deformation time series.

SBAS-InSAR Processing and Deformation Time Series Acquisition
In each Sentinel-1 track, we generated interferograms by every SAR image with its two sequential acquisitions, that is, for Scene n, with Scene n + 1 and n + 2 using InSAR Scientific Computing Environment (ISCE) software  (ISCE2, 2023).Multilooking with six pixels in the azimuth and 25 pixels in the range was applied in generating the interferogram.The network of unwrapped differential interferograms was generated after InSAR processing, and it was then transformed into a displacement time series using a weighted least squares estimator.The weight of the interferogram is determined by the inverse of its phase variance, and the raw phase/displacement time series were resolved by minimizing the phase residual (Y.Zhang et al., 2019).The raw displacement time series are relative to the first scene of the data sets in temporal and relative to the reference point in spatial.Then, tropospheric delay correction (Jolivet et al., 2014), phase deramping, and topographic residual correction (Y.Zhang et al., 2019) were applied to the raw displacement time series to reduce noises and improve deformation retrieval accuracy.The aforementioned processing was carried out by MintPy (MintPy, 2023; Y. Zhang et al., 2019).The retrieved deformation was geocoded and finally presented in a stack of 100 m × 100 m spatial grid.To reduce the impacts of noise-contaminated extreme values, we also applied a 3-size moving window filter on the deformation time series.The processing details and parameter settings can be referred to our previous works (Wang et al., 2022b).The InSAR-derived deformation is relative to the reference point.We set two reference points for the entire study area, one in the overlapping area of orbit 150 and orbit 77, and one in the area of orbit 4. They are located on mountain ridges and have interferometric coherence close to 1.The longitudes and latitudes of the reference points are 92.92032°E,34.71586°N, and 95.22896°E 35.46801°N, as marked in Figure 1b.

Extraction of Long-Term Deformation Rates and Seasonal Deformation
The permafrost terrain surface experiences both long-term subsidence/uplift deformation trend and seasonal oscillations (frost heave-thaw subsidence) (Daout et al., 2017;Li et al., 2015).Therefore, we used a linear trend combined with a sinusoidal (seasonal) function to represent the deformation time series as Equation 1.
where v is the long-term deformation rate, t is the time interval relative to the first SAR image acquisition date, A is the periodic/seasonal amplitude, T is the period of the seasonal undulation (assumed to be 1 year), φ is the initial phase, and c is the residual term.The parameters v, A, φ, and c can be solved by the deformation time series.The periodic/seasonal peak-to-peak (heave-subsidence) deformation magnitude is 2|A|, which is twice A in Equation 1.The seasonal deformation appeared in this work refers to the peak-to-peak (heave-subsidence) deformation magnitude value (2|A|).Supplementary information Text S1 and Figure S1 in Supporting Information S1 provides the performance assessment of the deformation model.
For flat terrain, deformation is caused mainly by permafrost activities, that occur predominantly in the vertical direction.Hence, assuming no horizontal movement, the InSAR-measured deformation in the line-of-sight direction can be translated into the vertical direction using Equation 2 divided by the cosine of the sensor's incidence angle.
where Defor LOS is the deformation in the sensor's line-of-sight (LOS), Defor vertical is the deformation in the vertical direction, and θ is the incidence angle of the sensor.
We extracted the long-term deformation velocity v and seasonal deformation 2|A| from the displacement time series pixel by pixel in each orbit and then merged the results from the three orbits together.The deformation values shown in this study all refer to deformations in the vertical direction.

Translation From Terrain Deformation to Ground Ice Meltwater Release
The volumetric ice content in frozen soil can exceed the natural pore volume of the soil (i.e., excess ground ice).Upon thawing, the excess ice (water) volume is squeezed out and drains from the ground material and this leads to the settling and consolidation of the ground material.The released meltwater is either lost as surface or subsurface runoff, or it pools up, forming an inundated land surface such as a thermokarst pond or lake (Westermann et al., 2016).The depressions in the landscape in ice-rich permafrost environments are evidence of excess ground ice melting.We assume that the long-term subsidence is equivalent to the thickness of excess ground ice released from the original ground material.The released ground ice meltwater can be calculated via the following conversion of Equation 3 (Wang et al., 2022b).
where ∆V water represents the ground ice meltwater released to the hydrological cycle, ∆V ice represents the corresponding changes in the ice volume, ∆h is the changes in terrain elevation brought by ground ice melting, C grid is the grid cell size, and d v denotes the long-term deformation velocity in the vertical direction.During the conversion, the areas with slope angles >10° and seasonal deformation magnitude ≤3 mm are masked out.The threshold settings can be referred to (Wang et al., 2022b).
It is worth noting that the loss of ground ice storage is, in reality, greater than the amount of ground ice meltwater indicated in Equation 3.This is because not all liquid water drains from the ground material; some of the water remains within the soil column.

Detection of the Variation of Seasonal Deformation Amplitude
Furthermore, we are interested in whether the seasonal deformation amplitude is constant or variant over the years since the seasonal deformation is proportional to the equivalent water depth in the active layer.First, we removed the linear trend from the displacement time series, and then calculated the seasonal deformation in each calendar year using the sinusoidal (seasonal) function.After the seasonal deformation is derived for each year individually (2017-2021, 5 years), we calculated the trend of seasonal deformation variation.Over the area where seasonal frost heave and thaw subsidence is insignificant, the seasonal variation is difficult to detect by this calculation method and the results would have large uncertainties; therefore, we only present the result of seasonal deformation variation in areas having distinct seasonal deformation >10 mm.

Accuracy of SBAS-InSAR Derived Terrain Deformation
Ground leveling measurements conducted at Wudaoliang leveling observation site are used to verify the InSAR-derived deformation (Zhou, Zhao, et al., 2022).The InSAR-derived deformations are interpolated to the date of leveling measurements through the linear interpolation of the two InSAR deformation values before and after leveling measurements.The SBAS-InSAR derived deformation curves and leveling measurements are shown in Figure 3. Pearson r between SBAS-InSAR monitoring and leveling measurements reach 0.95, and Root Mean Squared Error (RMSE) is 8.11 mm.It demonstrates that InSAR-derived deformations and leveling measurements have a good consistency.-2019: 20170109, 20170305, 20170524, 20170812, 20171218, 20180529, 20180815, 20181004, 20190109, 20190428, 20190815, 20190924 (format: YYYYMMDD).

Spatial Distribution of Long-Term Deformation Rates and Seasonal Deformation
Figure 4 displays the spatial distributions of long-term deformation velocities and the seasonal deformations.The SRYR exhibits extensive thaw subsidence: 55.3% of the terrain has subsidence >2.5 mm/a.As the Mann-Kendall trend test indicated, 89.8% of the total pixels exhibited significant trends in long-term velocity, as shown by p-values below 0.01.Furthermore, 86.8% of the pixels demonstrated even stronger significance, with p-values below 0.001 (Figure S2 in Supporting Information S1).Among the terrains that were subsiding (subsidence rates >2.5 mm/a), 1.1% of the area experienced subsidence rates greater than 30 mm/a, 8.4% experienced subsidence rates between 20 and 30 mm/a, 36% experienced subsidence rates between 10 and 20 mm/a, 33.3% had subsidence rates between 5 and 10 mm/a, and 21.2% had subsidence rates of less than 5 mm/a.Terrain subsiding is observed along the Tuotuo River, upstream of the Dangqu River, and the middle stream of the Chumaer River, but is less frequently seen along the large Tongtian River due to the relatively low altitude and high temperatures near large rivers.Tuotuo River subbasin undergoes the most extensive terrain subsidence, and 61% of the terrain surface endures subsidence greater than 5 mm/a.
Deformation properties are different in the five subbasins.Figure 5 depicts the density plots of long-term velocities and seasonal deformation within the five subbasins.Subbasins ① Dangqu River subbasin and ② Tuotuo River exhibit the most severe subsidence trends, with the average subsidence rate in the subbasin up to higher than 7 mm/a.The average subsidence value in the subbasins ③ Chumaer River and ④ Tongtian River I is between 4 and 5 mm/a.With only 1.2 mm/a, the lowest subbasin, subbasin ⑤ Tongtian River II, has a much smaller subsidence trend than the other four subbasins.Regarding seasonal deformation, the subbasins ③ Chumaer River subbasin, ② Tuotuo River, ① Dangqu River all have similar magnitudes of 12-14 mm, followed by the subbasin ④ Tongtian River I and subbasin (10.19 mm) ⑤ Tongtian River II (7.3 mm).The average seasonal deformation in the entire SRYR was 11.5 mm.

Variations of Seasonal Deformation During 2017-2021
The variations of seasonal deformation during 2017-2021 in the SRYR are shown in Figure 6.The seasonal deformation's variation magnitude is much smaller than that of the long-term deformation trend.The majority of values fall within the small range, typically between −0.3 and 0.3 mm/a.The change trend is not significant.Based on the distribution of P-values resulting from the Mann-Kendall trend test (Figure S3 in Supporting Information S1), merely 5.1% of total pixels in the SRYR exhibited P-values below 0.1, with an even smaller percentage of only 1.2% demonstrating P-values below 0.05.According to the statistics of five subbasins shown in Figure 7, only subbasin ① Dangqu River exhibited a gently increasing pattern of seasonal deformation, and subbasin ② Tuotuo River exhibited a gently decreasing pattern.Figures 6b-6f illustrates displacement curves at three selected sites.These three sites can represent the variation characteristics within a certain surrounding area and exhibit different seasonal deformation variation trends.Site 1 is located in a flat landscape, and its seasonal deformation is slightly amplified from 2017 to 2021.Site 2 is in a fluctuating terrain, and its seasonal deformation amplitude is slightly diminished.Site 3 is on a hilly slope, and unlike the other two sites, it doesn't show a clear trend in its seasonal deformation.Although the three sites all manifest a distinctive long-term subsidence velocity, the seasonal deformation variation trend differs.

Melting Ground Ice
The terrain deformation velocity is converted into the release rate of ground ice meltwater, according to the method stated in Section 3.2.3.The map of the meltwater release rate is shown in Figure 8.The terrain subsidence, correspondingly ground ice melting and water release, are the dominant signal in the SRYR.Weak uplift signals, and correspondingly positive numbers, are also occasionally observed in certain places.However, the uplift signal doesn't always relate to permafrost aggradation/ground ice accumulation.In most places in the study area, the uplift signal is more likely caused by sediment accumulation (Wang et al., 2022b) or variations in the soil moisture content (Reinosch et al., 2020).The magnified views of the yearly ground ice melt water are shown at two sectors aline with Landsat images in Figures 8b-8e.In sector 1, positive values are distributed along the river banks, reflecting the sediment accumulation.In sector 2, positive values are distributed in the wetland encircled by a network of waterways, indicating the impact of variations in the soil moisture content.The positive values in both two sectors are unrelated to ground ice aggradation.Additionally, the uplift values are very small, so their impact on the estimation of the total amount of ground ice meltwater is minimal.
The volumetric rate of meltwater release is converted into the runoff depth in the watershed (Table 2).The release rate of ground ice meltwater is 589.8 × 10 6 m 3 /a in the entire SRYR, yielding an increase in runoff depth of 4.3 mm/a, which is quite small when compared to the annual streamflow depth of 151 mm (2017-2021, Figure 2).The meltwater release rate is the highest in the subbasin ① Dangqu River subbasin (6.5 mm/a), followed by subbasin ② Tuotuo River subbasin (6.3 mm/a), subbasin ④ Tongtian River I subbasin (4.2 mm/a), subbasin ③ Chumaer River subbasin (3.9 mm/a), and subbasin ⑤ Tongtian River II subbasin (1.4 mm/a).

Water Storage Variations in the Active Layer
The seasonal deformation magnitude is proportional to the active layer's total water content (water storage) (J.Chen et al., 2020).Generally, the higher the water content in the active layer, the larger the seasonal heavesubsidence deformation.Additionally, if the total water content in the active layer varies from year to year, the seasonal deformation's variations shall reflect these changes.According to the map of seasonal deformation variation (Figure 6) and the statistics of five subbasins (Figure 7), the seasonal deformation variation is very small in the SRYR.The seasonal deformation variation trend is estimated only by five amplitude values (2017-2021, one seasonal deformation amplitude each year).It has large uncertainties and only reflects that the total water content in the active layer doesn't change significantly and doesn't manifest a strong increasing or decreasing trend during the investigation period.It implies that most of the water from melting ground ice is released as surface or subsurface runoff and the supply to the active layer is a small amount during 2017-2021.

Comparison of Ground Ice Meltwater Derived From InSAR Monitoring and Modeling in the SRYR
Wang et al. ( 2023) evaluated the contribution of ground ice meltwater to river runoff using the Geomorphology-based Ecohydrological Model over the past four decades  across TP.The results indicate that in the SRYR, ground ice meltwater amounts to approximately 0.47 Gt/a.This number is slightly smaller than the estimate of approximately 0.59 Gt/a calculated based on ground surface deformation.These discrepancies can be attributed to various factors.Firstly, the two studies have different study periods.The model simulation spans four decades, and it's possible that the rate of permafrost thawing in the 1980s was slower than the present rate.Secondly, the primary variables influencing ground ice meltwater are the abundance of ground ice and the rate at which it is melting.Currently, the lack of detailed ground ice data in terms of spatial distribution and depth profiles poses challenges when parameterizing the model.

Comparison of Ground Ice Meltwater With Other Permafrost-Covered Watersheds
Ground ice meltwater exhibits significant variability across different watersheds.Taking SRYR for example, the contribution of the ground ice melt water to streamflow in SRYR is ∼3%; however, in the Tuotuo River subbasin (the annual runoff depth is ∼51.9 mm), the contribution of the ground ice melt water to streamflow is ∼12%.In this section, we compared the ground ice meltwater in the SRYR with other permafrost-covered watersheds for which ground ice meltwater estimates are available.
In the source region of the Yellow River (∼12.1 × 10 4 km 2 , ∼34% permafrost coverage, annual streamflow depth ∼167 mm during 1961-2017), Ma et al. (2020) calculated the potential release rate of ground ice meltwater by model estimation.Their findings revealed that the release rate of ground ice meltwater amounts to 2.2 mm per year, contributing approximately 1.3% to the streamflow (Ma et al., 2020).Notably, the headwater area of the Yellow River, located above the Huangheyan hydrological station, which has extensive permafrost coverage of 86%, has a very high release rate of ground ice meltwater of 5.6 mm per year.This significant contribution constitutes approximately 14.4% of Huangheyan's yearly streamflow.For the entire TP permafrost region, a recent model simulation estimated that ground ice meltwater amounts to 2.2 Gt per year from 1980 to 2019, contributing approximately 2.1% to river runoff (Wang et al., 2023).
In the TP endorheic basin (∼70.8 × 10 4 km 2 , ∼7.7 ± 0.6 Gt per year rapid increase in the lake water volume), G. Zhang et al. (2017) conducted a water budget study for this region during the years 2003-2009 using a modeling approach.Their study revealed that the majority of the water supply is provided by increasing precipitation (74%), followed by glacier mass loss (13%), and ground ice loss (12%) (G.Zhang et al., 2017).In the largest watershed of TP's endorheic basin, the Selin Co watershed (∼4.4 × 10 4 km 2 , ∼30.2% permafrost coverage, ∼0.5 Gt per year increase in lake volume during 2018-2020), ground ice meltwater made up about 12% of the water supply for the lake volume increasement during the years 2017-2020 (Wang et al., 2022b).
Based on the available data, the contribution of ground ice meltwater to lake volume increase and streamflow is relatively modest, typically on the order of 10% or 5%.This estimation aligns with the findings of a previous study conducted in the Siberian Arctic.The study simulated the impact of ground ice melt on hydrological responses using the Community Land Model (CLM4.5)and concluded that ground ice meltwater is a small amount of water in relation to total runoff (Lee et al., 2014).

Contributions to Increasing River Flow in the SRYR
The increasing discharge in the SRYR over the past decades has garnered much attention, and many studies have attempted to quantify the contributions of various water sources.According to discharge monitoring at Tuotuohe and Zhimenda hydrological stations from 1964 to 2018, the annual average discharge at Tuotuohe station was 43-50 m 3 /s in the 1960s-1990s and increased to 86-98 m 3 /s in the 2000s-2010s, increasing by 97%, and the annual average discharge at Zhimenda station was 346-447 m 3 /s in the 1960s-1990s and increased to 476-499 m 3 /s in the 2000s-2010s, increasing by 25% (Jia et al., 2021).In areas underlain by permafrost, river discharge is primarily regulated by rain and snowmelt that is routed directly over the surface because groundwater recharge to, and discharge from, deeper aquifers is largely obstructed by the presence of permafrost (Bense et al., 2012).As evidenced in Figure 2, the river runoff in the SRYR is primarily controlled by precipitation.Based on the slope change ratio of cumulative quantity method, the contribution rate of precipitation to the runoff change is estimated to be 67.5%, and the comprehensive contribution of air temperature, glaciers, permafrost and other cryosphere elements to runoff change is 32.5% in the entire SRYR, while in its Tuotuohe subbasin, the contribution rate of precipitation to the runoff change is 23.0%, and the comprehensive contribution of air temperature, glaciers, permafrost, and other cryosphere elements reach 77.0% (Jia et al., 2021).Meanwhile, Tang et al. studied the relationship between changes in meteorological elements and large-scale circulation factors in the SRYR and detected that precipitation is the primary factor controlling river runoff, with a contribution rate of roughly 63% (Tang et al., 2018).
In the SRYR, the glacier ice reserve is 137 km 3 (Guo et al., 2015), and glacier mass loss is another important cryosphere element that contributes to river runoff.Glacier runoff estimations in existing literature exhibit significant levels of uncertainty.By using a degree-day approach, glacier runoff accounts for 11.0% of the stream flow during 1961-2000 (S.Liu et al., 2009).However, based on the glacier mass changes by comparing the digital elevation models (DEMs) generated by ZiYuan-3 tri-stereo sceneries with the SRTM DEM, the glacier mass loss contributes only 2% to the river discharge between 2000 and 2018 (L.Liu et al., 2020).Using a time series of DEMs derived from satellite stereo-imagery during 2000-2016, the glacier mass balance result indicates that the glacier mass loss is −0.5 ± 0.3 Gt a −1 , accounting for ∼3% of river discharge (Brun et al., 2017).Recent model simulations reveal that the contribution from glacier mass loss is smaller than that from permafrost thaw (Wang et al., 2023).
During the InSAR investigation period in this study, the annual runoff depth was 151 mm and clearly exhibited an increasing rate of 11.5 mm per year.Although precipitation data for the year 2021 is lacking, a 5-year record from 2016 to 2020 already indicates an overall increase in precipitation in the study area, with rates of 8 mm/a for Tuotuohe station, 39 mm/a for Qumalai station, 47 mm/a for Zaduo station, and 15 mm/a for Yushu station.Meanwhile, the average annual air temperature does not show significant signs of change.The ground ice meltwater release rate was approximately 4 mm/a, contributing roughly 3% of the river runoff.Based on the above analysis, the dominant factor behind the increase in river discharge appears to be the rising precipitation.Contributions from permafrost thaw and glacier melt, while present, are limited.However, given the substantial ground ice reserve (935 km 3 ) (Cheng et al., 2019) and permafrost thawing, the contribution of ground ice meltwater to river runoff should not be neglected.

Influencing Factors of Active Layer Water Storage
In the permafrost region on the Tibetan Plateau, the volumetric soil water content is generally around 10-30% at the near-surface during the thawing season, exhibits drying from evapotranspiration and rewetting from precipitation events, and is typically saturated >40% at the bottom of the active layer.Permafrost thawing leads to the redistribution of the soil water in the active layer and the drying of the near-surface soils (L.Zhao et al., 2019).In the SRYR, we didn't detect strong seasonal deformation variations, but different variation patterns were observed (Figure 6).Microtopography strongly influences the active layer groundwater flow (O'Connor et al., 2019) and the lateral redistribution of water in the active layer.We speculated that the microtopography might be related to the diverse patterns of seasonal deformation variation observed at the three sites in Figures 6h-6j.During the permafrost thawing process, if the extra ground ice meltwater from thawing permafrost can quickly drain through subsurface runoff, the amount of water stored in the active layer won't change much; in contrast, if the microtopography is very flat and the lateral flow is weak, the water storage in the active layer will increase because ground ice meltwater accumulates.We have observed a significant seasonal deformation amplification in the flat alluvial plain of Hoh Xil near the northern boundary of the permafrost region of the Tibetan Plateau during 2015-2020.Additionally, the soil material (coarse-grained vs. fine-grained deposits) or the hydraulic properties of soils greatly affect the drainage speed of ground ice meltwater.It might be a dynamic and interactive process between permafrost thawing and active layer water drainage, that permafrost thawing induces localized geomorphic changes, such as uneven ground surface subsidence and thaw slumps, and these changes in microtopography affect the drainage processes in turn.
Meanwhile, the variations of precipitation and evapotranspiration should not be neglected when examining active layer water storage variations because they significantly influence the active layer's water content.Compared to the years 2015-2016, the annual precipitation increased significantly after the year 2017 (Figure 2b).Thus, we have detected a large frost-heave and thaw subsidence seasonal deformation in the following year 2018 at all three sites (Figures 6h-6j).Limited by the short investigation period and large uncertainties in the seasonal deformation variation result, we didn't conduct a detailed analysis of active layer deformation variations and their influencing factors.However, the analysis so far has clearly shown that the total water content in the active layer did not exhibit a particularly strong increasing or decreasing trend throughout the short investigation period, although the water storage displays yearly dynamics.

Limitations and Prospects
Permafrost thaw and subsequent changes to regional hydrology vary across space and time (O'Donnell et al., 2016).Many long-term effects of permafrost thaw, for example, the reorganization of vegetation, hydraulic connections, and flow paths also have impacts on regional hydrology (Jorgenson et al., 2013;Walvoord & Kurylyk, 2016).Numerous thermokarst ponds/lakes are spread over the study area (Wei et al., 2021), and lake water might be quickly drained when the permafrost at the bottom of the lake is thaw penetrated (Jones & Arp, 2015;L. C. Smith et al., 2005;Șerban et al., 2021).These processes are not in the scope of this study and are not considered.
The subsurface water storage changes in permafrost regions on TP (Jing et al., 2019;Yang et al., 2023;Zou et al., 2022) are intricately linked with permafrost thawing, and ground surface deformation variations can provide valuable insights into understanding these changes.However, it's essential to acknowledge the limitations of this study.While the long-term deformation analysis benefits from data with ∼150 SAR acquisition dates, providing sufficient information to identify long-term ground subsidence or uplift trends, the same is not true for studying change trends in active layer water storage.The 5-year span for seasonal deformation studies can only represent a very small data set of five sample sizes, introducing a higher level of uncertainty.As indicated by the significance of Mann-Kendall trend test of the long-term velocity, where 89.8% of the total pixels in the SRYR exhibited P-values below 0.01 (Figure S2 in Supporting Information S1), which confirms the significance of the ground surface's long-term trend.However, for studying variations in seasonal deformation, the time span is insufficient to detect significant changes, as evident from the significance test results with merely 1.2% of total pixels exhibiting P-values below 0.05 (Figure S3 in Supporting Information S1).In future work, we plan to extend the investigation period and expand our research to the entire TP region.

Conclusions
This study presented a quantitative analysis of the effects of permafrost thawing on the water balance in the SRYR on the Tibetan Plateau by using ground terrain deformation monitored by Sentinel-1 SBAS-InSAR from September 2016 to December 2021.The long-term subsidence rates and seasonal deformation extracted from the deformation time series are used to explore the ground ice melting and the water storage in the active layer, respectively.It is the first time subsidence rates and seasonal deformation were used together to investigate the hydrological effect of thawing permafrost in the watershed.
Long-term deformation trends reveal that 55.3% of the terrain has subsidence >2.5 mm/a, and the terrain subsidence rates were typically 5-20 mm/a, indicating extensive thaw subsidence and ground ice melting in the SRYR.The rate of ground ice meltwater release is 4.3 mm per year in the entire SRYR, above 6 mm per year at the Dangqu River subbasin and Tuotuo River subbasin, 4 mm per year at the Chumaer River subbasin and Tongtian River I subbasin, and 1.4 mm per year at the Tongtian River II subbasin.Compared to the annual streamflow depth of 151 mm (2017-2021), the amount of released ground ice meltwater is not large (∼3%).However, it is essential to emphasize that despite its modest scale, this water resource is comparable in magnitude to glacier meltwater.Considering the substantial reserves of ground ice and the continuous process of ground ice melting, the significance of ground ice meltwater should not be overlooked.
Seasonal deformation was normally smaller than 50 mm, and mostly less than 30 mm in this region.The 2017-2021 5-year seasonal deformation variation magnitudes are typically between −0.3 and 0.3 mm/a, much smaller than the long-term trend magnitudes.It indicates that the total soil water in the active layer doesn't manifest a strong increasing or decreasing trend during the short investigation period.It might also hint that most of the water from melting ground ice is released as surface or subsurface runoff and the supply to the active layer is a small amount during the investigation period.The results provide a data basis for ground ice richness and loss information in the SRYR and help to understand the impact of permafrost thawing on the regional water cycle in the permafrost environment.

Data Availability Statement
The Sentinel-1 Level 1 single-look complex (SLC) images for InSAR deformation monitoring can be accessed from the Alaska Satellite Facility at https://search.asf.

Figure 1 .
Figure 1.Subfigure (a)  shows the Yangtze River's location and the permafrost distribution(Zou et al., 2017) on the Qinghai-Tibet Plateau.Subfigure (b) depicts the study area's topography, major rivers, and subbasins.The three Sentinel-1 tracks that cover the study area are labeled in black frames, and red crosses mark the locations of our two InSAR reference points.Subfigure (c) shows the field photos of the land surface.

Figure 2 .
Figure 2. (a) Mean annual air temperature and (b) annual precipitation at four meteorological stations (Tuotuohe, Qumalai, Yushu, Zaduo) within or near the Yangtze River source region, (c) Mean annual land surface runoff depth from 1965 to 2021, calculated from discharge measurements at Zhimenda gauging station.The locations of the Zhimenda gauging station and four meteorological stations are marked in Figure 1b.

Figure 4 .
Figure 4. Map of the long-term deformation trend velocity (a) and seasonal deformation (b), both in the vertical direction.The "na" in dark gray denotes that there the deformation could not be obtained due to lack of data or decorrelation.In subfigure (a), the positive and negative numbers represent the upward and downward movement of the terrain surface.

Figure 5 .
Figure 5. Density plots of (a) deformation trend velocities and (b) seasonal deformation magnitudes in different subbasins.

Figure 6 .
Figure 6.(a) Map of seasonal deformation variation.(b)-(d) show three sites' landscapes from Landsat 8 image (SWIR1-NIR-red RGB composite, acquired on October 2020), (e)-(j) show deformation curves in the LOS direction.(e)-(g) are the original deformation curves; in (h)-(j) the long-term trends are subtracted from the original curves.The vertical red letters label the annual seasonal deformation.

Figure 7 .
Figure 7. Density plots of seasonal deformation variations in different subbasins.

Figure 8 .
Figure 8.(a) Map of the release rate of ground ice meltwater.The grid's color denotes the yearly volumetric ground ice meltwater in the grid cell; negative values represent ground ice meltwater release, and positive values indicate conversion result from uplift deformation values, not necessarily ground ice accumulation.The subfigures (b)-(e) show the magnified views of the two sectors marked with black rectangles in subfigure (a), (b)-(c) for sector 1 and (d)-(e) for sector 2. The left column shows Landsat 8 images (SWIR1-NIR-red RGB composite, acquired on October 2020).The right column shows ground ice meltwater (sharing the same color bar as the map in subfigure a) overlaid on the hillshade.The black line in subfigure (c) delineates the track boundary of orbit 150.The seamless stitching of the deformation results from two nearby tracks indicates that the deformation properties extracted from the two orbits are almost the same, and the deformation results are robust.

Table 1
Characteristics of the Subbasins in the Source Region of the Yangtze River