The Aysén region of Chile is home to some 10 300 glaciers, with approx. 8700 km2 of glacial surface area (Direción General de Aguas, 2022). Glaciers have retreated dramatically over the past century, consequent to anthropogenic climate change (Dussaillant et al., 2019; Hugonnet et al., 2021). Since 1870, glaciers around the Northern Patagonia Icefield have lost 25 % (east margin of NPI) and 11 % (west margin of NPI) of their LIA ice extent (Carrivick et al., 2024). Glaciers along the eastern margin of the NPI have experienced stronger rates of decrease (Carrivick et al., 2024; Dussaillant et al., 2018) while glacial retreat along the western margin has been mitigated in part by increases in frozen precipitation, offsetting the net loss of glacier mass (Troch et al., 2024).

As glaciers retreat across Patagonia, the number and size of glacial lakes continue to grow, with a 27 % increase in glacial lake area from 1986 to 2016 (Wilson et al., 2018). This expansion has heightened concern over GLOFs, which have occurred hundreds of times in the region over the past century (Lützow et al., 2023). Many of the GLOFs that have occurred in Patagonia are located in relatively remote valleys with few downstream populations, but have killed livestock and destroyed barns, houses, bridges, and roadways (Iribarren Anacona et al., 2015, 2023; Jacquet et al., 2017).

However, GLOF hazard exposure is expected to increase due to ongoing development and shifting land use dynamics (Dussaillant et al., 2010; Hock et al., 2019). In Aysén, a recent influx of external investment and infrastructure expansion is rapidly transforming long-standing social and territorial relationships. These changes are especially pronounced in struggles over land governance, where global economic forces have collided with local traditions and rural livelihoods (Blair et al., 2019). The convergence of accelerated socio-economic development and climate change has prompted profound shifts in state-society relations and reshaped the political landscape of land governance.

The 1989 GLOF occurred in a remote and sparsely populated valley in rural Aysén, with only a few individuals present to witness the event firsthand. To document these accounts, I conducted interviews with the three surviving witnesses who were present in the valley at the time. These interviews documented witness descriptions of flood magnitude, timing, spatial extent, and other qualitative characteristics to inform model parameterization and provide contextual interpretation of the reconstruction. Questions focused on the individual's experience of the event, where they were located, and flood characteristics at specific points in the valley (e.g., depth, duration, magnitude of flood extent), and the immediate aftermath. Additional questions addressed perceived impacts, government response, and adaptive actions, offering insights into how the event was understood locally and how institutional responses unfolded in its wake.

The accounts that were shared were evaluated for relevance to the modeling process, and for qualitative context of the flood processes. Information describing flood duration, timing, and valley-wide inundation was used to parameterize the hydrologic models. Descriptions of flood depth at specific locations and qualitative observations of flood behavior were retained for contextual comparison with model outputs and interpreted alongside geomorphic observations. These observations were not treated as independent validation because they originate from the same witness accounts used during model parameterization.

Hydrodynamic simulations were conducted using the Copernicus GLO-30 m DEM which provided a common topographic dataset for both the HEC-RAS 2D and GeoClaw models. Using the same DEM for both simulations ensured that differences between model outputs reflected differences in model structure and parameterization rather than terrain inputs. The Copernicus DEM is derived from the TanDEM-X mission, with data acquired between 2010 and 2014 and released for public use in 2019. The DEM covers approximately 480 km2, encompassing the full drainage pathway from Laguna del Cerro Largo through Valle Soler to Lago Plomo.

Because the DEM postdates the 1989 GLOF, it reflects topography modified by flood erosion, deposition, and subsequent geomorphic change. Reconstructing pre-event topography was beyond the scope of this study, as historical topographic data of sufficient resolution were unavailable. We expect the influence of this limitation on the reconstruction to be minor, for two reasons. First, flood routing here is governed by valley-scale confinement rather than by channel-scale geometry. These first-order controls are not sensitive to post-1989 channel migration or sediment redistribution. Second, the scale of any plausible channel adjustment is small relative to the 30 m resolution of the Copernicus DEM, such that sub-grid changes in channel position or bed elevation would not be resolvable within the model terrain. Adjacent Patagonian river systems have likewise experienced only moderate natural channel adjustment over recent decades (Bañales-Seguel et al., 2020). The simulations therefore assume that these changes did not substantially alter the valley-scale controls governing flood routing.

This study compared two flood models, HEC-RAS-2D and GeoClaw (version 5.9.0, 2018) to simulate the March 1989 flood. The two models share a common topographic dataset (Sect. 2.2) but differ in their required input parameters and representation of breach initiation, resulting in differences in simulated flood behavior. The comparison had two objectives: (1) to estimate key flood characteristics – including flood extent, timing, and peak discharge – and (2) to evaluate the influence of model structure by comparing simulations generated from comparable input conditions where possible. The implications of these differing input requirements, along with initial assumptions, are discussed in detail in the following subsections.

Witness testimonies state that the propagating flood wave filled the valley from wall to wall and contained a slurry of sediment-laden water and woody material. This is plainly evident in the initial stretches (< 10–12 km) of the valley (Fig. 1), but these indicators, characteristic of GLOF flows, are more difficult to detect from remote sensing technology and noticeably diminish at ∼ 15 km downstream. At ∼ 12 km downstream, visible signs of sediment deposition largely disappear, seemingly inconsistent with the magnitude of the GLOF. Emmer (2023) discusses the complexity of vanishing evidence of GLOF events, a paradox which motivated the current analysis. For this reason, a sedimentological and dendrologic sampling field campaign was undertaken to compare with the results of the numerical modeling of the GLOF event.

Sampling sites were selected based on anticipated flood impacts, including low-energy environments conducive to sediment deposition and high-energy zones likely to produce dendrogeomorphic evidence of disturbance (Fig. 3). Samples were collected during the same field campaign and are discussed separately in continuation.

We collected two samples (n=2) of buried vegetation from stratigraphically distinct organic horizons within downstream sediment deposits to constrain the age of a previously undocumented GLOF event inferred to have originated from Laguna Turbio (Fig. 1) and whose geomorphic impacts are visible in historic remote sensing imagery. Using the RPA, we recovered two distinct organic layers from a stratigraphically continuous sequence, each representing a buried vegetation horizon. The shallower organic horizon directly underlies the upper clastic unit attributed to the 1989 GLOF, whereas the deeper organic horizon underlies a second, older clastic unit interpreted as evidence of an earlier flood. Our objective was to test whether the upper horizon yielded a modern age consistent with the 1989 event and whether the deeper horizon provided a maximum age constraint for the older flood deposit. The two samples were processed as bulk carbon and analyzed in the University of Colorado Boulder AMS Radiocarbon Prep Laboratory. Although the sample size is limited, additional radiocarbon analyses were beyond the scope of this study due to analytical costs.

We used witness accounts of the flood to constrain key parameters used as model input (full interviews are found in Part S1 in the Supplement), using their locations and descriptions of flood behavior. According to those who witnessed the event, there were only four people in the valley when the flood occurred, three of whom are surviving. I spoke with all three of the surviving witnesses to reconstruct the flood using the hydrologic models. From the interviews, we gathered important discrepancies from descriptions of the previous scientific accounts. First, two brothers and a third helper were working in the upper section of Valle Soler when they saw the rushing wave of water. They describe a “rapidly descending flurry of water” that “filled from valley wall to valley wall” (de cordillera a cordillera). One worker had to rapidly climb a tree, some 6–8 m high to avoid being swept away, and the brothers scrambled up a nearby hill.

These accounts contrast dramatically with the hydrologic model results of peak discharge estimates by Hauser (1993) which calculated, from empirical relations, the discharge to be around 2000 m3s-1. Modeled results with Hauser's peak discharge were insufficient to replicate the magnitude and depth of flooding witnessed (Burton et al., 2020). Second, the mode of moraine failure differed among accounts. From the initial report, Hauser posits an iceberg likely blocked the outlet causing the eventual failure of the dam, either from overtopping or compromising the dam as the water moved around the blockage. This differs from the witness accounts, all of whom did not mention an eventual increase of floodwater but describe the advancing flood wave as “all-at-once”, leaving them with little time to escape. Two witnesses stated they thought the moraine must have failed from below to release so much of its waters so rapidly.

Further downstream, the third witness describes his account of the flood (Part 1b in the Supplement), which was used to contextualize the rapid onset of the descending flood wave, as opposed to a gradual deepening more consistent with the original estimation of flood discharge.

Hydrologic model outputs of the 1989 event largely coincide between the two approaches, yet important differences arise based on the different initial assumptions (Fig. 5, videos S2 in the Supplement). As noted by Turzewski et al. (2019), the assumption from GeoClaw of instantaneous dam failure leads to higher peak discharges and shorter time to peak discharge than would otherwise be expected. This is evident through an apparent 15–30 min advance in the GeoClaw simulation compared with HEC-RAS-2D at the same Δt.

The GeoClaw simulation fully floods the downstream valley filling the canyon walls with 10 m of water, reaching the Rio Cacho tributary (Fig. 1) within the first 15 min. The flood loses energy upon reaching the wider stretches of the middle Valle Soler, and the advancing flood wave attenuates to a height of 4–6 m. A portion of the flood is diverted north, moving upstream of Rio Cacho. Within only 60 min, the wave front has nearly reached the end of the valley while the main bulk of water volume accumulates into a trailing pulse of water growing and diminishing in height with the contours and constrictions of the valley walls. By the end of the first hour, the bulk volume of the flood reaches the furthest stretches of Valle Soler, where an end moraine creates an abrupt constriction with two near 90° diversions that causes the water to pool 10 m deep as it moves through the constriction and drains into Lago Plomo.

HEC-RAS-2D, by contrast, includes a dam-break module that allows users to define initial conditions replicating the physical characteristics of a dam. In this case, the gradual erosion of the terminal moraine moderates the downstream propagation and reduces the depth of the advancing flood wave (Fig. 5). The simulation initiates with shallow overtopping that begins flooding the upper section of Valle Soler, followed by the main flood wave breaching the moraine and overtaking the initial inundation front. Notably, there is a time lag of approximately 30 min in the arrival of the bulk flood pulse in the HEC-RAS simulation compared to GeoClaw for comparable downstream locations. The bulk flood pulse similarly reaches depths exceeding 10 m, dependent on the constrictions of the valley width, accumulating at the final constriction at 90 min. The bulk flood pulse pools at this constriction for some 3–4 h as the outlet slowly drains the floodwater into Lago Plomo.

The sediment deposited by the 1989 Soler GLOF displayed a range of depositional environments (Fig. 6). Sediment ranged from silt- and clay-sized particles (approximately φ = 4–6) to coarse sand (approximately φ = 1–3), interspersed with organic horizons. When extracting sediment samples, we interpreted the first non-organic lens of clastic sediment to be most likely deposited by the 1989 GLOF, informed by several field workers' (herein referred to as campesinos) accounts' that Rio Soler had not flooded its banks since the 1989 GLOF. Although our field samples and analysis did not permit an in-depth sedimentological analysis (e.g., grading or sedimentary structures), the interpretation of this unit as a single-event deposit is supported by its consistent occurrence as a discrete clastic layer bounded by organic horizons across all sampled cores, together with the modern radiocarbon age and the absence of any known overbank flood of comparable magnitude since 1989. The organic horizons similarly varied in thickness, which we attributed to the various geomorphic environments in which they were found (i.e., low-lying wetland, seasonal pasture, raised floodplain). While individual core thicknesses varied between sites, reflecting local depositional environments (wetlands, raised floodplains, backwater settings), the upper clastic unit was present throughout the sampled reach. This consistency suggests that the 1989 GLOF deposited a valley-wide depositional unit, rather than isolated overbank deposits. Farther downstream, the cores also record a progressive fining from coarse sand toward finer sand and silt, consistent with the modeled attenuation of flow energy as repeated valley expansions promoted lateral spreading and sediment deposition. Rather than abrupt changes in depositional style, grain size exhibits an overall downstream fining trend at the valley scale, while local variations in deposit thickness and texture reflect depositional environment (e.g., wetlands, backwater settings, raised floodplains, and areas adjacent to topographic constrictions).

In all, we collected six sediment samples using the RPA (Part S3), and another nine using the sediment probe (Fig. 7). At each site, we typically extracted two pilot probes using the sediment probe to gain a snapshot of the substrata, but a malfunction with the rotating blade on the RPA inhibited further use at several locations, as the sediment was either too dry (would crumble apart) or oversaturated with water (and would not hold its shape once extracted).

A total of 50 cores were extracted from N. pumilio (Lenga, n=34), N. betuloidus (Coihue, n=10), and N. dombeyi (Ñirre, n=6). Where several studies have successfully related growth disturbances of trees with geomorphological processes such as flash floods (Casteller et al., 2015) and debris flows (Bollschweiler and Stoffel, 2010), few studies have related dendrogeomorphic samples with GLOFs (Gorsic et al., 2025), and this is still a developing science. In the field, we found many trees to have a strong outer bark, but to be soft and rotten on the inside beyond a few centimeters (∼ 6–10 cm). We later found out that Nothofagus species are impacted by a spreading fungus which produces similar symptoms of inner rot, particularly in wetter environments (Cwielong and Rajchenberg, 1995) that significantly inhibited a more robust sampling for dendrogeomorphologic impacts. Indeed, upon subsequent analysis, we found fourteen of the samples to be partially rotten to such a degree that we were unable to extract reliable ring counts from them and had to be discarded. Seven samples from these did not reach the desired temporal extent to capture the 1989 event and were similarly unusable. Of the remaining n = 29 samples, we were only able to obtain a series intercorrelation of r= .343 convergence on the tree ring growth disturbances. For this reason, we opted to abandon the dendrogeomorphologic data.

Some older living trees retained large scars on the upstream side of their trunks, likely sustained by the initial impact of the debris-laden flood wave. We were unable to extract dendrogeomorphic samples from these scarred trees, however: those with visible scarring were either rotten from the fungus described above or sufficiently large that our tree corers could not reach the relevant growth rings.

Although the ring-count analysis was abandoned, the physical scars and stranded debris themselves record useful information on flood stage, independent of the cross-dating statistics. At five sites we measured the height and orientation of flood-related scars and stranded debris as approximate paleostage indicators (Table 3; Figs. 8 and 9). Scars occurred consistently on the upstream, river-facing aspect of trunks, indicating the passage of a debris-laden flow advancing down-valley. Measured impact heights ranged from approximately 3 to 7 m above the present channel, including remnant debris stranded ∼ 7 m above the river surface (Fig. 8D). The stranded trunks were uniformly decayed and piled several meters above the current Río Soler high-water bank, indicating that they had accumulated over some time (decades) and were unlikely to have been deposited by normal seasonal high flow. These heights are consistent with the several-metre flow depths reconstructed at each location. Where scar heights fall below the modeled depth, most notably at the WD4 bottleneck (Table 3), this is expected, since debris caught by the passing leading edge of the wave records local arrival stage rather than the deeper peak inundation that developed later at the constriction.

Finally, the topographic constriction in the lower valley caused floodwater to pool temporarily before being funneled through the narrow channel toward Lago Plomo. The prominent eddy-bar deposit at this constriction (Fig. 9), situated approximately 8–12 m above Río Soler, was likely emplaced as the impounded water drained – representing the highest paleostage indicator recorded and consistent with the > 10 m pooling depth reproduced by both models at this location.

Clear geomorphic indicators of the 1989 flood are still present in the valley. However, these same features complicate the interpretation of additional evidence pointing to an earlier GLOF event that also impacted the landscape. For example, an analysis of sediment transportation during the 1989 event showed a counterintuitive coarsening of grain size in the initial 5 km of Valle Soler, before a gradual fining below the outlet of Lago Turbio (Fig. 1). While planning for the field expedition, remote sensing analysis suggested a prior outburst flood originating from the adjacent Laguna Turbio. Field observations supported this hypothesis, revealing multiple indicators of a past GLOF: a breached terminal moraine, a visibly lowered lake level, a downstream impact zone, and a consistent reduction in tree size across the suspected flood path – suggestive of uniform re-colonization following disturbance (Emmer, 2023).

We extracted and analyzed organic material from sediment samples extracted with the RPA to date the deposition of clastic sediment on top of the vegetation layer that we assumed as preexisting peat vegetation present before the 1989 GLOF. The sample was extracted near RA 3 (White “C” – Fig. 6) and was approximately 50 cm deep. The 14C sample taken from the organic horizon below our target clastic depositional layer returned “modern age”, with a fractional carbon at 1.1921 (Fig. 10). This indicates a high likelihood that our collected 14C sample was taken from organic material deposited in or relatively close to the 1989 flood. The second 14C sample measured a final fractional carbon ratio of 0.9845, with an estimated age of 125 ybp (SD 20), which, relative to the time of analysis (2024), would place the GLOF event took place in 1899 ± 20 years.

The combined witness accounts, hydrodynamic simulations, and geomorphic evidence indicate that the 1989 Laguna del Cerro Largo GLOF resulted from rapid catastrophic moraine failure rather than progressive outlet blockage. Together, these complementary lines of evidence point to an abrupt breach that generated a high-energy flood wave capable of transporting exceptionally coarse sediment throughout the upper valley. Witnesses did not observe a gradual onset of floodwater that would have signaled an impending flood; rather, they describe the main flood surge quickly overtook the initial wave front, with the advancing flood wave likely reaching several meters in height. The rapid advancement of the flood wave is corroborated by both hydrologic models, as well as witness testimony.

Hauser (1993) proposed that outlet blockage by an iceberg initiated overtopping and progressive moraine incision. While that mechanism cannot be excluded, it is less consistent with the evidence assembled here. All eyewitnesses independently described an abrupt flood onset with little warning, and both hydrodynamic reconstructions reproduce a rapidly propagating flood wave rather than a prolonged period of increasing discharge. Together with the late austral summer timing of the event and the absence of a substantial ice apron on the cirque headwall, these observations are more consistent with catastrophic overtopping initiated by a large icefall-generated displacement wave, as has occurred in other regions of the world (Colavitto et al., 2024; Hubbard et al., 2005; Shugar et al., 2021).

The Soler GLOF illustrates how rapidly GLOF rheology can evolve during downstream propagation. Mobilization of boulders exceeding 8 m in diameter (Fig. 11) indicates that the initial flood behaved as a dense, sediment-rich flow capable of transporting exceptionally coarse material. However, multiple valley expansions, abrupt channel bends, and decreasing valley gradient promoted rapid lateral spreading and energy dissipation within the first ∼ 12 km downstream. As sediment deposition progressively reduced the suspended sediment concentration, the flood transitioned toward a more water-dominated flow. Coarse gravel and sandy deposits attributable to the 1989 event were continuously identified over approximately 12 km downstream from the breach, with deposition bars found more dispersed farther downstream. The sediment cores independently record this downstream evolution. Although individual deposits varied in thickness according to local depositional setting, the upper clastic unit attributed to the 1989 GLOF was laterally continuous across all sampled sites and consistently ranged from coarse sand in the upper valley to finer sand, silt, and clay downstream. This pattern indicates that first-order depositional processes were governed by valley-scale topographic controls, with repeated expansions and constrictions progressively attenuating flow energy while producing a broadly consistent depositional signature throughout the lower valley.

Regional comparison places the 1989 Laguna del Cerro Largo event among the largest documented GLOFs in northern Patagonia (Table 4). Although direct comparison between historical events is complicated by differences in reconstruction methodology and dam type, Laguna del Cerro Largo occupies the upper end of the magnitude spectrum for moraine-dammed GLOFs in the region. This comparison also highlights the limitations of applying empirical peak-discharge relationships beyond the range of events from which they were derived. In this case, extrapolation of commonly used parametric equations substantially underestimated the magnitude of the 1989 flood, suggesting that physically based hydrodynamic reconstruction may be more appropriate for exceptionally large moraine-dammed lakes that lie outside the calibration range of existing empirical relationships.

The exceptionally large, reconstructed discharge must, however, be interpreted within the assumptions of the modeling framework. Both hydrodynamic models solve the shallow-water (Saint-Venant) equations and therefore assume Newtonian flow, an assumption least valid in the upper valley where boulder mobilization (Fig. 11) indicates the flood likely behaved as a dense, debris-rich slurry rather than clear water. A non-Newtonian flow of higher bulk density and yield strength conveys a given discharge at greater depth than the models assume; conversely, reproducing the observed inundation depths with a Newtonian model requires a larger water discharge than the true water-plus-sediment flux. This is consistent with our interpretation that the reconstructed peak discharge represents an upper bound. Importantly, this bias is spatially concentrated: as the flood attenuated and spread laterally within the first ∼ 12 km, losing turbidity and sediment load, the flow approached the clearer, more Newtonian regime for which the shallow-water assumption is best suited – the same downstream reaches where our sediment and witness-based evaluation is concentrated. We therefore treat the upper-valley discharge estimates as first-order approximations of a debris-influenced flow, while regarding the downstream reconstruction, against which the field evidence is compared, as less affected by rheological simplification. Full debris-flow rheological modeling was beyond the scope of this event reconstruction but represents a valuable direction for future work, especially as recent work has highlighted these important discrepancies (Lehnigk et al., 2026).

The lack of easily accessible and decipherable evidence of previous GLOF impacts in the lower reaches of Valle Soler raises important questions regarding the legibility of GLOF indicators, similarly analyzed by Emmer (2023). Beyond the initial sediment redistribution of upper Soler, the lower portions of the valley remained seemingly undisturbed, with fewer geomorphological impacts visible from available remote sensing (Fig. 12). Moreover, compounding GLOF events, such as occurred with LCL and Laguna Turbio, further complicate the legibility of even large-magnitude GLOF events. The relative uniformity of the upper clastic unit despite substantial differences in local geomorphic setting also illustrates how major GLOFs can leave deceptively subtle downstream sedimentary signatures, particularly once vegetation has recovered.

Conventional methodological approaches to estimating peak discharge in GLOF events can significantly misrepresent the true magnitude of these extreme events, particularly when applied beyond the empirical range of their original datasets. In this case, the initial empirical equations used by Hauser (1993) to estimate peak discharge (Qp) underestimated the flood by nearly an order of magnitude. This discrepancy likely stems from the unusually large volume of Laguna Cerro Largo (LCL), which exceeds the scale of events used to derive those equations. The formula that was applied in the official report relates discharge (Qp) to the volume of water drained, yet the case studies collated into this database were collected from ice-dammed lakes (Clague and Mathews, 1973), skewing the results in the 1989 Soler GLOF. Other formulae, like Costa and Schuster (1988) were based on data linking Qp to the calculated potential energy of the impounded lake. Those data points ranged from approximately 10¹⁰ to 10¹² J, whereas calculations with data taken from Burton et al. (2020) suggest a potential energy closer to 10¹⁴ J – 2 orders of magnitude greater.

The limitations of empirical formulae become more evident when compared to eyewitness testimony. All accounts, given by individuals located at different points in the valley, consistently describe a sudden onset of a high-energy, torrential flow that stood several meters tall. Witnesses reported that the floodwater rapidly inundated the entire valley floor, reaching depths of several meters at their respective locations. These descriptions sharply contrast with Hauser's estimated peak discharge (Qp) of 1800–2000 m3s-1. At that rate, and assuming sustained peak flow, it would have taken approximately 31 h to drain his estimated flood volume of 229 × 10⁶ m3 (Hauser, 1993). In contrast, interviewees recalled that the bulk of the flood passed through in just 8 h – long enough for one campesino to spend the night in a tree, which he had climbed 6 to 8 m above the ground to escape the rising waters. This discrepancy suggests that the original discharge estimate significantly underestimated the true magnitude of the event.

While these empirical equations have been reanalyzed and calibrated in the decades since, it is important to note that the applied results of these formulae are often not, with lasting consequences for policy and hazard planning. Local policymakers and scientists were more familiar with estimates in the 2000–3000 m3 range, citing Hauser (1993) and the cyclical ice-dammed GLOFs from Cachet II (Jacquet et al., 2017; Personal Communication, 2024). Discrepancies like these comprise a potentially dangerous miscommunication that impacts subsequent development of policy and emergency preparedness plans.

GLOFs have been studied predominantly from glaciological and other Earth science perspectives. While contributing a foundational geophysical understanding of the complex process chains and downstream impacts, the remote nature of these processes complicates analysis where data are often sparse or non-existent. As illustrated by Williams et al. (2022), triangulating an understanding of geomorphic events such as GLOFs using diverse data can help reconstruct geomorphic processes in data sparse areas.

The 1989 Soler GLOF is readily identifiable in satellite imagery, with distinct geomorphic signatures detectable via remote sensing. However, the temporal resolution is constrained by the infrequent revisit times of early satellite platforms at mid-latitudes and by persistent cloud cover, customary for this region. Additionally, repeat events, such as the overlaid GLOF deposition from the Soler and Turbio GLOFs, obscure the prior events and their visible indicators.

Witness testimonies were integral to providing crucial data for the hydrologic models. Further, their local expertise and intimate familiarity with the landscape provided additional details that would be otherwise exceedingly difficult to extract via remote sensing. For example, the dense vegetation present in the valley has all but obscured the most obvious of geomorphic features of the outburst flood. Moreover, local familiarity with the land is an underappreciated source of knowledge in scientific research (Balay-As et al., 2018; Cadag and Gaillard, 2012). Campesino accounts of a rapid and catastrophic moraine dam failure aligned more closely with the results of the hydrologic simulation than with the moderated descriptions and initial reports. Community members also demonstrated awareness of other potentially hazardous moraine-dammed lakes within the valley. Yet these histories were seemingly underappreciated from the initial surveys, as the official report contains only fleeting references to witness testimonies in passing (Hauser, 1993).

This research illustrates an important first step in approaching a more holistic understanding of GLOF risks, similar to other mixed-methods reconstructions of flood events (Williams et al., 2022). Given the limited empirical data and uncertain dynamics of glacier-lake systems in the region, establishing how, when, and where floods occur is a crucial first step in contextualizing the lived, political, and governance dimensions of hazard. These findings therefore make an essential – if partial – contribution toward a more holistic understanding of GLOF risk that integrates material processes with the social and institutional conditions through which risk is produced and experienced.

However, there are also many challenges and limitations that arise from this type of work. The inability to statistically correlate samples taken from impacted trees, despite this methodology has been successfully applied in similar contexts (Gorsic et al., 2025) underlines the difficulty of interdisciplinary research. Reconstructing extreme events through hydrologic modeling requires extensive and accurate input data, and may inherently contain uncertainty (Westoby et al., 2014; Williams et al., 2022). For example, the peak discharge reported here is likely an overestimate, reflecting the underlying assumptions and simplifications of the hydrologic model. As with empirical formulas that require context-specific calibration, it is important to acknowledge that scientific results often take on a political life of their own that may persist unexamined long after the initial analysis is complete.