Geology topics

Benjamin B. Mirus

Publications and source records attributed to Benjamin B. Mirus.

13 recordsLinked to original sources

USGS2026

SlideDetect: Spatio-temporal landslide detection using a three-dimensional convolutional neural network

Landslides pose a serious and ongoing threat to both human lives and infrastructure worldwide; therefore, it is of interest to predict where and when landslides are likely to occur. Advances in machine learning techniques have spurred numerous studies aimed at estimating relative landslide propensity, but are limited to spatial (as opposed to temporal) prediction due to the sparsity of landslide timing data. We address this data gap by training SlideDetect, a 3-dimensional convolutional neural network (3D CNN), to identify landslides based on their spatial and temporal occurrence within multitemporal image stacks. We use an inventory of landsides triggered by the 2018 Hokkaido earthquake and two years of monthly composite optical imagery spanning this event. The model can identify not only landslide location but also landslide date with an area under the precision-recall curve (PR-AUC) of 0.84. We further present a new standard for presenting PR curve results that explicitly compares model performance at different confidence thresholds, allowing for clearer model evaluation and comparison. Our new approach to constraining landslide timing paired with this more consistent and objective method for evaluating model performance shows considerable promise, and with further application and testing, SlideDetect could enhance the data availability and tools needed to advance landslide hazard and risk assessments.

JGR Machine Learning and Computation↗

USGS2026

Asynchronous landslide seasonality across the United States

Mid-range landslide outlooks can facilitate weather-related landslide preparedness and disaster response planning, but seasonal landslide activity remains poorly quantified at continental scales. Leveraging >55,000 reported landslides from across the United States (U.S.), we used circular statistics to quantify landslide seasonality in 67 National Weather Service County Warning Areas (CWAs). We found regional differences in landslide season timing and duration, with transitions between domains variably corresponding to climate class or river basin. We assessed differences in seasonality by movement type for slides, flows, and falls, detecting apparent, but uncertain, differences between slide and fall seasonalities in 27 of 35 (77%) of CWAs with both types reported. In the Pacific Northwest, where long records exist, we found a credible shift toward a later mean landslide season in western Washington from 1990 to 2020, but no trend in western Oregon. Our results can provide emergency planners a resource to assess seasonal landslide probability nationwide.

Geophysical Research Letters↗

USGS2026

From landslide susceptibility to risk assessment in the conterminous U.S.

Understanding the spatial distribution of landslide prone-areas and what consequences they may have is important for risk management and land-use planning. In the United States, although landslides occur in every state, a comprehensive landslide risk assessment is still missing. Existing efforts, such as the Federal Emergency Management Agency (FEMA)’s National Risk Index, rely on aggregated products and coarse cartographic units, limiting their geomorphological and practical accuracy. In this study, we present a methodological advance for landslide risk assessment across large areas with incomplete and sparse data. We apply our procedures to the conterminous United States by integrating geomorphologically meaningful partitions and spatial and temporal probability data-driven models. Landslide susceptibility is estimated using a Generalized Additive Mixed Model incorporating a bias capture/correction scheme to account for inventory inaccuracies (reference Area Under the Curve = 0.75). The exceedance probabilities of landslide occurrence are defined for three temporal scenarios (2, 5, and 10 year). Then, we explore the associated potential economic consequences for human settlements and agricultural areas. The findings indicate that the spatial variability of risk is primarily controlled by exposure rather than by susceptibility/hazard alone. The mean risk increases by ∼170% from the 2-year to the 10-year scenario. Beyond its quantitative outcomes, this study offers a blueprint for continental or sub-continental scale landslide risk assessments, demonstrating both the opportunities and current limitations.

Engineering Geology↗

USGS2025

Insights into widespread landsliding in southern Appalachia from Hurricane Helene

Between 23 and 27 September 2024, antecedent rain followed by Hurricane Helene produced one of the most damaging weather events in southern Appalachia history. The back-to-back storm events resulted in a maximum cumulative rainfall of 848 mm and hurricane-force wind gusts over 170 km/h in western North Carolina, eastern Tennessee, and southwestern Virginia. The resulting regional flooding, landslides, and tree blowdown caused over 100 fatalities, damaged or destroyed critical infrastructure and thousands of structures, and severed connectivity across the region. Over the next several weeks, a multi-agency landslide response produced a rapid hazard assessment and mapped 2217 landslides, 55% of which damaged infrastructure or property. Orographic uplift enhanced rainfall, resulting in concentrated landsliding along the ~250 km swath of the Blue Ridge escarpment in western North Carolina. Landslides initiated predominantly on windward-facing (southeast-facing) slopes, and localized clustering of initiation points indicated a strong influence of hillslope-scale meteorological and geomorphic factors. Many shallow landslides mobilized into larger, highly mobile, and damaging debris flows that graded into floods. Here, we put our preliminary observations in the context of historical storm-driven landslide events and open new avenues for investigating the nature and extent of landslides and their effects in southern Appalachia and similar environments.

Georgia, North Carolina, South Carolina, Tennessee↗

USGS2025

The influence of postfire root strength decay on shallow landslide susceptibility in western Oregon

Severe wildfires can increase the potential for postfire landslides, partly due to the loss of vegetation and root reinforcement. This study investigates the multi-year dynamics of root strength following wildfire in coniferous forests of western Oregon and evaluates implications for postfire slope stability. Root tensile strength was measured through laboratory testing of 969 roots collected from 26 test pits over four postfire timeframes: unburned conditions, 4–12 months postfire (Cedar Creek, 2022), 24–35 months postfire (Holiday Farm, 2020), and 59–67 months postfire (Eagle Creek, 2017). Variables analyzed include root diameter, tensile thread strength, stiffness, progressive tensile strength with displacement, time-dependent strength and ductility, root area ratio, and cohesion. Results indicate a 50% reduction in root strength after wildfire, with the lowest values occurring around 4 years postfire. Projections indicate it could take 10–22 years following fire for root strength to recover to 70% and 90% of the prefire original root strength, respectively. This timeline indicates a multi-year window of vulnerability for shallow landslides that reaches a maximum approximately 4 years following fire. Brittle failure behavior was observed in burned root systems compared to unburned root systems, signaling a more abrupt loss of strength at yield. Simple slope stability analyses show that reduced root strength can lead to instability on more gentle slopes and under less saturation in comparison to unburned conditions. These findings highlight the critical role of root reinforcement in postfire slope stability and the long-term implications of wildfire disturbance on landslide susceptibility.

Oregon↗

USGS2025

Constraining landslide frequency across the United States to inform county-level risk reduction

Informative landslide hazard estimates are needed to support landslide mitigation strategies to reduce landslide risk across the United States. Whereas existing national-scale landslide susceptibility products assess where landslides are likely to occur, they do not address how often , which is a critical element of landslide hazard and risk assessments. In particular, the U.S. Federal Emergency Management Agency's National Risk Index (NRI) requires landslide frequency estimates to inform expected annual loss estimates. We present county-level landslide frequency (landslides per area per year) estimates for the 50 US states. We applied Bayesian negative binomial regression to estimate both the expected (average) reported landslide frequency and full distribution of annual landslide counts for each county. We compared a suite of models that used combinations of landslide-susceptible area, probability of potentially triggering earthquakes, frequency of potentially triggering precipitation, and ecological region as predictors. We trained our models with landslide inventory data from counties with the most comprehensive records available nationwide and used zero-inflated negative binomial distributions as an incompleteness model to correct for temporal reporting gaps. We selected a preferred frequency model to inform the NRI based on information criteria and physically plausible parameter estimates. The model showed that average annual reported landslide frequencies vary by 5 orders of magnitude across US counties, ranging from 0.002 (0.00015–0.05) landslides 1000 km −2 yr −1 in Kusilvak Census Area, Alaska, to 29 (19–46) landslides 1000 km −2 yr −1 in Lake County, California, reflecting the country's strong variations in landslide susceptibility, earthquake probability, and other factors for which ecological region serves as a proxy. Counties with estimated frequencies in the top 20 % of all counties are predominately along the West Coast of the continental United States, in mountainous regions of the Pacific Northwest and Intermountain West, in locally steep or earthquake-prone regions of the Midwest and Southeast, along the Appalachians, in southern and southeastern Alaska, and on some Hawaiian islands. By examining the number of landslides predicted in 99th percentile years for each county, we identified that 26 % of US counties likely have potential for widespread landsliding with more than 10 landslides 1000 km −2 yr −1 , even when such large events have not been reported in the training data for that county. Overall, our results better represent the range of possible landslide frequencies and spatial variations than previous national-scale estimates reported in the NRI, and our approach can inform other risk-reduction and loss-mitigation efforts across the United States and globally.

Natural Hazards and Earth System Sciences↗

USGS2025

Dynamic feedbacks between river meandering and landsliding in northwestern Washington glacial terraces

Landsliding in river valleys poses unique risks for cascading hazards and can damage infrastructure and cause fatalities. In postglacial valleys, many landslides are posited to occur in relation to lateral river erosion, but the dynamics of fluvial-hillslope interactions are not well understood. Here, we investigate a section of the Nooksack River in western Washington State where the channel is flanked by landslide-prone glacial terraces similar to those that failed in the 2014 State Route 530 “Oso” landslide. We map 216 landslides through time across 17 aerial imagery data sets (1933–2022) and analyze them in relation to river meandering and curvature. We observe dynamic feedbacks between lateral river meandering and valley-adjacent landsliding. Terrace lateral retreat rates of up to 25 m/year owing to combined fluvial erosion and slope failure occur on pinned, outer meander bends immediately downstream from peaks in river curvature (>0.0075 1/m); these locations are predisposed to both shallow and deep-seated landslides. Deep-seated landslides extending 17%–32% of the active valley width into the floodplain can displace the river away from the floodplain margin and change the channel planform. River-displacing landslides relocate meanders up- or downstream, thereby conditioning the location of subsequent landslides. This conceptual model of coupled landslide-driven meander displacement and valley-adjacent landsliding is exemplified across western Washington river systems. The distance between up- and downstream valley-adjacent landsliding scales with valley width, meander wavelength, and terrace height. Our results can advance our understanding of the river-hillslope interface in landscape evolution and can be used to inform hazard management in river corridors.

Washington↗

USGS2025

Invited perspectives: Integrating hydrologic information into the next generation of landslide early warning systems

Although rainfall-triggered landslides are initiated by subsurface hydro-mechanical processes related to the loading, weakening, and eventual failure of slope materials, most landslide early warning systems (LEWS) have relied solely on rainfall event information. In previous decades, several studies demonstrated the value of integrating proxies for subsurface hydrologic information to improve rainfall-based forecasting of shallow landslides. More recently, broader access to commercial sensors and telemetry for real-time data transmission has invigorated new research into hydrometeorological thresholds for LEWS. Given the increasing number of studies across the globe using hydrologic monitoring, mathematical modeling, or both in combination, it is now possible to make some insights into the advantages versus limitations of this approach. The extensive progress demonstrates the value of in situ hydrologic information for reducing both failed and false alarms, through the ability to characterize infiltration during, as well as the drainage and drying processes between major storm events. There are also some areas for caution surrounding the long-term sustainability of subsurface monitoring in landslide-prone terrain, as well as unresolved questions in hillslope hydrologic modeling, which relies heavily on the assumptions of diffuse flow and vertical infiltration but often ignores preferential flow and lateral drainage. Here, we share a collective perspective based on our previous collaborative work across Europe, North America, Africa, and Asia to discuss these challenges and provide some guidelines for integrating knowledge of hydrology and climate into the next generation of LEWS. We propose that the greatest opportunity for improvement is through a measure-and-model approach to develop an understanding of landslide hydro-climatology that accounts for local controls on subsurface storage dynamics. Additionally, new efforts focused on the subsurface hydrology are complementary to existing rainfall-based methods, so leveraging these with near-term precipitation forecasts is a priority for increasing lead times.

Alaska↗

USGS2025

Overcoming the data limitations in landslide susceptibility modelling

Data-driven models widely used for assessing landslide susceptibility are severely limited by the landslide and environmental data needed to create them. They rely on inventories of past landslide locations, which are difficult to collect and often nonrepresentative. Furthermore, susceptibility maps are most needed in regions without the means to assemble an inventory. To overcome these challenges, we develop a method for assessing shallow landslide susceptibility based on a probabilistic morphometric analysis of the landscape’s topography, rather than the characteristics of landslides. The model assumes that hillslopes with higher relief and gradient compared to the surrounding landscape are more prone to landslides. We demonstrate the superior performance of this approach over contrasting data-driven models across the northwestern United States. As our morphometric model only requires elevation data, it overcomes the major limitations of data-driven models and facilitates the creation of effective susceptibility models in areas where it was previously unfeasible.

Oregon, Washington↗

USGS2024

Parsimonious high-resolution landslide susceptibility modeling at continental scales

Landslide susceptibility maps are fundamental tools for risk reduction, but the coarse resolution of current continental-scale models is insufficient for local application. Complex relations between topographic and environmental attributes characterizing landslide susceptibility at local scales are not transferrable across areas without landslide data. Existing maps with multiple susceptibility classifications under-represent landslide potential in moderate and gently sloping terrain. We leverage an extensive landslide database ( N = 613,724), a high-resolution digital elevation model (10-m), and high-performance computing resources, to develop a new nationwide susceptibility map for the contiguous United States, Hawaii, Alaska, and Puerto Rico. We calculate four alternative linear and nonlinear thresholds of topographic slope and relief using an objective split-sample calibration. We down-sample our results to a 90-m grid to account for uncertainty in the digital elevation model and landslide position, and evaluate these thresholds' ability to differentiate areas of greater susceptibility. The less conservative nonlinear model optimally balances our priorities of capturing observed landslides (99%) while minimizing area covered by susceptible terrain (43%). Independent evaluation with four statewide landslide inventories ( N = 172,367) reinforces our model selection but highlights spatially variable performance. Therefore, we propose a novel approach to susceptibility classification using the concentration of landslide-prone terrain within each down-sampled grid. While landslides are possible within any cells containing susceptible terrain, those with the highest concentration capture the majority of observed landslides. Our new map characterizes landside susceptibility more consistently than prior models; our transparent classification approach also provides flexibility for accommodating different tolerances in risk reduction measures.

AGU Advances↗

USGS2024

RegionGrow3D: A deterministic analysis for characterizing discrete three-dimensional landslide source areas on a regional scale

Regional-scale characterization of shallow landslide hazards is important for reducing their destructive impact on society. These hazards are commonly characterized by (a) their location and likelihood using susceptibility maps, (b) landslide size and frequency using geomorphic scaling laws, and (c) the magnitude of disturbance required to cause landslides using initiation thresholds. Typically, this is accomplished through the use of inventories documenting the locations and triggering conditions of previous landslides. In the absence of comprehensive landslide inventories, physics-based slope stability models can be used to estimate landslide initiation potential and provide plausible distributions of landslide characteristics for a range of environmental and forcing conditions. However, these models are sometimes limited in their ability to capture key mechanisms tied to discrete three-dimensional (3D) landslide mechanics while possessing the computational efficiency required for broad-scale application. In this study, the RegionGrow3D (RG3D) model is developed to broadly simulate the area, volume, and location of landslides on a regional scale (≥1,000 km 2 ) using 3D, limit-equilibrium (LE)-based slope stability modeling. Furthermore, RG3D is incorporated into a susceptibility framework that quantifies landsliding uncertainty using a distribution of soil shear strengths and their associated probabilities, back-calculated from inventoried landslides using 3D LE-based landslide forensics. This framework is used to evaluate the influence of uncertainty tied to shear strength, rainfall scenarios, and antecedent soil moisture on potential landsliding and rainfall thresholds over a large region of the Oregon Coast Range, USA.

Journal of Geophysical Research: Earth Surface↗

USGS2024

Constraining mean landslide occurrence rates for non-temporal landslide inventories using high-resolution elevation data

Constraining landslide occurrence rates can help to generate landslide hazard models that predict the spatial and temporal occurrence of landslides. However, most landslide inventories do not include any temporal data due to the difficulties of dating landslide deposits. Here we introduce a method for estimating the mean landslide occurrence rate of deep-seated rotational and translational slides derived solely from high-resolution (≤3 m) elevation data and globally available estimates of the diffusion coefficient for sediment flux. The method applies a linear diffusion model to the roughest landslide deposits until they reach a representative non-landslide roughness distribution. This estimates the time for a landslide deposit to be unrecognizable in high-resolution digital elevation data, which we term the mean lifetime of the landslide. Using the mean lifetime and number of landslides within an area of interest, we can estimate the mean occurrence rate of landslides over that domain. We validate this approach using a comprehensive temporal inventory of landslides in western Oregon created using age-roughness curves that are calibrated with high-resolution elevation data and radiocarbon data. We find good agreement between our diffusion method and the existing age-roughness-derived estimates, producing mean lifetimes of 4500 and 5200 years (4% difference), respectively. Hazard maps produced using the two methodologies generally agree, with the maximum differences in landslide probability reaching 0.1. Due to the relative abundance of high-resolution elevation data compared with age-dated landslides, our method could help constrain landslide occurrence rates in areas previously considered unfeasible.

Oregon↗

USGS2024

A benchmark dataset and workflow for landslide susceptibility zonation

Landslide susceptibility shows the spatial likelihood of landslide occurrence in a specific geographical area and is a relevant tool for mitigating the impact of landslides worldwide. As such, it is the subject of countless scientific studies. Many methods exist for generating a susceptibility map, mostly falling under the definition of statistical or machine learning. These models try to solve a classification problem: given a collection of spatial variables, and their combination associated with landslide presence or absence, a model should be trained, tested to reproduce the target outcome, and eventually applied to unseen data. Contrary to many fields of science that use machine learning for specific tasks, no reference data exist to assess the performance of a given method for landslide susceptibility. Here, we propose a benchmark dataset consisting of 7360 slope units encompassing an area of about 4,100 km 2 "> 4,100 km 2 in Central Italy. Using the dataset, we tried to answer two open questions in landslide research: (1) what effect does the human variability have in creating susceptibility models; (2) how can we develop a reproducible workflow for allowing meaningful model comparisons within the landslide susceptibility research community. With these questions in mind, we released a preliminary version of the dataset, along with a “call for collaboration,” aimed at collecting different calculations using the proposed data, and leaving the freedom of implementation to the respondents. Contributions were different in many respects, including classification methods, use of predictors, implementation of training/validation, and performance assessment. That feedback suggested refining the initial dataset, and constraining the implementation workflow. This resulted in a final benchmark dataset and landslide susceptibility maps obtained with many classification methods. Values of area under the receiver operating characteristic curve obtained with the final benchmark dataset were rather similar, as an effect of constraints on training, cross–validation, and use of data. Brier score results show larger variability, instead, ascribed to different model predictive abilities. Correlation plots show similarities between results of different methods applied by the same group, ascribed to a residual implementation dependence. We stress that the experiment did not intend to select the “best” method but only to establish a first benchmark dataset and workflow, that may be useful as a standard reference for calculations by other scholars. The experiment, to our knowledge, is the first of its kind for landslide susceptibility modeling. The data and workflow presented here comparatively assess the performance of independent methods for landslide susceptibility and we suggest the benchmark approach as a best practice for quantitative research in geosciences.

Earth-Science Reviews↗