Glacial lake area and glacier terminus position were manually mapped from optical satellite images (Table S1 in the Supplement). Landsat imagery (TM/ETM+/OLI, 30 m resolution) was used for the period from 1990 to 2015, and PlanetScope images (3 m resolution) were used for 2016 to 2025.

The area uncertainty associated with lake-boundary delineation from optical imagery primarily depends on image quality and can be approximated as ±0.5 pixel (Salerno et al., 2012). In addition, the uncertainty introduced by manual delineation can be assumed to follow a Gaussian distribution (Hanshaw and Bookhagen, 2014). We estimated the uncertainty (δ) in glacial lake area using Eq. (1) (Hanshaw and Bookhagen, 2014): 1δ=PG×G22×0.6872 where P is the perimeter of the glacial lake (m), G is the spatial resolution of the images used (m), 0.6872 is the revised coefficient under 1σ (i.e. approximately 69 % of peripheral pixels are subjected to errors).

We used 10 PlanetScope images (Table S2) to derive annual surface velocity fields for Jinwu Glacier from 2017 to 2025. To capture displacement over a full calendar year, we prioritized image pairs acquired in the second half of each year. After October, terrain shadows often obscure large parts of the lower glacier (Fig. S1 in the Supplement), reducing image texture and degrading feature tracking. We therefore mainly selected scenes from September and early October to minimize shadow effects. We also prioritized images with minimal cloud and snow cover to improve tracking reliability and spatial coverage.

Surface velocities of Jinwu Glacier were derived using the Coregistration of Optically Sensed Images and Correlation (COSI-Corr) technique implemented in ENVI (Leprince et al., 2007). This approach has been widely applied to glacier motion monitoring and has shown robust performance (Gao et al., 2025; Li et al., 2023; Singh Jasrotia et al., 2024). We used a correlation window of 150m×150m (50×50 pixels) with a step size of 30m×30m (10×10 pixels). The search range was set to 90m×90m (30×30 pixels). Systematic biases in the east–west and north–south displacement components were corrected using stable terrain. Stable terrain was defined as non-glacier surfaces unaffected by clouds, lakes, or topographic shadows, and was further restricted to pixels with slopes <30° (Fig. S2 in the Supplement). To suppress outliers, we excluded pixels with velocities exceeding four times the normalized median absolute deviation (NMAD) (Fig. S2). Glacier velocity (GV) was then calculated using Eqs. (2) and (3): 2D=Dx2+Dy23GV=D×AN where D is the displacement measured between each image pair, Dx and Dy are the east–west and north–south displacement components (m), respectively, N is the time interval between the image pairs (days), and A is the number of days per year (da-1), taken as 365. To further improve robustness, we retained only velocity pixels with a signal-to-noise ratio (SNR)>0.7. Local extreme outliers were then identified and removed using a 3×3 pixel moving window. For each window, the centre pixel was identified as an outlier and removed when its absolute deviation from the local median exceeded 1.5 times the median absolute deviation (MAD). Subsequently, velocity profiles along the glacier centreline were extracted at 50 m intervals, and profile points showing velocity differences greater than 15 ma-1 relative to adjacent points were excluded. To further evaluate the reliability of the derived velocity fields, we manually selected visually traceable surface features from the original image pairs for independent validation, as described in Sect. 5.2.

We assessed velocity uncertainty using residual displacements over stable, non-glacier terrain (Mattea et al., 2025; Paul et al., 2022). The mean residual displacement over stable terrain was 2.65 ma-1. NMAD values ranged from 0.82 to 2.58 ma-1 (Table S2). We additionally used the open-access global glacier surface velocity dataset ITS_LIVE (annual mosaics, 1986–2022, 120 m resolution) as a benchmark for comparison (Gardner et al., 2025).

The DEMs used in this study were generated from ASTER L1A stereo data using the NASA Ames Stereo Pipeline (ASP) (Beyer et al., 2018). This approach has been widely applied to glacier surface elevation measurements (Brun et al., 2017a; Bhattacharya et al., 2021). We produced three DEMs for the study area, dated 2 October 2002, 1 November 2014, and 13 June 2025. Surface elevation change was derived by DEM differencing. Because suitable ASTER stereo pairs over this glacier are limited, the available DEM epochs do not provide a strict pre- and post-GLOF comparison. In particular, the 2014–2025 differencing period spans both several years before and after the 2020 GLOF. Therefore, the DEM-differencing results are used to characterize multi-year surface elevation changes, rather than to isolate the direct event-scale elevation response to the GLOF. DEM biases may arise from limitations in DEM generation and from differences in processing strategies across datasets (Ke et al., 2020). To reduce these biases, we co-registered DEMs from different epochs following Nuth and Kääb (2011) and then removed the residual median vertical offset over stable terrain (Fig. S3 in the Supplement). Stable terrain was defined as pixels with slopes below 30° and absolute elevation differences smaller than four times the NMAD (Mattea et al., 2025). Pixels with absolute elevation differences larger than 150 m were excluded (Hugonnet et al., 2021).

We quantified the systematic bias and random errors of DEM differencing using stable, non-glacier areas. For 2002–2014, the mean elevation difference over stable terrain was 0.05 m and the NMAD was 5.66 m. For 2014–2025, the mean elevation difference was -1.68 m and the NMAD was 6.46 m. We modelled the dependence of DEM-differencing errors on terrain slope and maximum absolute curvature. Elevation differences over stable terrain were grouped into two-dimensional slope–curvature bins, and the NMAD was calculated within each bin. The binned NMAD values were interpolated to generate pixel-level elevation-change uncertainty maps. Elevation differences were standardized by these local uncertainties, and an empirical variogram was calculated from the standardized residuals and fitted with a double-range exponential model (Fig. S4 in the Supplement). The fitted variogram was used to estimate the effective number of independent samples, which was combined with the pixel-level uncertainties to estimate the uncertainty of spatially averaged elevation changes (Hugonnet et al., 2022; Mattea et al., 2025). We also used two open-access glacier surface elevation change datasets that cover the study area as independent references, spanning 2000–2019 (Hugonnet et al., 2021) and 2000–2016 (Brun et al., 2017a), as described in Sect. 5.2.

During 1990–2025, the terminus of Jinwu Glacier showed an overall retreating trend (Fig. 2). Measurements along the glacier centerline indicate that the glacier length decreased from 5699.00 m in 1990 to 4938.41 m in 2025, corresponding to a cumulative retreat of 760.59 m and a mean retreat rate of 21.73 ma-1 (Fig. 3a). Before the GLOF, the glacier terminus retreated continuously. From 1990 to 2019, the cumulative terminus retreat was 702.38 m, with a mean retreat rate of 24.22 ma-1. After the GLOF, changes in terminus position became smaller. In 2021, the glacier terminus advanced downstream by 23.02 m relative to 2020 (Fig. 4). It then retreated again during 2022–2025, with a cumulative retreat of 33.87 m and a mean retreat rate of 8.47 ma-1.

Over the same period, the area of Jinwuco increased from 0.371 ± 0.004 km2 in 1990 to 0.577 ± 0.004 km2 in 2019, representing an increase of approximately 56 % (Fig. 3b). After the 2020 GLOF, the lake area rapidly decreased to 0.272 km2, corresponding to a reduction of approximately 53 %, consistent with the results of Zheng et al. (2021b). During 2020–2025, the lake area remained relatively stable at 0.25–0.27 km2, with an area of 0.251 ± 0.004 km2 in 2025.

Following the June 2020 GLOF, ice flow near the glacier terminus shifted toward a lower velocity regime (Fig. 5a). In the 100–200 m section from the glacier terminus, mean ice-flow velocities decreased gradually from 48.43 ± 2.53 ma-1 in 2017 to 38.53 ± 3.36 ma-1 in 2020, followed by a marked decline to 27.03 ± 7.63 ma-1, accompanied by a change in the along-flow velocity gradient (Fig. 5b and c). The mean velocity was 42.97 ± 4.38 ma-1 during 2017–2020, but decreased to 24.09 ± 6.71 ma-1 (∼44 %) during 2021–2025. In the upstream 650–1600 m section, no pronounced slowdown was observed after the GLOF (Fig. 5d). The mean velocity changed from 37.55 ± 9.56 ma-1 during 2017–2020 to 35.24 ± 9.37 ma-1 during 2021–2025, representing a reduction of approximately 6 %. Over the longer term, ITS_LIVE suggests an overall deceleration of Jinwu Glacier since 2000 (Fig. 5c and d), although ice speed in the downstream differs between ITS_LIVE and our measurements. These discrepancies are further discussed in Sect. 5.2.

Over the past two decades, Jinwu Glacier has experienced sustained surface lowering (Fig. 6). The terminal 0–550 m section showed enhanced thinning, with the mean surface elevation change rate changing from -2.90 ma-1 during 2002–2014 to -3.71 ma-1 during 2014–2025. By contrast, thinning in the upstream 600–1550 m section slowed, with the mean rate changing from -1.17 to -0.87 ma-1 over the same period. This reduction was mainly driven by a slowdown in thinning within the 500–750 m section, where the mean surface elevation change rate changed from -1.20 ma-1 during 2002–2014 to -0.57 ma-1 during 2014–2025. A similar spatial pattern is also evident in the dataset of Hugonnet et al. (2021) for 2000–2019 and in that of Brun et al. (2017a) for 2000–2016 (Fig. 6). Notably, relatively large discrepancies are found among the different datasets within the terminal 0–550 m section (Fig. 6c), as further discussed in Sect. 5.2.

Following the 2020 GLOF, the downstream tongue of Jinwu Glacier showed a clear slowdown (Fig. 5a). The rapid loss of lake contact likely reduced hydrostatic pressure and buoyant support at the glacier front, thereby increasing basal and frontal resistive stresses, which contributed to the reduction in ice-flow velocity (Sutherland et al., 2020; Truffer and Motyka, 2016; Benn et al., 2007). After 2021, the near-terminus section maintained a relatively stable low-velocity state (Fig. 5c). In addition, PlanetScope imagery indicates degradation of the terminus ice cliff after the GLOF, while the glacier terminus position exhibited a slight advance during 2020–2021 (Fig. 4). This slight advance occurred together with ice-cliff degradation observed in PlanetScope imagery and ice-cliff collapse documented by UAV and field investigations on 30 October 2021 (Peng et al., 2023). These observations suggest reduced frontal stability shortly after the loss of lake contact and indicate that the mapped terminus change during 2020–2021 likely reflected short-term frontal morphological adjustment rather than a sustained advance of the glacier. These observations indicate that the geometry and stability of the glacier front changed after the drainage event. A comparable response has been reported at Longbasaba Glacier, where engineering works lowered the lake level by about 5.6 m in 2005. The glacier subsequently experienced accelerated retreat during 2005–2008 and an increase in ice speed above its long-term mean in 2006 (Liu et al., 2020). Liu et al. (2020) suggested that lake-level lowering may increase the surface slope near the grounding line, enhance grounding-line flux, and promote calving into the lake. Although Jinwu Glacier experienced a more abrupt and complete loss of lake contact, this case illustrates that rapid lake-level lowering can perturb glacier terminus dynamics. Taken together, the loss of lake contact modified the stress balance at the glacier terminus. The removal of hydrostatic pressure and buoyant support likely increased basal and frontal resistive stresses, contributing to the observed near-terminus slowdown, while also reducing the stability of the frontal ice cliff. Ice-cliff degradation, collapse activity, and the short-lived mapped terminus advance during 2020–2021 may therefore reflect the early adjustment of the glacier front to the changed terminus forcing. As the direct lake forcing was removed and a new terminus force balance established, subsequent glacier changes may have become increasingly governed by climatic forcing.

Surface elevation changes in the upstream sector are more robust than those near the terminus (Sect. 5.2). Within the 500–750 m reach, surface lowering slowed from -1.20 ma-1 during 2002–2014 to -0.57 ma-1 during 2014–2025. Hugonnet et al. (2021) reported a rate of -1.09 ma-1 for the same reach during 2015–2019 (Fig. S5 in the Supplement). Ice-flow velocity in the upstream sector remained broadly stable after the GLOF (Fig. 5d). An icefall separates the near-terminus region from the upstream glacier tongue. Previous studies have shown that steep icefalls and abrupt changes in ice thickness or bed geometry may reduce longitudinal stress transmission between glacier sections and limit the upstream propagation of dynamic perturbations originating near the terminus (McNabb et al., 2012; Durkin et al., 2017). Observations at Yahtse Glacier showed that interannual dynamic changes originating near the terminus were strongly attenuated above an icefall, suggesting that the icefall may inhibit upstream transmission of stress perturbations (Durkin et al., 2017). Regional observations have also shown that the upstream limit of seasonal velocity variations often coincides with the base of icefalls, suggesting that icefalls may separate glacier sections with different dynamic responses (Armstrong et al., 2017). Jinwu Glacier shows a spatial pattern consistent with these observations. The near-terminus section below the icefall (100–200 m) experienced a pronounced slowdown, whereas mean velocity within the 650–1600 m section upstream of the icefall changed only slightly. Given the potential for dynamic separation across the icefall, the upstream velocity record is therefore insufficient to infer a direct response to the GLOF. Because the 2014–2025 DEM-differencing period spans several years both before and after the GLOF, the slower thinning in the 500–750 m section cannot be directly attributed to either the GLOF or the near-terminus slowdown.

Proglacial lakes are commonly regarded as a significant non-climatic forcing factor influencing lake-terminating glaciers (Carrivick and Tweed, 2013). In contrast, a GLOF represents an event-driven perturbation that can abruptly alter glacier terminus boundary conditions by lowering lake level and removing ice–lake contact. The response of Jinwu Glacier further suggests that such events can modify the velocity structure and frontal morphology of the terminus sector. Given the increasing hazard potential of moraine-dammed glacial lakes under ongoing climate warming (Zheng et al., 2021c; Zhang et al., 2023b; Zhang et al., 2025), GLOFs should be considered event-driven processes in future assessments of lake-terminating glacier dynamics and associated hazards.

In this study, uncertainties in surface velocity and elevation change were assessed using residual statistics over stable terrain adjacent to the glacier, which allowed both systematic and random errors to be quantified. Nevertheless, these estimates cannot account for all potential error sources. For surface velocity, image matching can be degraded by repetitive or self-similar glacier surface features, such as crevasses, seracs, and landslide or avalanche deposits, which can increase noise and produce mismatches (Paul et al., 2015). For surface elevation change, additional uncertainties may arise from terrain dependent biases linked to slope and curvature, as well as from residual co-registration errors (Li and Hajnsek, 2025; Nuth and Kääb, 2011; Hugonnet et al., 2022).

In the velocity field, relatively high noise occurs within 200–450 m from the glacier terminus. This reach coincides with an icefall (Fig. 1e) and dense crevassing (Fig. 4), where image matching is likely less stable. Moreover, Jinwu Glacier is north facing, and this reach is more frequently affected by topographic shadowing (Fig. S1). This may partly explain the elevated noise in this area and the lower ITS_LIVE velocities observed there (Fig. 5b). To constrain the accuracy of velocity estimates, we performed an independent validation based on manual interpretation (Sato et al., 2022; Hyde et al., 2025). We selected 36 clearly identifiable tie points on the image pairs (Fig. S6 in the Supplement). The validation shows a root mean square error (RMSE) of 3.61 ma-1 between the PlanetScope velocities and the manually derived estimates (Fig. S7 in the Supplement). Two targeted validations further confirmed the robustness of the centerline velocity profiles and the near-terminus slowdown signal. Along the lower 0–500 m glacier tongue, 20 repeatedly traceable points covering the 2017–2025 annual velocity fields showed strong agreement between manual and automatic estimates, with an R2 of 0.92 and an RMSE of 3.84 ma-1 (Fig. S8a and b in the Supplement). In the near-terminus reach where the most pronounced slowdown was observed, nine additional validation points showed even higher consistency, with an R2 of 0.96 and an RMSE of 2.14 ma-1 (Fig. S8c and d). These validations suggest that, despite locally elevated noise in parts of the velocity field, the main velocity patterns reported in this study, especially the slowdown in the near-terminus section (Fig. S9 in the Supplement), are robust. In addition, we tested the robustness of the near-terminus velocity-gradient change using manually identifiable feature points along the glacier centerline from the 2017–2018 and 2023–2024 image pairs (Fig. S10 in the Supplement). Before the GLOF, velocities were relatively high and varied only slightly along the centerline, whereas after the GLOF they clearly decreased toward the glacier terminus. This distinct near-terminus deceleration gradient supports stronger resistive effects after lake drainage and provides direct evidence that the loss of lake–ice contact affected the near-terminus dynamic adjustment.

Surface elevation change estimates from different data sources show relatively poor agreement in the terminus sector (0–550 m, Fig. 6c). To quantify the consistency among datasets, we calculated the pairwise RMSE between elevation-change profiles at corresponding centerline sampling points and used the mean standard deviation among the profiles to represent their overall spread. Pairwise RMSE values within 0–550 m range from 0.75 to 0.97 ma-1, and the mean standard deviation among the three profiles is 0.532 ma-1. By contrast, agreement is improved in the upstream sector (600–1550 m), where RMSE decreases to 0.18–0.29 ma-1 and the mean standard deviation drops to 0.162 ma-1. This contrast indicates that surface elevation change estimates are subject to markedly greater uncertainty near the glacier terminus, whereas estimates from the upstream sector are comparatively robust. A likely explanation for this discrepancy is the complex and steep topographic setting of the Jinwu Glacier terminus. DEM elevation errors generally increase with slope (Hugonnet et al., 2022), and residual horizontal offsets in DEM differencing can be amplified over steep terrain, where they are translated into larger apparent elevation differences (Nuth and Kääb, 2011; Li and Hajnsek, 2025). In addition, the highly crevassed surface, pronounced local shadowing, and complex surface texture of the terminus may degrade DEM extraction, image matching and DEM co-registration quality (Berthier et al., 2014). These effects likely accumulate in the terminus sector and reduce the consistency of elevation-change estimates among datasets. Although different datasets yield somewhat different thinning near the terminus, they consistently indicate substantial surface lowering. Therefore, the conclusion of pronounced thinning near the glacier terminus is robust. However, because the 2014–2025 DEM-differencing interval spans several years both before and after the 2020 GLOF, it cannot isolate an event-scale elevation response. We use the DEM-derived elevation changes to characterize the longer-term thinning pattern of Jinwu Glacier. We further assessed the statistical uncertainty of our DEM-differencing results using stable, non-glacier terrain (Hugonnet et al., 2022; Mattea et al., 2025) (Fig. S11 in the Supplement). The uncertainty is generally larger in the terminus sector than in the upstream section, consistent with the stronger disagreement among independent datasets near the glacier front. In the 500–750 m section where thinning weakened most clearly, the uncertainty ranges of the two periods are largely separated, suggesting that the reduced thinning signal is unlikely to be caused solely by DEM-differencing uncertainty.