The review, led by Muhammad Shareef Shazil of the University of Campania Luigi Vanvitelli together with Thom Bogaard of Delft University of Technology and Roberto Greco of the University of Campania, analyzed 235 studies drawn from an initial pool of 470 records in the Scopus database. The authors found that previous major reviews, including landmark syntheses from 1988 and 2002 and more recent assessments, have concentrated overwhelmingly on the geomorphological factors of dam formation and failure. Yet the processes that actually kill these dams, overtopping, seepage-induced piping and slope failure, are all activated by water. Rainfall, snowmelt and glacier melt control how fast a dammed lake fills, how much pressure builds against the dam body, and how quickly internal erosion channels can carve their way through loose material.

The stakes are far from academic. The historical record contains chilling examples. In 1786, the catastrophic failure of a landslide dam on the Dadu River in China sent a flood wave roughly 1,400 kilometers downstream and killed an estimated 100,000 people. More recently, the 2010 Attabad landslide in northern Pakistan blocked the Hunza River, killed 20 people and submerged several villages under a growing lake. Classic analyses reported that about half of all landslide dams fail within ten days of formation, and inventories compiled in recent years show that roughly 7 percent collapse within a single day, about 11 percent within a month, while just over 20 percent survive longer than 20 years. But longevity is notoriously hard to predict: the Kummer dammed lake in South Tyrol, Italy, held for 370 years before it failed, destroying parts of the city of Merano.

The failure mechanisms themselves are well documented. Overtopping, in which rising lake water spills over the dam crest and erodes the downstream face, accounts for more than 90 percent of recorded failures. Piping occurs when water infiltrating through the porous dam body enlarges internal channels until the structure collapses, a process especially dangerous in fine-grained dams. Slope failure arises when rapid water level changes alter pore pressures within the dam, triggering sliding along internal failure surfaces. What these mechanisms share is a hydrological trigger: each one is set in motion by high water levels, extreme inflows, or sudden shifts in the water balance of the upstream catchment. A dam’s geometry and material composition matter, but without knowing how much water arrives, and when, the probability of failure remains deeply uncertain.

To illustrate the danger of ignoring hydrology, the review points to the Ram Creek landslide dam in New Zealand. Geomorphic indices, empirical formulas built from dam volume, catchment area and valley geometry, had classified the dam as stable. It failed after just a few hours of intense rain. Such indices, from early volume-to-catchment ratios proposed in the 1980s to the widely used dimensionless geomorphic index combining dam height, volume and catchment area, remain valuable first-order screening tools. But they are static. They cannot capture the dynamic, seasonal and extreme hydrological forcing that so often flips a dam from stable to unstable. The review notes one partial exception: the Hydromorphic Dam Stability Index, which integrates discharge and fluid density, represents one of the first attempts to bridge geomorphology and hydrology in stability assessment.

Geotechnical research has added its own layer of insight. Particle size distribution has emerged as a key control on dam behavior: dams built from coarse, blocky rock-avalanche debris tend to be highly permeable, allowing rapid seepage that can actually reduce overtopping risk, while dams of unconsolidated fine material are far less stable. Heterogeneous deposits containing weak layers of contrasting permeability are particularly prone to internal erosion. Newer quantitative tools, including the Material-based Morphological Index, the Dam Composition Index and a dimensionless parameter describing material distribution, attempt to fold these material properties into stability predictions. Discrete element simulations have shown that longer sliding paths, higher base friction and gentler valley slopes increase the internal heterogeneity of dam deposits. Yet even these sophisticated indices suffer from a chronic shortage of material data from real, heterogeneous natural dams.

The heart of the review is its argument that upstream catchment hydrology deserves equal billing. Inflow from rainfall, snowmelt, glacier melt and groundwater directly controls how quickly the lake fills and how much load presses on the dam. Evidence for this link comes from multiple directions. Discriminant and logistic regression analyses of Japanese landslide dams identified historical peak flow as a critical stability factor. In Italy, a potentially unstable dam survived a period of high river flows only because a pre-existing hydropower bypass diverted 80 percent of the peak discharge. Bayesian lifetime analysis showed that catchment area, together with dam shape, controls dam longevity, and machine learning studies have recently confirmed the importance of hydrological factors. In the Karakoram, the Attabad dam, initially judged highly vulnerable because of its size and enormous catchment, has now stood for more than 16 years, aided by an excavated spillway and by a hydrological regime in which precipitation falls mainly as snow and meltwater is released gradually, damping discharge peaks and limiting overtopping potential.

Because most landslide dams form in remote, high mountain regions where gauges are sparse or absent, the review devotes considerable attention to how remote sensing and modelling can fill the data gap. Satellite imagery can track the surface area and volume of dammed lakes over time; researchers used Sentinel-2 images processed in Google Earth Engine to detect and monitor new lakes formed after the 2011 Kaikōura earthquake in New Zealand. Radar technologies such as Sentinel-1 SAR and the SWOT satellite altimetry mission promise measurements of water surface elevation and even discharge in ungauged catchments. Reanalysis products and snow and glacier datasets supply the meteorological inputs needed for water balance calculations. On the modelling side, parsimonious lumped models such as the Budyko framework and the GR6J rainfall-runoff model can simulate catchment response where data is limited, while general circulation models allow exploration of future climate scenarios, though uncertainties from coarse spatial resolution and complex topography remain a challenge.

The review also warns that climate change is poised to reshape the entire hazard landscape. Warming temperatures accelerate snowmelt and glacier retreat, shifting the timing of inflows so that lake level peaks arrive sooner after dam formation and increasing the risk of early failure. More intense rainfall extremes raise both landslide occurrence and overtopping probability. Glacier retreat exposes new valley walls and debris sources, creating conditions for future dams, while meltwater can accelerate lake filling or destabilize existing barriers through increased seepage. Even vegetation dynamics play a role: loss of cover through drought or wildfire enhances runoff generation, altering the volume and timing of water reaching dammed lakes. Because climate change acts mainly through hydrological processes, the authors argue, it amplifies the very mechanisms that existing stability assessments most often overlook.

The authors close with a set of recommendations aimed at turning hydrology from an afterthought into a core component of landslide dam hazard assessment. They call for continuous monitoring of lake area, volume, inflows and outflows; systematic use of remote sensing products for evapotranspiration, snowmelt, runoff and terrestrial water storage; hybrid approaches that combine physically based hydrological models with machine learning trained on satellite, reanalysis, geotechnical and geomorphological data; dedicated field campaigns to measure discharge, sediment dynamics, infiltration rates and dam composition; and, crucially, the expansion of global landslide dam inventories to include hydrological information. No single hydrological variable, the review stresses, can predict dam stability across all environments, but integrating multiple dynamic variables offers the clearest path toward reliable early warning systems. For the communities living downstream of mountains that hold thousands of these natural time bombs, that shift in perspective could not come soon enough.

Subject of Research: Hydrological controls on the formation, stability and failure of landslide dams

Article Title: Hydrological perspective of landslide dams: a review

Article References: Shazil, M. S., Bogaard, T., & Greco, R. (2026). Hydrological perspective of landslide dams: a review. Natural Hazards, 122(18), Article 620. https://doi.org/10.1007/s11069-026-08393-2

Image Credits: AI Generated

DOI: 10.1007/s11069-026-08393-2

Keywords: landslide dams, hydrology, dam stability, overtopping, seepage, remote sensing, dammed lakes, outburst floods, climate change, snowmelt, catchment hydrology, natural hazards