Event · · ~17:00 (UTC−3)
Numerical reconstruction of the San Carlos de Apoquindo debris flow
English version translated from the Spanish original as of 13 October 2026; the Spanish site is updated first.
An independent study simulating the 8 October 2026 debris flow using public sources, observational data, and explicit modeling assumptions: the foothill terrain, all 42,018 buildings in the area, measured rainfall, and mudflow physics.
The hypothesis selected before the test using held-out observations (main channel, southern branches, and rainfall on the hillside) scores 72/100, and that test yielded an unfavorable result. In subsequent tests, which are for fitting rather than validation, allowing mud to enter houses and distinguishing the turbid water on campus and on La Plaza Sur from the mud in the main branch raises the index verified with a fine time step to 76–78/100. The integrated hypothesis obtains 82/100 with an assumed rainfall of 35 mm/h for one hour (verified with a fine time step), 81/100 with rainfall measured at the Río Molina station with its pulse shape, and 79/100 with that same rainfall spread evenly over 4 hours: it is therefore reported as a range, 79–82/100 depending on rainfall. Fire severity measured with Sentinel-2, a log-jam test, and a chronological account of the events with each fact attributed to its source have been added (see the paper). This is not predictive accuracy: the index summarizes streets with mud, affected facilities, reach to Padre Hurtado, and mud within the official affected area from IDE Chile, and serves to compare hypotheses. Details of each simulation.
Recorded from the viewer; preliminary simulation.
01 / Context and scale
The event and the study, in figures
- Rainfall on 6 and 7 October
- ≈100 mmSantiago foothills, according to CR2 (72 mm at Quinta Normal); the mayor spoke of 144 mm accumulated, without an identified station · 42.8 mm on 8 October at Río Molina (CR2)
- Homes affected
- 2,01555 uninhabitable · municipal assessment of 9 October
- Contributing catchment
- 10.1 km²0 °C isotherm ≈ 2,800 m (reported): supports the assumption of rainfall across the entire catchment
- Buildings in the model
- 42,018Mesh of ≈852 thousand triangles
Mud volume · calculated, not measured≈300,000 m³ of mixture · independent estimate adopting the reported saturation: median 295,000 m³ (5th and 95th percentiles: 181,000–468,000, under the adopted assumptions)
In the scenarios tested and under their assumptions, that volume allows the reach criterion to Padre Hurtado to be reproduced, within the order of magnitude of an independent rainfall-based estimate. Both figures are results calculated under assumptions, not measurements.
Source of the figures on this home page: hechos.json · version 2026-10-13.
02 / Research question
How can the debris flow be reconstructed from terrain and flow physics?
What it seeks to answer
Which simulated flow paths, volumes, and mechanisms are compatible with public information about the event, by comparing hypotheses about its origin using a numerical reconstruction informed by public sources and explicit assumptions.
The model calculates how a layer of mixture moves over the terrain and incorporates resistance to the onset of motion, viscosity, and turbulence (ANUGA 4.0.1 with O'Brien and Julien rheology; details in Method).
Scope and limitations
This is neither an expert assessment nor an official report, and it does not attribute responsibility. The study is independent and is in development.
The study uses only public data. It distinguishes what is observed (reported by authorities, the press, and official mapping), what is calculated (model outputs under assumptions, such as volumes and discharges), and hypothetical scenarios (variants tested, such as culvert capacities). Nothing calculated or hypothetical is a measurement or an official conclusion; all require independent scientific validation.
03 / Working scenarios
Hypotheses compared
Study hypotheses; these are not conclusions about the event. The figure is the exploratory compatibility index (0–100), not an accuracy measure.
- H1
- All flow exits through the main channel onto the alluvial fan, with the conduits blocked. 68/100 with the full city (all 42,018 buildings).
- H2
- Overflow only at the Francisco Bulnes Correa culvert inlet. 16/100: ruled out as the sole explanation.
- H3
- Distributed inflow: main channel + southern branch (U. de los Andes campus) + short ravines toward La Plaza Sur.
- H3 + rainfall
- Selected for the held-out test · 72/100 (unfavorable test result). H3 plus rainfall falling directly on the burned hillside and descending through ravines and roads toward Clínica U. de los Andes and La Plaza Sur.
- Integrated
- Subsequent fitting · 79–82/100 depending on rainfall (82 with an assumed 35 mm/h, verified with a fine time step; 81 with rainfall measured at Río Molina; 79 with that rainfall spread over 4 hours; its components, 76–78/100). H3 + rainfall with houses that allow mud to enter, turbid water on the lower hillside, and features not resolved by the terrain: the hill road, the confined carriageways of La Plaza Sur, and university entrances. It meets all four original criteria with assumed rainfall and three of four with measured rainfall.
- Log jam
- Does not explain the southern sector · 78/100 (79 without a log jam). A 3 m accumulation of logs at the end of the line of trees in the main ravine does not divert flow southward: the Álvaro del Portillo and campus sector has the same 20.2 ha wetted with and without a log jam. It retains and distributes mud near the alluvial fan.
- Fire
- Quantified precondition. Without fire, the balance gives about 62 % of the mud volume, 43 % of the solids, and about one-quarter of the debris-flow probability. In the city, the mud wets 13–19 % less area, but the index does not fall (80–82): the metric does not distinguish this effect.
- Bike park
- The trails do not change the route taken by the water; their area does contribute water toward La Plaza Sur. Their possible sediment contribution cannot be evaluated with this model, which does not simulate erosion.
04 / Preliminary findings · 13-Oct-2026
What the simulations show so far
Simulations with the full city (all 42,018 buildings); subject to final runs and satellite validation.
- 1
- CalculatedThe city channels the mud. With all buildings, mud flows deeper along streets: with only the main channel, the model calculates 1.16 m around Clínica UC Christus (the consulted source reported about 1 m in the emergency department; a maximum near the facility does not necessarily represent its interior) and 1.9 m beside Duoc UC, and reproduces La Plaza, General Blanche, and Camino El Alba. When houses retain mud (integrated hypothesis), the thickness beside UC Christus falls to 0.30 m: no configuration simultaneously reproduces that meter and the extent within the official area.
- 2
- ObservedCalculatedAbout 300 thousand m³ (calculated), with reported saturation and a high isotherm. Observed: about 100 mm fell in the foothills on 6 and 7 October according to CR2 (the mayor spoke of 144 mm accumulated, without an identified station), and the reported isotherm was near 2,800 m, above the catchment summit; no snow was detected in the cloud-free part of the catchment in the 21 September Sentinel-2 image. On 31 January, according to T13, 75 mm fell in 80 minutes (Interferencia mentions about 20 minutes of intense rainfall) and the event was much smaller: the comparison is compatible with an influence of antecedent saturation, although by itself it does not establish its weight relative to intensity.
- 3
- CalculatedSouthern branches and rainfall on the hillside. In the simulations, those inflows produce flow beside Clínica U. de los Andes (0.14 m at the control point) and on La Plaza Sur, compatible with the exterior impacts reported by the clinic (other sources reported mud reaching the emergency department; the accounts differ) and with Senapred, which names La Plaza Sur among the most affected streets.
- 4
- ObservedHypothetical scenarioBulnes Correa culvert. Observed: official hydrographic mapping shows the channel continuing along Quebrada Honda toward Dipreca, downstream of the culvert inlet. Calculated: with the full city (all 42,018 buildings), discharge at the culvert inlet reaches ~65 m³/s. Hypothetical scenarios: under those assumptions, a capacity of 30 m³/s is insufficient, and one close to ~65 m³/s, with complete capture, would prevent overflow. The actual capacity and condition of the culvert are not public: this does not evaluate the structure.
- 5
- CalculatedHouses that allow mud to enter. Representing buildings in the official area as porous increases the share of urban mud within the official area from 34 % to 54 % and wets 40 % of the affected properties (verified index 76/100). Trade-off: thickness in the UC Christus emergency department falls to 0.30 m.
- 6
- ObservedCalculatedTurbid water on campus and on La Plaza Sur. A Meganoticias video shows turbid water, not thick mud, filling a sunken courtyard on campus. In the detailed simulations, overflow from Álvaro del Portillo does not reach it; water from the hillside and upper slopes, channeled by the carriageways at different levels on La Plaza Sur, enters through a clearing and fills the courtyard to 12–30 cm, within the range in the video, with rainfall of 15 to 35 mm/h, and wets 41–87 % of the avenue depending on intensity. Carving the main bike park trail does not change that route, but its area does contribute water. With a fine mesh on campus within the city simulation and measured rainfall (m36), the courtyard reaches 0.17 m and La Plaza Sur is wet along 58 % of its length: the difference at city scale was due to resolution. A fitting hypothesis based on contributed observations.
- 7
- CalculatedFire measured by satellite. According to the Sentinel-2 dNBR index (23 December 2025 versus 6 February 2026), 82 % of the contributing catchments shows a burn signal, but only 10 % with moderate-high or high severity. With that severity, the empirical volume from Gartner (2014) falls from 374 thousand to 148 thousand m³. Under the balance assumptions, without fire the pulse would have had about 62 % of the mud volume and 43 % of the solids, and about one-quarter of the debris-flow probability (M1 model of Staley et al., 2017, used for illustration). In the city, without fire the mud wets 13–19 % less area and leaves 0.19–0.23 m at UC Christus, versus 0.30 m, but the index does not fall (80–82).
- 8
- ObservedCalculatedMeasured rainfall with its temporal pattern. Hyetograph from the Río Molina station (DGA), digitized from figure 3 of the CR2 analysis: 31.8 mm in 8 intervals of 30 minutes between 15:00 and 19:00, with a maximum of 10.9 mm between 17:30 and 18:00. With that rainfall, the integrated hypothesis obtains 81/100.
- 9
- Hypothetical scenarioLog jam. With this terrain, a 3 m accumulation at the end of the line of trees does not divert flow southward; it retains mud near the alluvial fan. It does not explain the southern sector.
- 10
- Unresolved. The test with held-out observations had an unfavorable result and subsequent trials are fitting: validation with new data is still needed (for example, Sentinel-1 or official level marks). Open issues remain: the thickness at Clínica UC Christus, which no configuration reproduces together with the extent within the official area, and what the model does not simulate: erosion and deposition, entrained logs and vehicles, urban drainage, and infiltration.
04b / Confidence level
How confident we are in each result
Qualitative, provisional assessment of confidence in each result, conditional on the assumptions and available evidence.
| Result | Confidence level |
|---|---|
| General mudflow corridors (El Alba–General Blanche and Quebrada Honda) | Relatively high: compatible with multiple scenarios and with the official affected area |
| Need to represent channels and culvert inlets | Robust finding on model sensitivity |
| Order of magnitude of the volume (~300 thousand m³) | Conditional estimate, with a wide range (181–468 thousand m³) |
| Actual culvert capacity | Undetermined: hypothetical scenarios only (10–200 m³/s tested; in these, overflow is prevented from ~65 m³/s) |
| Mechanism of water flow on campus and on La Plaza Sur (channeled local runoff) | Fitting hypothesis: compatible with a video and contributed observations; requires field verification |
| Exact depth in front of each building | Exploratory: depends on terrain and parameters |
| Velocities, pressures, and arrival times | Preliminary indicators, without instrumental validation |
| Thickness in the Clínica UC Christus emergency department | Not consistently reproduced: ~1 m without porous houses and 0.30 m in the integrated hypothesis, versus about 1 m reported |
| Effect of fire on volume and flow probability | Conditional estimate: severity measured with Sentinel-2; coefficients by severity are assumed and the probability model is illustrative |
| Index of the integrated hypothesis (79–82/100) | Fitting, not validation: depends on assumed or measured rainfall and was developed after the held-out test |
05 / Study workflow
From evidence to results
Evidence and evidence log
More than 280 source consultations recorded with date and hash, including automated satellite-catalog queries.
Terrain
30 m terrain resampled to 4 m, 42,018 buildings, and channels carved according to the official hydrographic network.
Physics
ANUGA 4.0.1 and O'Brien and Julien rheology.
Volume
Rainfall, isotherm, reported saturation, and fire severity measured with Sentinel-2: calculated median of 295 thousand m³.
Calibration 79–82/100
Four criteria set in advance; about 80 simulations (19 screening runs with the previous terrain, 37 with the full city, 5 for concentration quadrature, and 19 campus microsimulations); unfavorable held-out test result (72/100 for the selected hypothesis).
Verification
Stopping, convergence (including in the city), and volume balance.
Comparison with observations
Account of the events with each fact attributed to its source, official affected area (IDE Chile), hydrographic network, and Sentinel-2.
Final results In progress
Validation with new observations that were not used to construct the hypotheses (for example, Sentinel-1 radar or official level marks).
Methodological sequence. The numbering does not indicate that all preceding stages are complete.
06 / Explore the study
Visualization, papers, and data
Open access, no registration. CC BY-NC-SA 4.0 license; media use permitted with credit (see terms).
3D viewer
Scenarios with the full city (all 42,018 buildings) and their exploratory compatibility index against observations.
Paper · PDF · 13-Oct versionNumerical reconstruction
Technical document: data, model, verification, hypotheses, and comparison with observations.
(Spanish original) Sub-study · PDFCensus-based exposure
Homes and census population within the simulated propagation area with the full city (all 42,018 buildings; 11-Oct version). Calculated figures conditional on the model: they do not identify affected homes.
(Spanish original) MethodologyMethod
Physics, calibration, verification, and change log.
ReproducibilityData
Simulation table, code, and searchable evidence log.
07 / Frequently asked questions
Frequently asked questions about the Las Condes debris flow
Brief answers based on the study and its public sources (13-Oct-2026 version). Items marked as preliminary may change with the final simulations.
What happened in San Carlos de Apoquindo on 8 October 2026?
At around 17:00 (UTC−3), a debris flow—a flow of water, mud, and rocks—descended from the ravine draining the Andean foothills of San Carlos de Apoquindo, in Las Condes (Santiago de Chile), and traveled along residential streets for several kilometers. According to the municipal assessment of 9 October, 2,015 homes were affected (55 uninhabitable) and 6 people were injured; Las Condes was declared a disaster area. According to the mayor, about 144 mm had fallen, with the soil saturated (CR2 reports about 100 mm in the foothills on 6 and 7 October). The hillside had burned in the fire that began on 29 December 2025, with a preliminary area of 1,012 ha in the municipality according to the municipal authority; according to this study’s calculation using Sentinel-2 imagery, 82 % of the contributing catchments shows a burn signal, mostly of low or moderate-low severity.
Which streets and places did the Las Condes debris flow affect?
According to press reports and information from the municipality and Senapred compiled by the study, mud traveled along General Blanche, Quebrada Honda, Camino El Alba, and Av. La Plaza (including La Plaza Sur), reaching Av. Padre Hurtado and the area around Pueblito Los Dominicos. Affected facilities include Clínica UC Christus San Carlos de Apoquindo (about 1 m of mud in the emergency department), Clínica Universidad de los Andes (accounts differ: the clinic reported only exterior impacts and other sources, mud reaching the emergency department), the Duoc UC San Carlos de Apoquindo site, and the Universidad de los Andes campus. This is not an official list: for current road closures, consultation with Municipalidad de Las Condes and Senapred is recommended.
Why did Quebrada Honda street flood?
Observed: the sources consulted reported overflow at Francisco Bulnes Correa, and official hydrographic mapping shows the channel continuing along Quebrada Honda toward Hospital Dipreca, downstream of the culvert inlet. Calculated: outflow only at the inlet does not by itself explain the debris flow; instead, the city channels part of the mud emerging on the alluvial fan toward the inlet (about 93,000 m³, with peaks near 65 m³/s), and in the simulations flow that does not enter the culvert continues along Quebrada Honda. Hypothetical scenarios: in capacity trials that do not describe the real structure, a capacity of 30 m³/s is overwhelmed. The actual capacity and condition of the culvert are not public: the study does not evaluate the structure, determine legal causation, or attribute responsibility.
How did water enter Universidad de los Andes?
According to the detailed simulations, it did not enter through overflow from Av. Álvaro del Portillo, which flows elsewhere in the model. A Meganoticias video shows turbid water descending from the woods and filling a sunken courtyard beside a campus building. The most compatible hypothesis is local runoff from the hillside and the hill (dirt roads, a bike park, and the burned surface): Av. La Plaza Sur, with its carriageways at different elevations, channels it northward, and it enters the campus through a clearing among the trees. When these features are represented, water in the courtyard reaches 12–30 cm, within the range seen in the video, with rainfall of 15 to 35 mm/h, and La Plaza Sur is wet along 41–87 % of its length depending on intensity. Carving the main bike park trail does not change that route, but the area of the bike park and upper hillside does contribute water. This is a fitting hypothesis, based on contributed observations (video, Street View, and Google Earth); it requires field verification.
Where can I view the 3D map of the debris flow?
In the study's 3D viewer (aluvion.hirely.cl/visor/), which runs in a browser. It shows the mountains with the day's snow and isotherm, water flowing down the ravines, and simulated mud in the city, and allows scenarios to be compared using their exploratory compatibility index against observations, which does not measure predictive accuracy, in depth, velocity, dynamic pressure, and arrival time layers. This is a preliminary simulation, not an official survey of the affected area.
How much mud came down?
There is still no official measurement. In the scenarios tested, about 300,000 m³ of mixture allows the reach criterion to Padre Hurtado to be reproduced with the full city (all 42,018 buildings), and an independent estimate based on rainfall, reported saturation, and the 10.1 km² catchment gives a median of 295,000 m³ (5th and 95th percentiles: 181,000–468,000 m³). Under the calculation assumptions, the upper bound on the snowmelt contribution is less than 5 %: no snow was detected in the cloud-free part of the catchment in the Sentinel-2 image of 21 September. Both figures are preliminary. With fire severity measured by satellite, the balance estimates that without fire the pulse would have had about 62 % of that mud volume (between 51 and 76 %) and 43 % of the solids.
How reliable is this reconstruction?
It uses public sources and explicit assumptions—terrain, 42,018 buildings, measured rainfall, the isotherm, census data, and the official hydrographic network—and the hypothesis selected for the held-out test meets three of the four original acceptance criteria. Separately, the exploratory compatibility index summarizes those components on a continuous scale: 72/100 (streets with mud 77, affected facilities 75, reach to Padre Hurtado 100, and mud within the official affected area from IDE Chile 34); this is neither predictive accuracy nor independent validation. A test using held-out observations, defined in advance (49 officially affected facilities), yielded an unfavorable result: the hypothesis reaches 19 of 49 (39 %), because the model leaves parts of the official area without mud. In subsequent tests, which are for fitting rather than validation, allowing mud to enter houses and distinguishing the turbid water on campus and on La Plaza Sur from the mud in the main branch raises the verified index to 76–78/100; the integrated hypothesis obtains 79–82/100 depending on rainfall (82 with an assumed 35 mm/h, verified with a fine time step; 81 with measured rainfall; 79 with measured rainfall spread evenly). The index does not distinguish the effect of fire either: without it, the mud wets less area, but the index does not fall. Without porous houses, the model calculates around 1 m near Clínica UC Christus, where the source consulted reported about 1 m in the emergency department; with the integrated hypothesis it falls to 0.30 m, and no configuration reproduces both things at once. The model passed numerical checks, including a check that, when the time step is reduced by a factor of four, the 95th percentile of the difference in maximum depth in the city is less than 1 cm; that check does not validate the reconstruction. Only qualitative observations are available (no published discharge measurements or high-water marks), with a fine mesh on campus, the city simulation reproduces the water in the U. de los Andes courtyard and on La Plaza Sur, and the simulated mud extends beyond the official area, west of Padre Hurtado and toward Dipreca. This is neither an expert assessment nor an official report.
What is a debris flow?
It is a mixture of water and sediment—mud, stones, and debris—that flows down a ravine. When the solids concentration is approximately 20 % to 45 % by volume (mudflows; in debris flows it reaches 50–80 %), the mixture has a yield stress: it moves like a thick fluid and stops when the slope is no longer sufficient to drive it. Its density in mudflows is usually between 1,330 and 1,740 kg/m³, higher than that of water, which increases its capacity to carry material.
Who conducted this study?
This is an independent technical study by Lester Wood, with no official status, using public and open sources (FABDEM, Copernicus GLO-30, OpenStreetMap, Overture Maps, Censo 2017, the Geoportal de Chile hydrographic network, the IDE Chile emergency viewer, Dirección Meteorológica de Chile, the CR2 analysis using the DGA Río Molina station, Sentinel-2, and Open-Meteo) and a log of each source consulted. It is published under the CC BY-NC-SA 4.0 license; the citation format is in the “How to cite” section.
Reference
How to cite
Wood, L. (2026). Reconstrucción numérica del aluvión de San Carlos de Apoquindo, Las Condes (8 de octubre de 2026). https://aluvion.hirely.cl · CC BY-NC-SA 4.0