Glaciers, debris-covered glaciers, rock glaciers (active/inactive) and perennial snowfields across the Argentinean Andes and South-Atlantic islands cover an area of 8484 km² and are documented in the National Inventory of Glaciers conducted by the Argentine Institute for Snow, Ice and Environmental Sciences (IANIGLA-CONICET) in collaboration with the Argentine Ministry of the Environment and Sustainable Development (Zalazar et al., 2017). With the aim of preserving glaciers and the periglacial environment, the dataset was established based on satellite imagery and ground-truthing as requirement of the law on Minimum Standards for the Preservation of Glaciers and the Periglacial Environment (span. Régimen de Presupuestos Mínimos para la Preservación de los Glaciares y del Ambiente Periglacial) (ibid.).

We rely on the National Inventory of Glaciers for glacial and periglacial feature boundaries in our study area and conduct our vertical surface changes and horizontal velocity analysis within these boundaries. For our analysis, we treat rock glaciers indifferent of their mapped state of activity allowing us to rely on our measured horizontal velocity quantification rather than the visual interpretation of surface features as conducted during the establishment of the inventory (Zalazar et al., 2017, 2020).

The Pléiades 1A and 1B satellites were launched on 17 December 2011 and 2 December 2012 respectively. With a sun-synchronous orbit type and repeat cycle 26 d, they offer panchromatic (resolution 0.5 m) and multispectral (resolution 2 m) imagery in stereo and tristereo mode (ASTRIUM, 2012). Achieving the (tri)stereo cover during one pass over the area, a homogeneous product which allows for most precise DEMs to be generated is acquired despite the challenging terrain of the Andes – and allows for a suitable base-to-height (B / H) ratio. In contrast to radar imagery, the optical signal does not penetrate snow and ice, increasing suitability for cryospheric surface change detection (Berthier et al., 2014). All Pléiades data used in this study were tasked for austral summers, in stereo (2019) or tristereo (2022–2025) mode and processing level 1 corresponding to the primary product (ASTRIUM, 2012), Table 1.

We process all Pléiades (tri)stereo pairs automatically using Ames Stereo Pipeline (ASP) software (Beyer et al., 2018). Processing is conducted without ground control points (GCPs) relying on rational polynomial coefficients (RPCs). We follow Berthier et al. (2024) and Cusicanqui et al. (2025) in using a STRM DEM as seed DEM during stereo processing and a semi-global matching strategy. Here, either both stereo acquisitions or both pairs of the tri-stereo triplet are included. As a result, we generate DEMs at 1 m resolution. Given the data quality we do not fill gaps and proceed without correcting for sensor undulations.

We conduct 78 repeated Differential Global Navigation Satellite System (DGNSS) measurements on Dos Lenguas and El Paso rock glaciers as well in the Agua Negra Glacier forefield using Trimble DGNSS equipment (R8 base, R2s rover, TSC3 handheld, RTK), Table 2. Located on the landforms as well as surrounding terrain, the measurements are conducted in at least 2 consecutive years in the austral summers of 2022, 2023 and 2024; within maximum 2 consecutive days. Measured point locations are marked on flat surfaces of selected large boulders (> 2 m). The coordinates of the base stations are 424101.83, 6654084.621, 4247.137 m for Dos Lenguas rock glacier; 422350.62, 6655630.667, 4723.248 m for El Paso rock glacier and 422783.389, 6661216.068, 4726.973 m for the Agua Negra Glacier forefield – all provided in WGS 84 UTM zone 19S. DGNSS measurements on Dos Lenguas rock glacier have been used for validation and georeferencing (Stammler et al., 2024, 2025a) and are published in Stammler et al. (2025b). We publish the additional DGNSS measurements used in this study in the paper-accompanying dataset (Stammler et al., 2026).

We buffer the inventoried landforms and generate rectangular bounding boxes, resulting in bounding boxes with minimum 500 m distance to each of the (debris-covered) glacier and rock glacier polygons to clip our Pléiades-based DEMs. Consequently, we co-register the younger to the older clipped DEMs following Nuth and Kääb (2011) in Demcoreg (Shean et al., 2016), while masking the cryospheric landforms based on the National Inventory of Glaciers (IANIGLA-CONICET, 2018). Single Pléiades tiles are processed separately to prevent distortion from mosaicking. If available for both acquisitions, each tile is co-registered respectively, e.g., 2023T to 2022T, 2023B to 2022B, 2023R to 2022R. For acquisitions with different extents and/or number of tiles, co-registration is only possible where data is available and is conducted as, e.g., 2022T to 2019, 2022B to 2019, 2022R to 2019. Co-registering clipped rasters allows for an adaption to the local setting of each landform, preventing larger distortion patterns to imprint on the DEM-based analysis while reducing processing times. Further, it enables temporal and spatial investigation of the x, y, z correction factors used during co-registration.

For the calculation of the LoDs of our vertical surface changes, we extract vertical surface change at 1000 random points distributed outside and in vicinity of each landform polygon and derive their median. Terrain outside the landform polygons is assumed to be predominantly stable. Having controlled vertical surface change in the areas outside the landform surfaces during co-registration, we directly accept the medians as LoDs. For vertical surface change quantification by DEM differencing, we subtract the co-registered newer DEMs from the original older DEMs. Depending on the figure, vertical surface change of the cryospheric landforms in the Rodeo basin is compared as total change (sum of annual median change, Fig. 2A) or as vertical surface change normalized to full years, both calculated as median for the landforms' surfaces, Figs. 2B–D and 5A. Rock glaciers are attributed positive when vertical surface change plus and minus the LoD are above zero, and negative when both are below zero. Further, we derive elevation and slope from the Pléiades DEMs, all at a resolution of 1 m.

We use the projected panchromatic imagery at 0.5 m resolution to conduct feature tracking on all rock glaciers for which we have data for all time periods (47 out of 59) following the approach by Schwalbe and Maas (2017) which matches image patches between orthoimages with two different time stamps by applying cross-correlation for an estimation of a pixel-precise shift, and a least-squares matching for the achievement of sub-pixel accuracies. Originally implemented in the Environmental Motion Tracking (EMT) software, we adapt this approach to semi-automatically process a large quantity of rasters in python (Ebert and Rehn, 2026). For our stereo dataset (2019), we select the first image of the pair with a view angle tilted towards south. For the tristereo datasets (2022–2025), we select the second image of the triplet which is closest to nadir view (90°).

Similar to the DEMs, we clip the panchromatic orthoimages by our bounding boxes prior to feature tracking. After conducting an affine transformation for offset correction, identifiable pixels are tracked on the landform surface (no front and side slopes) using a set of equally, 5 m spaced points – leading to a 5 m spatial resolution of the calculated velocities. This approach benefits from using grey-scaled input which is independent of light conditions (Dall'Asta et al., 2017; Fleischer et al., 2021). To reduce the impact of the polygon boundaries as included in the National Inventory of Glaciers (IANIGLA-CONICET, 2018) on the feature tracking outcome, we apply a 200 m buffer on all polygons to determine the reference area used for aligning the images while we use a buffer of 50 m around the polygons for the actual image tracking. Horizontal rock glacier velocity is presented normalized to full years. To enable the comparison of horizontal velocity values with varying LoDs, we calculate the median exceedance from the corresponding LoD for each rock glacier and time step, Fig. 7A.

Similar to the calculation of the LoDs for vertical surface changes, surface motion is tracked at 1000 randomly distributed points located outside of the landform surface for the calculation of the LoDs of the horizontal rock glacier velocities. We accept the median of the feature tracking results at these 1000 random points as LoD. Selecting the median serves as a precautionary measure against singular of the randomly distributed points coincidentally placed on moving objects.

We track the residuals of the affine transformation during our feature tracking approach to develop a criterion for the quality of the image alignment, which directly affects the feature tracking results. We detect the correlation of these residues in x, y space and use low correlation coefficients (0 to 0.5) as indication of a high-quality feature tracking and high correlation coefficients (0.5 to 1) as indication of a poorer feature tracking quality. The implementation of the correlation coefficient of the residues that arise during feature tracking as a quality control is based on the hypothesis that high correlation is indicative of a technical error, e.g., an image distortion, while true residues are expected to be independently distributed. We include this quality criterium as a metric and as categories.

The area's aridity and extremely limited cloud and vegetation cover yield perfectly suitable conditions for change analysis with remotely sensed optical imagery, such as Pléiades imagery. The summer conditions with very scarce and, if present, non-persistent snow coverage allow for largely uncovered terrain suitable for co-registration, Fig. S1. For co-registration, X correction factors are of least magnitude in median through time, while Z correction factors are of highest, Table 3. A comparison of the T, B, and R tiles indicates no consistent pattern of spatial differences in co-registration factors. All correction factors used during co-registration are independent of the landform type, Fig. S2.

Vertical surface changes as extracted at 1000 randomly distributed points surrounding the landform polygons are independent of the landform type and time period and correspond to the respective LoD of our vertical surface changes, Table 4. They vary between ±0.1 and ±10 cm yr⁻¹ when compared as median for all landforms of each landform type (glaciers, debris-covered glaciers, rock glaciers). Vertical change LoDs are lower than the LoDs calculated for horizontal velocities. Median horizontal velocities at 1000 randomly placed points in vicinity of the rock glacier surfaces range between ±16 and ±61 cm yr⁻¹ and represent our LoDs for horizontal velocities. The calculated LoDs for horizontal velocities are low for the 2019–2022 and 2023–2024 time periods, compared to 2022–2023 and 2024–2025.

Correlation factors of the residues arising during our feature tracking approach range in median over time for the selected rock glaciers between 0.27 (ID44) and 0.64 (ID39), Table 5. This is representative for all polygons, with the least and highest correlation being 0.12 (ID25) and 0.66 (ID47), Table S2. In total, 45 of the 57 polygons are characterized by low correlation coefficients of the residues (0 to 0.5), with 12 polygons exceeding a coefficient of 0.5 but none reaching a coefficient of 0.7. In median and independent of the different rock glacier polygons, the time period 2023–2024 is characterized by least correlation of the residues both for the selected rock glaciers (2023–2024: 0.26) as well as all polygons (2023–2024: 0.28). Thus, three out of the four time periods are characterized by low correlation coefficients with one exceeding a coefficient of 0.5 but remaining below 0.6, both for the selection as well as for all polygons.

Vertical surface change across the cryospheric landforms in the Rodeo basin is highest in magnitude for glaciers, second highest for debris-covered glaciers and least for rock glaciers – independent of the time period, Fig. 2A. Calculated LoDs are very low compared to the vertical surface changes on glaciers, low compared to debris-covered glaciers and vertically dynamic rock glaciers, and high compared to rock glaciers with minimal vertical surface change when calculated as median for the landforms' surface, Table 4.

All glaciers monitored in our study are characterized by negative vertical surface changes, leading to negative total vertical surface change increasing in magnitude, Fig. 2A. Total vertical surface change is of highest magnitude for Agua Negra Glacier (1.09 km², 5012 m a.s.l., 2nd largest glacier in the study area) amounting to a vertical elevation change in median for the glacier surface of -8.99 m for 2019–2025, detected with a LoD of ±0.11 m yr⁻¹. Increase in time-normalized annual vertical surface change is not linear, highest for 2023-20-24 (-1.50 m yr⁻¹, LoD ± 0.01 m yr⁻¹) and lowest for 2022–2023 (-0.20 m yr⁻¹, LoD ± 0.001 m yr⁻¹), Table 4. Time-normalized annual vertical surface changes of Agua Negra Glacier are higher than of the largest glacier monitored with our Pléiades imagery for the full time period (ID71, 1.80 km², 5335.5 m a.s.l.), Table 6. In general, smaller glaciers are characterized by higher time-normalized median vertical surface change than large glaciers, Fig. 2B. Further, vertical surface change is highest in magnitude for glaciers at lower elevations, Fig. 2C, and independent of slope, Fig. 2D.

Only three debris-covered glaciers are present in the Rodeo basin with median vertical surface changes up to -22 cm yr⁻¹, detected with a LoD of ±10 cm, Table 6.

Rock glacier vertical surface change is minimal on Dos Lenguas and El Paso rock glaciers. This is representative for all rock glaciers monitored, Fig. 2A and Table 4. Rock glacier vertical surface changes are not correlated with median elevation or slope, Fig. 2C-D. However, rock glaciers in the Rodeo basin are characterized by an interannual variability of vertical surface change, with 47 rock glaciers alternating at least once between positive and negative median annual surface change between the observed time episodes (6 always negative, 4 always positive).

Spatially, time-normalized vertical surface changes on Agua Negra Glacier are heterogeneously present with highest magnitudes on its west side and towards the glacier tongue, Fig. 3A. Vertical changes on the largest glacier (ID 71) are also heterogeneously present, with highest magnitudes reached in the centre and southern part, Fig. 3B. The glacier located next to El Paso rock glacier exhibits vertical surface changes particularly in its centre, Fig. 3C. Rock glacier vertical surface changes are spatially heterogeneous and often follow a ridge- and furrow morphology with alternating positive and negative areas, Fig. 3D. Rock glacier fronts are characterized by a coherent positive vertical surface change, Fig. 3C–D. For vertical surface changes of all monitored cryospheric landforms, see Fig. 3E.

Rock glaciers in the Rodeo basin exhibit differences in the magnitude of their horizontal velocity in space with median rock glacier surface velocities being heterogeneously present in the Rodeo basin, Fig. 4. Their LoDs are independent of a rock glacier's location, size, and horizontal velocity. Out of the 47 rock glaciers, 1 rock glacier exceeds median velocities of > 1.25 m yr⁻¹ for the time period 2019–2025 (Dos Lenguas rock glacier), while 7 rock glaciers fall between 1 and 1.25 m yr⁻¹ (e.g., El Paso rock glacier), and 14 fall in the classes between 0.75–1 and 0.5–0.75 m yr⁻¹, respectively. The LoD exceedance of median rock glacier surface velocities based on all rock glaciers varies between 0.12 and 0.29 m yr⁻¹, Table 4 (HVe minus respective LoD), non-linear in time.

Correlation coefficients support that larger rock glaciers are faster than those characterized by a smaller surface area, that rock glaciers located at higher location are faster than rock glaciers located at lower elevation, and that smaller rock glaciers are characterized by higher slope than larger rock glaciers. For all correlation coefficients, see Table S3. Larger rock glaciers (> 0.1 km2) in their vast majority exhibit velocities above 0.25 m yr⁻¹ in all time periods, while smaller rock glaciers (< 0.1 km2) are characterized by various horizontal velocities from 0.0 to 1.23 m yr⁻¹, Fig. 5A. Rock glaciers located at elevations below 4100 m a.s.l. do not exceed horizontal velocities of 0.25 m yr⁻¹, while all rock glaciers located above 4300 m a.s.l. exceed horizontal velocities of 0.09 m yr⁻¹ and reach up to 1.38 m yr⁻¹, Fig. 5B. Slope and rock glacier velocity are not correlated, Fig. 5C, and variability of median vertical change is dependent on the time period – with 2022–2023 characterized by more negative and 2023–2024 by more positive median vertical surface changes, Fig. 5D. For a full visualization of the concurrence of elevation and slope separated by time periods and colour-coded for median horizontal velocity and rock glacier surface area, see Figs. S3 and S4.

Rock glacier velocities are spatially heterogeneously present on the various landform surfaces, Fig. 6. El Paso rock glacier (ID 5) is in its upper part characterized by linear ridges in line with the flow direction that exhibit a faster horizontal velocity than the surrounding rock glacier surface, Fig. 6A. Highest velocities of up to 1.09 m yr⁻¹, Table 7, are reached in its lower part. Dos Lenguas rock glacier and rock glacier ID 9 are characterized by extensional flow in the upper area and c-shaped ridge and furrow morphologies oriented perpendicular to the direction of flow, commencing in the centre area, Fig. 6B–C. For both, highest velocities of up to 1.38 m yr⁻¹ (2024–2025, LoD ± 0.64 m yr⁻¹) and up to 1.17 m yr⁻¹ (2024–2025, LoD ± 0.95 m yr⁻¹) are reached in the upper part. The rock glacier with ID 40 is characterized by a lower magnitude of horizontal velocity compared to the other three rock glaciers, Fig. 6D. The different rock glaciers fastest for a specific time period are Dos Lenguas in 2019–2022 and 2024–2025, Table 7, ID 23 in 2022–2023 (1.16 m yr⁻¹, LoD ± 0.87 m yr⁻¹), and ID 40 in 2023–2024, Table 7.

The temporal evolution of rock glacier velocities for the 47 rock glaciers for which we have horizontal velocity data for all time episodes shows unchanged velocities over time, Fig. 7A, independent of rock glacier size and magnitude of velocity. Our quality control for our feature tracking approach, the residue correlation coefficients, indicate higher quality feature tracking for the time periods after 2022 consistent for the three rock glaciers, Fig. 7B–D, representative of all landforms, Fig. S5.

Vertical and horizontal errors of our DGNSS measurements describe their measurement quality. The magnitude of acceptable error depends on the magnitude of the surface changes investigated. Our DGNSS vertical errors at all three sites (El Paso and Dos Lenguas rock glaciers and Agua Negra Glacial Forefield) range from 0.008 to 0.028 m (2022, median 0.019 m), 0.01 to 0.076 m (2023, median 0.021 m) and 0.009 to 0.041 m (2024, median 0.022 m). Horizontal errors are lower than vertical errors and range from 0.005 to 0.017 m (2022, median 0.009 m), 0.006 to 0.036 m (2023, median 0.012 m), and 0.006 to 0.018 m (2024, median 0.011 m). With very few exceptions on Dos Lenguas rock glacier (2023), all horizontal errors are below 0.02 m yr⁻¹ and are comparable between the landforms. For a spatial distribution of the errors, see Fig. S6.

Based on the DGNSS measurements, vertical surface changes and horizontal rock glacier velocities are calculated, serving as independent, in situ based validation dataset for the Pléiades-based results, Tables 8 and 9. While vertical surface changes on the El Paso rock glacier surface are positive in median, they are consistently negative for Dos Lenguas rock glacier. Both rock glaciers are characterized by horizontal velocities of comparable magnitude, particularly considering the effect of LoDs. In all cases, measurements outside the rock glacier surfaces are of much lower magnitude than vertical surface changes and horizontal velocities on the rock glacier surfaces. In Agua Negra Glacial Forefield, the DGNSS measurements indicate vertical and horizontal surface stability with median surface changes close to zero. For a spatial distribution of the DGNSS-based surface changes at all three sites, see Fig. S7.

With the above-presented in situ dataset, we can compare our Pléiades-based surface changes and determine differences between the quantified vertical surface changes and horizontal rock glacier velocities derived using the two datasets, Table 10 and 11. Differences on both rock glacier surfaces and in the Agua Negra Glacial Forefield correspond to less than a pixel. Given our feature tracking approach for quantifying horizontal velocities, only DGNSS points within the landform polygon and the 50 m buffer can be compared.

The offset between the acquired panchromatic Pléiades imagery is highest in z dimension compared to the x and y dimension, Table 3 and Fig. S2. Linear co-registration shifts calculated over stable terrain excluding the landform surfaces do not exceed ±0.20 m (x) and ±0.59 m (y), Table 3. They indicate relatively good alignment between the panchromatic Pléiades acquisitions in x and y dimension even prior to our co-registration. Co-registration for the z dimension is of higher importance with correction factors up to -1.55 m (2024–2025, tile B) and 2.62 m (2019–2022, tile R), Table 3, respectively. While our shift in z direction is high compared to the x and y component, it is low compared to Rieg et al. (2018) (offsets x, y, z of 4, -3.2, and 4.8 m, Pléiades-based DEMs in ERDAS IMAGINE) and Beraud et al. (2023) (offsets x, y, z of up to -9.89, 7.79, and 12.40 m, Pléiades-based DEMs in ASP). As expected, our linear shifts are independent of the targeted location within the catchment, of time, and of the landforms type – supporting the technical rather than geomorphological nature of the need for co-registration. We do not see a difference in the magnitude of the correction factors needed to co-register the stereo dataset (2019), compared to the tri-stereo datasets (2022–2025). Based on the similarity of the 2019–2022 time period to the other time periods, cf. Figs. 2 and 7, we conclude the stereo acquisition to be suitable for vertical and horizontal rock glacier surface change monitoring in the Dry Andes. In comparison, Berthier et al. (2014) conclude a moderate effect of the tristereo benefit, including the reduction of the percentage of data voids.

Artefacts or missing values in our DEMs occur in areas with steep slope. These artefacts are not related to our DEM generation but to the Pléiades data-take which includes no data values for these steep slopes due to mismatches during image correlation. In the Rodeo basin and with our processing approach, we do not see generated spikes in the DEMs as described by Ruiz and Bodin (2015). The number of artefacts in our DEMs is very small. This is mainly due to the vegetation-freeness of the study area as well as the semi-global matching strategy which tend to smooth the surface of the reconstructed DEMs. With a DEM resolution of 1 m, the DEMs used in this study are of higher resolution than used in other studies (Beraud et al., 2023; Falaschi et al., 2023, 2025; Pitte et al., 2022).

Related to our DEM generation workflow, vertical surface change cannot be monitored on all glaciers. This is due to limited heterogeneity of glacier surface characteristics leading to gaps in the DEMs as the photogrammetric processing fails to generate elevation information for these areas, cf. 4.3, Fig. 3E. According to Berthier et al. (2014), the Pléiades imagery's radiometric range (12 bit) limits these effects compared to SPOT1-5 and ASTER. The effect is, however, also encountered by other studies (Beraud et al., 2023; Falaschi et al., 2023).

LoDs calculated as median vertical surface change at 1000 random points in the vicinity of the landforms lead to low LoDs in comparison to the magnitude of vertical surface changes, particularly for glaciers, cf. Table 4. Investigations of vertical surface change of rock glaciers are scarce (Vivero and Lambiel, 2024), particularly for studies using Pléiades imagery (Falaschi et al., 2025). Thus, a comparison of our LoD calculation with other studies is extremely limited. For rock glacier velocities, all horizontal velocities quantified exceed our LoDs, Figs. 6 and 7. Comparatively low LoDs highlight the stability of surfaces assumed to be stable and support the suitability of panchromatic Pléiades imagery for feature tracking.

Residue correlation coefficients, here used as quality control of the feature tracking approach, cf. 3.5, are in general below 0.5 in median – independent of the respective rock glacier polygon or time period. Their variability between the time periods is higher than between the landforms, highlighting the technical nature of their correlation, cf. Tables 5 and S2. The analysis of residue correlation characterises the time period 2023–2024 as particularly high in quality, not corresponding to the Pléiades tiles with the least co-registration needed, cf. Table 3. Though not fully portrayed in the residue correlation coefficient, we experience the effect of image distortion to be higher on rock glaciers located on steep slopes compared to locations closer to valley bottoms. This corresponds often with smaller landforms, as larger landforms with their elongated tongues “flatten” the topography by building up bodies of rock and ice, cf. Fig. 7D.

While the use of the National Inventory of Glaciers for the landform boundaries allows upscaling our analysis, it restricts the landform surface to a static measure, rather than the surface being adapted for each year. This introduces error on the calculation of median vertical surface changes and horizontal velocities and prevents the potential detection of new landforms. We reduce the impact of the inventory by calculating vertical surface change for bounding boxes with 500 m distance to the landform polygon and with buffered polygon outlines during feature tracking. We highlight the need to display the complex and variable spatial distribution of surface changes within the landform, cf. Figs. 3 and 6, next to the calculated statistics and further agree with Ferri et al. (2020) that highlight the effect of the choice of an inventory, e.g., a national versus a global one.

Difference of our Pléiades-based vertical change to the vertical change measured with the DGNSS equipment is of lowest magnitude for the stable forefield of Agua Negra Glacier, cf. 4.2 and Fig. 3A. Higher difference correlates with a higher magnitude of vertical surface change dynamics at Dos Lenguas and El Paso rock glaciers, cf. 4.2 and Fig. 3C–D. However, difference between the DGNSS measurements and the Pléiades-based vertical surface changes is of lower magnitude for the rock glacier surface compared to the DGNSS measurements located off-site the rock glacier surfaces, cf. 4.2. We hypothesize this difference to be caused by higher vertical accuracies of our DGNSS measurements located at the rock glacier surfaces compared to the surrounding terrain, see Fig. S6 – paired with a very dynamic off-surface environment for El Paso rock glacier, cf. Fig. 3C. Further, we hypothesize the higher accuracies of the DGNSS measurements on the rock glacier surfaces compared to their surroundings to stem from the elevated location on the surface allowing for a good connection between the DGNSS base and rover, compared to the off-surface DGNSS measurement locations partially obscured by the rock glacier body. Similar to the comparison of DGNSS- and Pléiades-derived vertical surface changes, differences between the DGNSS- and Pléiades-derived horizontal velocities are smallest for the Agua Negra Glacier forefield, cf. 4.2, highest for El Paso rock glacier (-0.35 m, 2023–2024) and low in general (all < 1 pixel). The present spatial pattern of differences in magnitude between DGNSS measurements, Fig. S7, is equivalent to our Pléiades-based approach, supporting the feasibility of the method.

Our Pléiades-based vertical surface changes reveal glaciers at higher elevation and of larger size to be less prone to surface lowering, cf. Fig. 2B–C. This is in agreement with the temperature gradient and Al-Yaari et al. (2023) that find small glaciers to be affected by more pronounced loss – supporting the suitability of our approach. We refrain from calculating glacier mass balances given our limited in situ knowledge on glacier ice density. Before comparing our results to the literature, we highlight the effect of glacier surface delineation (cf. 5.1), as also discussed by, e.g., Ferri et al. (2020). The glacier we refer to with ID 71 as mapped by IANIGLA-CONICET (2018), cf. Fig. 3B, is split in two WGMS glacier IDs (32927, 32920). Based on the geodetic method, its mean annual elevation change is attributed values ranging from -0.2 m yr⁻¹ (2000–2012) to -0.5 (2009–2014) for the first WGMS ID and predominantly around -0.4 m yr⁻¹ for different measurement periods between 1999 and 2019 for the second (Dussaillant et al., 2019; Braun et al., 2019; Hugonnet et al., 2021) – contrasting with our Pléiades-based quantification of surface changes, Table 6, due to these differences in delineation.

Pitte et al. (2022) calculate mass balances ranging between -0.79 m w.e. (2014–2025, cumulative for entire glaciological year) and -3.67 m w.e. (2020–2021, see before) for Agua Negra Glacier based on the glaciological method (Cogley et al., 2011). Their annually repeated measurements are all negative and increase in time with one exception (2016–2017), indicating accelerated downwasting in agreement with literature (Dussaillant et al., 2019; Masiokas et al., 2020; Ferri et al., 2020). Based on DEM differencing using, among others, Pléiades-based DEMs, Pitte et al. (2022) report a generalized thinning with high magnitudes at lower elevations which we can attest with our data, cf. Fig. 3A. Pitte et al. (2022) present vertical surface changes of ca. 1 m yr⁻¹ between 2013–2019 in the Agua Negra centre and upper parts and changes of ca. 2 m yr⁻¹ between 2013 and 2019 in the lower and western parts. This corresponds in magnitude with our vertical change results for Agua Negra Glacier as well as in pattern, Table 6 and Fig. 3A. Pitte et al. (2022) further report a 23 % reduction of the Agua Negra surface area between 1959 to 2019. While not part of this study we highlight the suitability of panchromatic and even multispectral Pléiades imagery to continue the assessment of surface area changes using our Pléiades acquisitions, cf. Table 1. Ayala et al. (2025) estimate a 35 % area loss of the debris-free area of Tapado glacier between 1956 and 2024 located in the neighbouring catchment in Chile – elucidating a similarity of glacier response on both sides of the Andes. In contrast, they observe an increase of the debris-covered area for Tapado glacier which reduces comparability to Agua Negra glacier that does not have a debris-covered part.

As expected, given their higher ice content compared to rock glaciers, vertical surface change is higher on the debris-covered glaciers in the Rodeo basin compared to the rock glaciers, Fig. 2A. This is in agreement with Ferri et al. (2020) that find mass balances of higher magnitude for debris-covered glaciers compared to rock glaciers based on the ASTERIX method (Dussaillant et al., 2019) in the Central Andes (30 to 37°). For this study, however, we highlight that the three debris-covered glaciers are too little in number to draw representative conclusions upon their vertical surface change behaviour – particularly on detailed ablation patterns or the development of supraglacial ponds or ice cliffs as conducted, e.g., in Ayala et al. (2025) and Falaschi et al. (2021). Ayala et al. (2016) find similar streamflow contributions of debris-covered glaciers compared to glaciers, highlighting their hydrological significance. This significance contrasts with a strong underrepresentation of studies on debris-covered glaciers in glaciological studies (Masiokas et al., 2020).

Ferri et al. (2020) detect for the Central Andes (30 to 37°) highest mass balance losses for partly debris-covered glaciers, followed by clean ice glaciers and contrasting with completely debris-covered glaciers and rock glaciers characterized by almost zero mass balances (e.g., rock glaciers: -0.02 ± 0.19 m w.e. yr⁻¹, 2000–2018). Aside the partly debris-covered glaciers which we do not address here, this is well reflected in our results, cf. Fig. 2A. It highlights the strong differences in vertical surface change behaviour between the landforms. The question here is the comparability between the meaning of vertical surface changes on glaciers and rock glaciers, which is why we focus on rock glacier horizontal velocities as indicator of the (in)stability of permafrost conditions in the next chapter.

For Dos Lenguas rock glacier, our Pléiades-based rock glacier vertical surface changes, Table 6, and horizontal velocities, Table 7, are in good agreement with UAV-based rock glacier vertical surface changes and horizontal velocities for Dos Lenguas rock glacier available for 2016-2018 (Halla et al., 2021) and 2016–2024 (Stammler et al., 2025a), both in terms of magnitude, Fig. 8, and pattern. In terms of pattern, highest values of 1.5 to 2 m yr⁻¹ are reached in the upper zone, both in Fig. 6B as well as the literature (Halla et al., 2021; Stammler et al., 2025a). Strozzi et al. (2020) quantify horizontal surface velocities ranging between 1.5 and 2 m yr⁻¹ (2015–2020) for the upper part of Dos Lenguas rock glacier, based on interferometric synthetic aperture radar (InSAR) and offset tracking. These projected values are in good agreement with our Pléiades-based horizontal velocities, Fig. 8. While we acknowledge the challenges surrounding the comparison of optical and radar imagery-based rock glacier surface change quantification (view angles, different sources of error, different time periods, etc.) and caution a detailed comparison other than the comparison of general magnitude, we highlight the inter-methodological agreement of magnitude between the UAV-, Pléiades-, and InSAR-based investigations.

Except for Dos Lenguas rock glacier, no other rock glaciers in Rodeo basin have been monitored for surface change. This highlights the great benefit of the satelliteborne analysis and its scalability compared to, e.g., UAV-based rock glacier surface change monitoring, as the number of rock glaciers investigated is unprecedented and such comparisons would else not be possible. Median horizonal surface changes for all rock glaciers investigated in this study, Tables 4 and 6, are in agreement in terms of magnitude with average velocities of 0.54 ± 0.03 m yr⁻¹ for rock glaciers in the neighbouring La Laguna catchment (Chile) as detected by Robson et al. (2022). We do not detect a regional trend in increasing rock glacier velocities in the Rodeo basin between 2019 and 2025, Fig. 7. All horizontal velocities quantified with our method remain unchanged within the time period monitored. This absence of such a trend is independent of a rock glaciers' size, median horizontal velocity, elevation, or slope. Apart from the median horizontal velocity exceedance from LoD which facilitates a comparison among the different rock glaciers, Fig. 7B–D as well as Fig. S5 portray the unprecedented strength of the dataset including the median horizontal velocity for each of the rock glaciers for each time period together with the respective LoD and our quality control – the residues' correlation coefficient. A plausible explanation for the absence of a basin-wide acceleration is that, in this semi-arid Andean setting, persistently low precipitation may limit both seasonal snow insulation and liquid-water input into the active layer, thereby damping interannual variability in ground thermal conditions and hydro-mechanical softening that can otherwise promote speed-ups in rock glacier creep (Cicoira et al., 2019). This interpretation is consistent with evidence that aridity strongly constrains permafrost thermal regimes in the Dry Andes (Koenig et al., 2025) and with recent findings that precipitation scarcity can contribute to comparatively stable rock glacier behaviour (Stammler et al., 2025a). Under this hypothesis, the slightly higher velocities for large rock glaciers in 2023–2024 would reflect short-lived departures from typical moisture limitation (e.g., an anomalously wet season) similar to the variability described in Halla et al. (2021), a possibility that requires confirmation using local precipitation and/or snow proxies and longer kinematic time series.

The lack of a regional trend in increasing velocities determined in this study elucidates stable permafrost conditions in Rodeo basin during 2019–2025. While we acknowledge the limitations of our short monitoring period, we highlight the data scarcity in this region of the world – particularly also for rock glacier monitoring (Hu et al., 2025). The detected lack of a regional trend is in agreement with a longer-term monitoring (1968–2023) by Blöthe et al. (2024) who identify unchanged rock glacier velocities in the Valles Calchaguíes region (24–25° S, Argentinean Andes) and Falaschi et al. (2025) who report a mixed signal of acceleration and deceleration of rock glacier velocities in Central Patagonia (47° S, Argentinean Andes) between 2018 and 2023. It contrasts with findings in the Alps (Kellerer-Pirklbauer et al., 2024; Manchado et al., 2024; Marcer et al., 2021) and North America (Kääb and Røste, 2024). Based on borehole measurements, Koenig et al. (2025) find no clear warming indication for ground temperatures in the Andes (27–34°) – further supporting the implication on permafrost conditions of our quantification of rock glacier velocities. We highlight the strong need for continued monitoring of rock glaciers in this basin in this potentially dynamic point in time.