The oldest available data set for this region are the Trimetrogon aerial photographs acquired by the U.S. Army Air Force (USAF) in the austral summer of 1944/45 (Figures 2a,b). The aerial survey, conducted in January and February of 1945, represents the first comprehensive airborne coverage of the CDI and Cloue Icefield and provides a valuable basis for assessing glacial and geomorphologic change over the past eight decades (Lliboutry, 1998). The survey was strategically carried out in mid-summer to ensure optimal conditions for snow and ice melt, which in most areas allows glacial lakes to be clearly distinguished from ice and land areas. However, the spatial resolution of the USAF Trimetrogon photographs is limited by the variable image quality, incomplete coverage of some areas, the flight line spacing and the oblique view of the left- and right-looking acquisitions, even resulting in gaps in some places that make it difficult to distinguish these features. In addition, the 1:250,000 scale preliminary maps produced by the Instituto Geográfico Militar (IGM) of Chile based on these photographs often lack precise delineations of glacial and hydrological features, further complicating their interpretation (Lopez et al., 2010).

The original USAF Trimetrogon photographs, from 1944/45, served as the basis for the sketch maps published by Louis Lliboutry in his seminal work “Nieves y Glaciares de Chile” (Lliboutry, 1956). One of these maps covers the entire CDI and its surroundings and provides a detailed representation of the extent of the glaciers and the location of the glacial lakes as they existed in 1945, providing an important historical reference for the study of glacier dynamics in the region. For this purpose, Lliboutry’s sketch map was georeferenced with 132 off-glacier tie-points manually identified on high-resolution “ESRI Imagery” available via “QuickMapServices” in the QGIS software (v3.40.5-Bratislava). In the case of the two icefields and considering that the Cloue Icefield was not included in Lliboutry’s sketch map, the preliminary location and extent of the glacial lakes identified with the sketch map were manually improved by (1) using the individual USAF Trimetrogon photographs (Table 1) and (2) using the previously mapped geomorphologic landforms (former lake shorelines from Izagirre et al., 2024).

It should be noted that Lliboutry’s work remains a cornerstone in the glaciological literature of the Southern Andes, as it not only documents the state of the CDI and surrounding mountain ranges in the mid 20th century, but also establishes a methodological framework for the integration of aerial photography into glaciological studies. Despite its historical significance, the map’s utility is reduced by the limitations of the original photographic data. Nevertheless, these early datasets are invaluable for reconstructing the long-term evolution of glaciers and glacial lakes, especially when combined with modern remote sensing technologies and field observations (Rivera et al., 1997; Harrison et al., 2007; Wilson et al., 2016; Piret et al., 2021; Gorsic et al., 2025).

To create the glacial lake inventories for the observation years 1986, 2006, 2019 and 2024, we used a combination of Landsat-5, Sentinel-2 and PlanetScope satellite imagery (Table 1; Supplementary Table S1). All Landsat-5 MSS and TM scenes with a multispectral spatial resolution of 30 m were orthorectified by the United States Geological Survey (USGS) and obtained as Level 1 Products through the USGS Earth Explorer interface (https://earthexplorer.usgs.gov/).

For the 2019 inventory, we analysed nine Sentinel-2 MSI scenes with a multispectral spatial resolution of 10 m. The Sentinel-2 MSI scenes were orthorectified by the European Space Agency (ESA) and downloaded as Level-1C Products from the Copernicus Browser interface (https://browser.dataspace.copernicus.eu/). The 2024 inventory was derived from five Sentinel-2 MSI scenes, complemented by 11 high-resolution PlanetScope scenes (∼3 m multispectral spatial resolution) from 4 March 2024 (Planet Team, 2024).

All satellite scenes were carefully selected based on minimal cloud cover and a late summer acquisition date to ensure optimal conditions for distinguishing surface boundaries such as ice/snow, glacier/rock, glacier/lake and lake/floating ice. However, due to challenges such as snow cover, cloud obstruction and mountain shadows, in some target years (e.g., 1986, 2006 and 2019) scenes from a ±1 year period had to be included to ensure adequate image availability and quality. This approach allowed us to maintain consistency in the identification and mapping of glacial lakes across the study period.

The first step in establishing our glacial lake inventory was to define which lakes should be included. Given that the study region was extensively covered by the Patagonian Ice Sheet during the last glacial cycle, many lakes in the area could be classified as glacial in origin (Davies et al., 2020). However, to focus on lakes most relevant to modern glacier dynamics, we restricted our analysis to the present-day ice-contact lakes of the Cordillera Darwin and Cloue Icefields and their surroundings, as well as the lakes that formed after the end of the LIA and are still in contact with or separated from the present-day glacier margins. Lakes in direct contact with glaciers are particularly dynamic and have likely undergone the greatest changes since the end of the LIA (Loriaux and Casassa, 2013; Izagirre et al., 2024). These lakes are also more likely to experience significant changes in size, volume and stability due to their interaction with glacial and paraglacial processes (Harrison et al., 2018; Wilson et al., 2018). To identify these lakes, we compared LIA glacier extents mapped by Davies and Glasser (2012), Meier et al. (2018) and Carrivick et al. (2024) with historical USAF Trimetrogon photographs and recent glacial landform mappings by Izagirre et al. (2024).

We classified glacial lakes into three main types based on their formation mechanisms: (1) moraine-dammed lakes, which form adjacent to glacier termini and are impounded by frontal or terminal moraines; (2) ice-dammed lakes, which are impounded by glacier ice, often in tributary valleys or marginal depressions on the side of the glacier; and (3) bedrock-dammed lakes, which occupy overdeepenings and depressions in bedrock. These classifications help to understand the geomorphic processes driving lake formation and their potential for future changes. In addition, we categorised the lakes according to their area size to facilitate the analysis of size-dependent behaviours. Small lakes are defined as lakes with an area of less than 0.5 km² and large lakes with an area of 0.5 km² or more. This classification corresponds to the thresholds used in comparable studies in Patagonia (e.g., Loriaux and Casassa, 2013) and effectively captures the different behaviours observed in the study region.

Glacial lakes were delineated using a combination of semi-automatic and manual mapping techniques. The semi-automatic approach employed the normalized difference water index (NDWI) and a ratio-based method (R) proposed by Gardelle et al. (2011), which distinguishes water from non-water surfaces using the formula: R = B G r e e n B N I R where B _Green and B _NIR represent the green and near-infrared spectral band values, respectively. While NDWI is widely used for water detection, it can misclassify shadowed areas as water, particularly along lake margins (Huggel et al., 2002). The ratio-based method (R) mitigates this issue but can struggle to differentiate between ice and water at calving fronts (Loriaux and Casassa, 2013). To address these challenges, we incorporated a terrain slope threshold derived from a 30-meter resolution digital elevation model (DEM) from the NASA Shuttle Radar Topography Mission (SRTM GL1) Global 1 arcsec degree of the year 2000 from Open Topography repository (NASA Shuttle Radar Topography Mission, 2013). Lakes were assumed to occupy areas with surface slopes below 10% (5.7°), accounting for DEM uncertainties and resolution limitations (Gardelle et al., 2011).

Manual refinement of the semi-automated classifications was essential to ensure accuracy, particularly in complex areas such as shadowed margins, calving fronts and regions with mixed ice-water boundaries. This labour-intensive process involved visually inspecting and correcting lake boundaries using false-colour composites of satellite imagery (Figures 2c–f; Supplementary Table S1). This step also ensured that turbid lakes with high suspended sediment loads—which reduce spectral contrast and can be misclassified using ratio-based methods—were correctly identified and included in the inventory. Manual refinement significantly improved the quality of the lake polygons, ensuring consistency and high accuracy in the final inventory (Wilson et al., 2018).

Mapping errors and uncertainties were inherent in our approach, particularly for earlier observation years (1945, 1986 and 2006) due to the coarser spatial resolution of the available image data (e.g., Landsat-5 MSS and TM in 1986 and 2006) and the exclusively manual delineation in 1945. In this respect, small lakes were probably underrepresented in the early inventories (1945 and 1986) as the coarser resolution of the images possibly increased the apparent growth rates. Errors were also influenced by factors such as cloud cover, snow cover and mountain shadows, which can obscure lake boundaries and thus lead to minor biases (Gardelle et al., 2011; Loriaux and Casassa, 2013; Wilson et al., 2018). To minimise these uncertainties, we used high-resolution PlanetScope imagery (3 m) for the 2024 inventory, which provides a more accurate reference for validating previous datasets. Despite these efforts, uncertainties in lake area calculations vary considerably across time periods due to differences in image resolution and validation sources. Earlier estimates (e.g., ± 20% in 1945 and ±15% in 1986) reflect the limitations of historical images with lower resolution, while modern measurements achieve much higher precision thanks to higher resolution data (e.g., ± 5% in 2019 with Sentinel-2 and ±3% in 2024 with PlanetScope). To account for these variations, we include a mapping error of at least 1 pixel along all lake perimeters.

In this study, we hypothesised that changes in glacial lakes between 1945 and 2024 could reveal previously undocumented drainage events in the Cordillera Darwin and Cloue Icefields. We classify these events into two categories based on hydrological and terrestrial geomorphological evidence: (1) Outflows, defined as usually rapid discharges of water from ice-dammed lakes that often leave stranded icebergs but have limited distal impact due to buffering by downstream water bodies (lakes or sea) and may occur in cyclic events; (2) Outbursts (or GLOFs), characterised by single or cascading, sudden high-magnitude water releases that produce clear geomorphological signatures beyond the lake basin, including breached moraines and downstream sediment deposits.

Cyclic events such as those at Lago Mateo Martinic are poorly documented in existing studies (Iturraspe, 2011; Iribarren Anacona et al., 2015b; Izagirre et al., 2024). To fill this gap, we combined remote sensing analyses with field observations conducted between 2016 and 2024 and focused on areas with known or suspected outflow or GLOF activity. These included the inner part of Fiordo Fouque (Cloue Icefield) and the Roncagli/Alemania glacier system (CDI).

Field data collection included site reconnaissance, photo-documentation and validation of remote sensing interpretations to constrain the timing and magnitude of GLOF events. This process included an extension of satellite image coverage to identify annual variations (Supplementary Table S1) and detailed geomorphologic mapping of diagnostic GLOF features, such as breached moraines, flood channels and sedimentary deposits, as described by Izagirre et al. (2024). In addition, we collected first-hand accounts from local sailors and residents of Puerto Williams that gave us valuable insights into the dynamics of GLOFs, particularly the events of 1997/98 and 2018 in the inner part of the Fiordo Fouque. These interviews helped reconstruct the sequence of events and assess the geomorphological changes associated with GLOFs.

In March 2016, we visited the area of Lago Covadonga in the inner part of Fiordo Fouque to document the landscape changes caused by the GLOF event in 1997/98. Using a Nikon Coolpix AW100 (16 Mp), we acquired 356 overlapping photos along survey circuits around the terminal moraine and the breached section. These images were processed using structure-from-motion (SfM) techniques in Agisoft Metashape Professional (v2.2.0) to create an initial 3D point cloud model. As the camera’s geotags provided only limited positional accuracy, the 3D model was subsequently georeferenced using 50 virtual ground control points (GCPs) from high-resolution satellite imagery, supplemented by handheld Garmin etrex30 GPS waypoints (±5 m elevation precision) collected along the lake shoreline and moraine ridge. From the georeferenced 3D point cloud, we derived a DEM that formed the basis for estimating lake level and volume change. While no independent GCP set was available for formal accuracy validation, the use of a large number of virtual GCPs, in combination with on-sited GPS waypoints and cross-comparisons of the lake shoreline with foliage patterns clearly visible in 2016, helped to minimise horizontal and vertical uncertainties. Using this reconstruction, we estimated the lowering of the lake level and the area change after the GLOF event and finally calculated the released water volume using a first-order approximation based on the mean area multiplied by the elevation change.

In order to document the geomorphological changes caused by the 2018 GLOF event in the inner part of Fiordo Fouque, we carried out an unmanned aerial vehicle (UAV) survey with a DJI Mavic 2 Pro in August 2024. The high-resolution aerial images were processed with SfM in Agisoft Metashape to create an initial 3D model covering an area of 1.33 km². Since there were no differential GNSS camera positions during our survey, we took special measures to correct lens distortions and minimise model warping. We created 11 synthetic GCPs along the shoreline of both the current and former areas of the lower lake (Lake 2). These synthetic GCPs—defined as points with x-y-z coordinates extracted from the initial 3D model—helped to compensate for elevation differences of several metres observed along the shoreline due to geolocation inaccuracies and imperfect lens distortion corrections. In the final 3D model, we constrained these GCPs to zero elevation (all are shoreline points) and set their positional uncertainties to 15 m (x, y) and 0.1 m (z) to prevent error propagation into the lens distortion parameters. The conservative horizontal value reflects the maximum expected drift of the UAV’s non-differential GNSS receiver, while the vertical threshold accounts for both pixel-level selection errors (≤0.1 m at 0.05 m resolution) and our use of instantaneous sea level as a zero-elevation reference.

This approach successfully generated flat water surfaces and reduced shoreline elevation differences to centimetre accuracy, with lens distortion residuals equal to or less than 1 pixel. Using the optimised model, we generated an orthomosaic with a resolution of 0.05 m and a DEM with a resolution of 0.20 m, allowing accurate quantification of geomorphological changes and key parameters, such as breach dimensions and the released water volume. For the comparison of pre- and post-GLOF, we manually digitised the lake boundaries using the UAV-generated orthomosaic (0.05 m, 15 August 2024) and WorldView-2 satellite imagery (0.50 m, 21 November 2013), identifying former shorelines based on vegetation patterns and bedrock colour changes. Elevation differences were calculated using the Copernicus DEM GLO-30 as a baseline, with estimates of released water volume derived from the mean area multiplied by the elevation change as a first-order approximation.

The inventory of glacial lakes in 2024 for the Cordillera Darwin and Cloue Icefields and their surrounding regions shows a total of 185 glacial lakes with a total area of 63.3 ± 1.9 km² and a mean lake area of 0.34 km² (Figure 1b; Table 2). The majority (81.5%) of the lakes mapped in 2024 are small lakes (<0.5 km²; 151 lakes totalling an area of 16.5 km²), but large lakes (≥0.5 km²; 34 lakes totalling an area of 46.8 km²) contribute disproportionately to the total area (74%).

Accounting for 75.5% (139 lakes) of the total number and 80.5% (50.9 km²) of the total area, moraine-dammed lakes dominate the inventory. These lakes are concentrated in regions where glaciers retreated significantly after the LIA, particularly in the CDI, which contains 83 lakes (37.4 km²), many of which formed and expanded in overdeepened basins exposed by retreating outlet glaciers. The largest are the proglacial lakes of Stoppani and Pigafetta, which have an area of 3.6 km² and 2.8 km², respectively. Ice-dammed lakes are less numerous (9.8% of the total number, 18 lakes) but still account for 14.8% (9.4 km²) of the total area, especially in zones with active glacier dynamics such as the Roncagli/Alemania glacier in the CDI, where the largest ice-dammed lake (Lago Mateo Martinic) has an area of 6.5 km². Bedrock-dammed lakes are less frequent (14.7% of the total number, 27 lakes) and account for 4.7% (3.0 km²) of the total area.

In spatial distribution, the CDI remains the epicentre of lake development. It is home to the largest lakes and accounts for 59.2% (37.4 km²) of the total glacial lake area (Figure 1c). The Cloue Icefield has 15 lakes (4.7 km²) and a mean area of 0.31 km² (Figure 1d). In contrast, the Monte Sarmiento massif (6.8 km²) and the Cordón Navarro range (4.2 km²) are home to fewer but larger lakes with a mean area of 0.85 km² (8 lakes) and 0.42 km² (10 lakes), respectively. Peripheral lakes, located in recently deglacierised adjacent areas, are hotspots for small lake formation, with 68 lakes (10.1 km²), 85% of which are smaller than 0.5 km² and have a mean lake area of 0.15 km² (Figure 1c).

Between 1945 and 2024, the number and area of glacial lakes in the study region increased considerably. The total number of lakes increased by 461% from 33 to 185, while the total area of glacial lakes increased by 124% from 28.2 ± 5.6 km² to 63.3 ± 1.9 km² (Table 3; Supplementary Table S2). In the case of the large lakes, they increased by 162% in number (from 13 to 34), while their area increased by 87% from 25.0 ± 5.0 km² to 46.8 ± 1.4 km². This growth was not linear; there were various phases of lake area acceleration and deceleration. The fastest increase in lake number occurred between 1945 and 1986 (136% growth), followed by sustained growth until 2006 (65%) and 2019 (43%), which stabilized by 2024 (Figure 3a). In contrast, the largest expansion of total lake area occurred between 1986 and 2006 (54.6% growth, annual rate of 2.7% yr⁻¹), followed by the 2019 to 2024 period (9.8% growth, annual rate of 2.0% yr⁻¹).

The dominance of ice-dammed lakes in 1945, which accounted for 71.6% (20.2 km²) of the total area, changed drastically during the study period (Figure 3b). In 1986, moraine-dammed lakes were already the predominant lake type accounting for 62.4% (19.1 km²) of the total area, compared to 26.9% (7.6 km²) in 1945. Ice-dammed lakes, on the other hand, declined sharply between 1945 and 1986, both in terms of area (from 71.6% to 35.6%) and number (from 36.4% to 3.8%). From 1986 to 2024, the difference between ice-dammed and moraine-dammed lakes widened, with the latter accounting for 80.5% (50.9 km²) of the total area in 2024, while the mean area of ice-dammed lakes decreased significantly, from 1.68 km² in 1945 to 0.52 km² in 2024. This transition again reflects the role of glacier retreat after the LIA, which exposed moraines that dammed up meltwater, while glacier thinning destabilized the ice dams. The proportion of bedrock-dammed lakes, while still small, steadily increased, tripling in both number (from 1 to 27) and area (from 0.4 km² to 3.0 km²), indicating a gradual exposure of stable overdeepenings and depressions.

The mean lake area decreased from 0.86 km² in 1945 to 0.39 km² in 1986 and 0.34 km² in 2024 (Figure 3a), mainly due to the disappearance of large ice-dammed lakes and the proliferation of small lakes, particularly in the adjacent areas of the Cordillera Darwin and Cloue Icefields. These peripheral lakes saw the largest relative increase in lake numbers, from 6 in 1945 to 49 in 2006 and 68 in 2024, but remained small after 1986, when the mean lake area decreased from 0.19 km² to 0.15 km², indicative of widespread newly emerging glacial lakes (Figure 3c).

The total area of glacial lakes in the CDI fluctuated from 1945 to 2006, decreasing by 17% from 1945 to 1986 and then increasing by 39% during the next observation period (1986–2006). Meanwhile, the emergence rate of new lakes increased from 0.3 yr⁻¹ between 1945 and 1986 to 1.2 yr⁻¹ between 1986 and 2006. From 2006 to 2024, the total area and number of glacial lakes increased further (1.2 yr⁻¹ in number and 39% in area). These observed increases in number from 1945 to 2024 occurred despite a 64% reduction in the mean lake size (from 1.24 km² to 0.45 km²), mainly due to the nearly complete disappearance of the large ice-dammed lakes observed in 1945. The total lake area of the CDI increased by 59% (from 23.5 km² in 1945 to 37.4 km² in 2024), reflecting the formation of smaller lakes in newly deglacierised terrain and small ice-dammed lakes at higher elevations after 2019, together with large moraine-dammed lakes that expanded at an accelerated rate after 2006 (Figure 3c). This applies to the proglacial lakes of the Videla, Pigafetta, Roncagli/Alemania and Stoppani glaciers.

The Cloue Icefield experienced a 15-fold increase in lake area (from 0.3 km² in 1945 to 4.7 km² in 2024), and the mean lake area also increased from 0.13 km² to 0.31 km², as the land-terminating glaciers transitioned into freshwater calving systems, which expanded considerably after 1986 (Figure 3c). The Monte Sarmiento massif experienced the fastest growth, with a 16-fold increase in lake area (from 0.4 km² in 1945 to 6.8 km² in 2024). Meanwhile, its mean lake area increased from 0.13 km² to 0.85 km² (Figure 3c), due to the expansion of large moraine-dammed lakes such as Schiaparelli, Emma, Conway and Lovisato. In contrast, the Cordón Navarro range showed a gradual but constant development of glacial lakes between 1945 and 2024. The number of lakes increased from 3 to 10, while the total lake area increased by 252% (from 1.2 km² to 4.2 km²). The mean lake area remained relatively stable and varied between 0.37 km² and 0.45 km² (Figure 3c), indicating that the growth of the lakes was due to the formation of new basins of similar size rather than the enlargement of existing ones.

Our analysis of multi-temporal aerial and satellite imagery, combined with field observations and local accounts, identified several previously undocumented outflow events and various GLOFs in the Cordillera Darwin and Cloue Icefields between 1986 and 2024 (Table 4). Some of these events, identified by lake extent changes in association with distinct geomorphic features such as breached moraines and extensive large-size (>1 m) boulder debris, represent the southernmost modern-day GLOFs recorded in South America.

The Roncagli/Alemania glacier area in the southeastern part of the CDI provides multiple examples of ice-dammed lake drainage (Figure 2). Here, one of the earliest documented outflows occurred between 14 May 1985 and 26 February 1986 at Lago Taranaki (Supplementary Figure S1) — a large ice-dammed lake named by a New Zealander mountaineering expedition during the austral summer of 1970/71 (Banks and Radcliffe, 1972). Due to the limited availability of satellite imagery prior to this drainage event, earlier events are difficult to identify. However, high-resolution aerial photographs from the austral summer of 1984 (acquired by the Servicio Aerofotogramétrico de Chile, SAF) and a low-resolution satellite image from 1985 show the lake at full capacity. In contrast, the satellite image from 1986 shows a completely drained lake that did not refill in the following years. Therefore, the former area of Lago Taranaki is now a side valley with rich glacial geomorphological features such as several former lake shorelines and an extensive glaciolacustrine deposit (Izagirre et al., 2024).

Upstream of the Roncagli/Alemania glacier, the ice-dammed Lago Mateo Martinic (6.5 km² in 2024) showed cyclic partial outflows between 1997 and 2024. Annual satellite observations revealed that the lake area repeatedly decreased by 40%–50% (from ∼10.5 km² to ∼4.0 km²) after each ice dam failure. At the same time, the lake required a shorter period of time in recent years to refill to its maximum storage capacity and outflow again (Table 4). Since 2010, the northern tributary has been separated from the main lake basin. A smaller ice-dammed lake, Lago marginal Este (0.4 km² in 2024), located on the opposite edge of the glacier, also showed repeated drainage events in the period 2016–2024 (Supplementary Figure S2).

At the southernmost edge of the study region, a first major GLOF was reconstructed based on interviews with sailors, who reported extensive driftwood at Fiordo Fouque, eastern edge of the Cloue Icefield, in March 1998. Satellite imagery and field observations at Lago Covadonga provide clear evidence of a large GLOF event that occurred between 5 March 1997 and 4 February 1998 (Figure 4). A comparison of coregistered aerial imagery (1996) and Landsat imagery (1997 and 1998) shows a 60% reduction in lake extent after the event (from 0.35 km² to 0.14 km²), with a conspicuous breached dam on the terminal moraine visible on the 2002 Landsat-7 imagery. Field observations carried out in March 2016 confirmed the outburst path and showed a drop in lake level of about 34 m and an approximately 185 m wide breach through the terminal moraine. This GLOF released a total water volume of 8.3 ± 1.2 x 10⁶ m³.

The most recent event was noted by a sailboat entering the inner part of Fiordo Fouque in late March 2019. The crew onboard reported extensive wood debris floating in the sea and a large opening in the moraine that allowed passage into the previously moraine-dammed lake by a lifting-keel sailboat. Once inside, they observed that another lake, located at a higher elevation and connected to the previous one by a stream, also had a breached moraine. Subsequent satellite imagery analysis revealed that the event occurred sometime between 4 and 10 November 2018 (Figure 5), when the higher-elevation moraine-dammed lake (informally named Lake 1) catastrophically drained into the lower moraine-dammed lake (Lake 2), which then outburst into the fjord, carving a large breached dam through the impounding terminal moraine (Supplementary Figure S3). Field surveys in 2024, supported by UAV-derived photogrammetry, quantified the breached moraine dimensions (130 m wide, 30 m high) and estimated a lake level lowering of 30 m and 22 m for Lakes 1 and 2, respectively (Figure 6). These two connected GLOFs released a total water volume of 28.3 × 10⁶ m³ (Table 5).

The evolution of glacial lakes in the Cordillera Darwin and Cloue Icefields between 1945 and 2024 highlights a regional transformation driven by glacier retreat after the LIA, climate warming and topographic controls. The shift from ice-dammed to moraine-dammed lakes—evidenced by the decline of ice-dammed lake area from 71.6% (1945) to 14.8% (2024) of total area in this study—aligns with global patterns of glacial lake evolution (Shugar et al., 2020; Veh et al., 2025), which have been observed elsewhere in Patagonia (Loriaux and Casassa, 2013; Wilson et al., 2018), and in other regions such as Alaska (Wolfe et al., 2014; Rick et al., 2022, 2023), Iceland (Zhang T, et al., 2024) and High Mountain Asia (Gardelle et al., 2011; G. Zhang et al., 2019; Shugar et al., 2020). This transition reflects the destabilisation of ice dams due to glacier thinning and the exposure of terminal moraines that impound meltwater in overdeepened basins. The near-disappearance of large ice-dammed lakes, such as those documented in 1945, mirrors processes described by Loriaux and Casassa (2013), where ice-dam collapse led to rapid drainage of ice-contact lakes and subsequent replacement by moraine-dammed lake systems.

The dominance of small lakes, constituting 81.5% of the 2024 inventory, highlights widespread glacier fragmentation and meltwater ponding in peripheral deglacierised terrain. However, large lakes disproportionately contributed to total area (74%), driven by the expansion of moraine-dammed lakes such as Stoppani (3.6 km²), Pigafetta (2.8 km²) and Roncagli/Alemania (2.4 km²) in the CDI, in the Monte Sarmiento massif, and in the Cloue Icefield. This contrast parallels findings in the Himalayas, where small lakes dominate numerically, but larger lakes control total water storage (Gardelle et al., 2011). Similar trends have also been documented in the Southern Patagonia Icefield, where moraine-dammed lakes proliferated as glaciers retreated into steeper terrain (Wilson et al., 2018). Conversely, the modest but steady growth of bedrock-dammed lakes signals the gradual exposure of stable basins, a phenomenon observed in the European Alps and Cordillera Blanca as glaciers receded exposing bedrock overdeepenings (Emmer et al., 2015, 2020).

Regional heterogeneity in lake development reveals the influence of local topography and glacier dynamics. The CDI, hosting 59% of the total lake area in 2024, emerged as the hotspot of growth in the number and area of glacial lakes since 1986 (2.4% yr⁻¹), driven by the retreat of large outlet glaciers into overdeepened valleys. A similar behaviour can be observed in the Cordón Navarro range (2.4% yr⁻¹), where a gradual increase in lake area has occurred through the formation of lakes of similar size sized lakes rather than the enlargement of existing lakes (as the mean lake area remains stable from 0.37 km² in 1986 to 0.42 km² in 2024). In contrast, the number of glacial lakes in the Monte Sarmiento massif remained stable, but the area increased significantly (3.2% yr⁻¹), resulting in a growth in mean lake area from 0.44 km² in 1986 to 0.85 km² in 2024. The Cloue Icefield doubled the number of glacial lakes and the area increased considerably (7.6% yr⁻¹), mainly due to the appearance of large moraine-dammed lakes in the southern slope of the icefield.

Temporal trends emphasise the non-linear response of glacial lakes to climatic influences. The accelerated expansion of glacial lakes between 1986 and 2006 (2.7% yr⁻¹) and 2019–2024 (2.0% yr⁻¹) coincides with the increased warming and glacier mass loss in the Southern Andes region in the late 20th and early 21st centuries (Falaschi et al., 2019; San Martín et al., 2021; Zemp et al., 2025), and has accelerated considerably in the last decade (2015–2024; Dussaillant et al., 2025). Similarly, the CDI-wide average annual climatic mass balance shows a relatively positive period during 2007–2016 and an accelerated negative period after 2016 (Temme et al., 2025).

The documented GLOFs and cyclic outflows in the Cordillera Darwin and Cloue Icefields between 1985 and 2024 reveal different drainage mechanisms operating within the evolving deglacierising environment. The case studies highlight the interplay of dam stability thresholds, water pressure and erosion in initiating outflows and outburst floods, while emphasising the unique constraints imposed by the region’s low relief and coastal setting.