This study evaluated the influence of overwash processes on the dynamics of the coastal system in southern Santa Catarina, Brazil. The analysis was conducted at two temporal scales: monthly and interannual, to investigate the controlling factors in the formation of washover fans within the sand spit barrier (2017–2023) and interdecadal, to understand the morphological behavior of the shoreline and the coastal system (1938–2021). The methodology involved the integration of remote sensing data, aerial photographs, tide gauge records, ocean-climatological data, and analyses of morphometric indicators. The study area was divided into five sectors based on foredune segmentation and the areas with the highest recurrence of overwash fan events. A total of 93 overwash events were identified, with a higher concentration during the autumn and winter months. The formation of washover fans was associated with a combination of storm surges, storm waves, and the presence of foredunes on the sand spit barrier during the passage of low-pressure centers (cyclones). Sectors 3, 4, and 5 exhibited greater vulnerability to overwash events, while Sector 1 was more affected by anthropogenic activities. On the interdecadal scale, the shoreline presented three evolutionary phases: a regressive phase (1938–1978), followed by two transgressive phases (1978–2005 and 2005–2021), with an intensification of the coastal system retreat after 2005. This study contributes to the understanding of how the high frequency of overwash events has driven a transgressive behavior across most of the Araranguá coastal system over the past 16 years, contrasting with the regressive pattern observed in the Holocene system as a whole.
This study develops a peak-oriented deep learning framework for one-day-ahead daily streamflow forecasting in the Balikhlouchai River watershed, Iran, using daily observations of streamflow, rainfall, and mean temperature from 1999 to 2024. A range of baseline and advanced deep learning architectures was evaluated, including CNN–RNN hybrids, U-shaped encoder–decoder models, attention-based models, and Transformer-based designs. Peak flow conditions were characterized using 95th-percentile upper-tail definitions, and model training followed a two-stage loss strategy, consisting of Huber pre-training followed by peak-weighted MSE fine-tuning, to emphasize peak flow errors while maintaining overall predictive skill. Model performance was assessed using general accuracy metrics together with peak-focused diagnostics that explicitly evaluate peak magnitude and event-detection reliability. The top-performing models were further examined through uncertainty quantification, separating epistemic uncertainty using Monte Carlo Dropout and aleatory robustness using meteorological input perturbations, and were interpreted using peak-conditioned SHAP analyses. Final model ranking was conducted using multi-criteria decision analysis (MCDA), integrating deterministic performance, peak-detection reliability, uncertainty behavior, and interpretability. Results show that U-shaped CNN–RNN architectures provide the most balanced performance for peak flow forecasting, particularly in terms of operational reliability. U-CNN-BiGRU achieved the best integrated rank, with an MCDA score of 0.69, reflecting balanced accuracy, low false-alarm tendency, and favorable epistemic reliability–sharpness behavior. U-CNN-BiLSTM showed the highest robustness under noisy meteorological inputs among the meteorology-enabled models. SHAP results indicate that short-lag antecedent streamflow, especially Q(t − 1), dominates one-day-ahead predictions, while rainfall and temperature provide secondary, event-dependent refinement rather than primary control. Overall, the proposed framework supports balanced, uncertainty-aware, and interpretable peak flow forecasting and can serve as a predictive component within data-driven early-warning systems.
Recent years have highlighted an increasing vulnerability to wildfires in Europe driven by climate, land use change, and socio-economic factors. Under future projected warming and drying patterns, Mediterranean areas emerge as one of the most wildfire-sensitive regions. This study combines high-resolution climate projections from three CMIP6 models (ACC, ECE, MPI) under three SSP scenarios with a wildfire modeling framework to assess wildfire danger and hazard for historical baseline (1985–2014) and a near-future period (2035–2064) in Sardinia, Italy. Climate variables were statistically downscaled to 1 km2 resolution and used to calculate Fire Weather Index (FWI) and Daily Severity Rating (DSR). Then, using the Minimum Travel Time algorithm and probabilistic fire simulations, we generated spatially explicit estimates of burn probability, fire intensity and crown fire probability at 150 m resolution. Despite model-dependent differences in magnitude, all scenarios consistently point to warmer, drier, and more fire-prone conditions. Results indicate a systematic shift toward more severe fire danger under future scenarios, with a marked increase in the days reaching very high FWI and DSR values, particularly from May to October, which evidences nearly doubled frequencies under SSP5-8.5 scenarios. Similarly, simulated wildfire hazard outputs show significant increases, with hotspots expanding towards western and inland Sardinia, particularly with ECE (+ 250
Rock avalanches often lead to severe casualties and economic losses, making real-time monitoring and dynamic analysis of paramount importance for disaster mitigation. Improving the characterization of event-scale dynamic behavior and precursory deformation is essential for effective disaster mitigation and early warning. With the increasing availability of video recordings from monitoring systems and public sources, video-based analysis has become a valuable complementary data source for investigating landslide dynamics. In this study, a joint analytical framework that leverages video-based computer vision analysis and seismic observations to characterize the dynamic evolution of large rock avalanches is proposed. The framework primarily employs optical-flow based motion analysis to extract image-plane deformation and kinematic patterns from video footage, while seismic signal analysis is used as an independent reference for temporal alignment and validation. The application of the framework to a representative large rock avalanche event in Southwest China demonstrates its practicality. Velocity distribution and motion descriptions at any moment during the avalanche can be obtained through the proposed four-stage architecture, which integrates data acquisition, preprocessing, joint analysis, and result visualization. The results demonstrate that the framework can provide complementary insights to those obtained from seismic signal analysis alone. This study represents a case-oriented and exploratory attempt to integrate visual and seismic information for rock avalanche analysis and highlights the potential of video-based computer vision techniques to enhance the interpretation of landslide dynamics and support future monitoring and early-warning research.
The concurrent rise of global sea levels and coastal population growth is intensifying exposure to climate-driven hazards. In Ireland, where most major urban centres are on, or connected to the coast, the nature and severity of hazard impacts vary across diverse coastal landscapes. A spatially explicit assessment of coastal exposure is therefore critical for effective risk management and climate adaptation planning. This study presents a national-scale Coastal Exposure Index for Ireland, developed using the Coastal Vulnerability Model within the InVEST® software suite. The index integrates physical and climatic variables, including 30 years of wind and wave records, sea-level rise rates, and storm surge levels. Coastal habitats, including dunes, machair, saltmarsh and seagrasses, were directly represented to assess their potentially protective role. Exposure levels of coastal populations and cultural heritage sites were also evaluated. A critical appraisal of the InVEST® approach was undertaken, including a sensitivity analysis to assess how model modifications affect outcomes. Approximately 8575 km of coastline was assessed at an alongshore spacing of 250 m, with exposure reported on a relative scale. Results reveal that 38
Forest fires pose a significant environmental hazard in southwestern France, where increasing fuel accumulation, human activities, and climatic extremes have intensified fire occurrences in recent decades. This study developed a high-resolution forest fire vulnerability map for the Gironde department by integrating remote sensing, GIS-based modeling, and the analytical hierarchy process (AHP). A comprehensive geodatabase comprising twelve conditioning factors related to vegetation and fuel characteristics(e.g., forest fuel type and land cover), climate and weather variability, topographic constraints, and human influence was developed to support the vulnerability assessment. A 12 × 12 AHP pairwise comparison matrix generated a consistent weighting scheme (CR = 0.046), and the resulting criteria weights were applied through a weighted sum model to derive the forest fire vulnerability index (FFVI). A key contribution of this study is the integration of multi-source geospatial data with a validated multi-criteria decision-making framework to produce a high-resolution forest fire vulnerability assessment, an approach not previously applied to the Gironde region. The model demonstrated strong predictive performance when validated using 699 active fire detection points obtained from the fire information for resource management system (FIRMS). The ROC analysis produced an AUC of 0.721, indicating good discriminatory performance, and 87
Landslide dams, created when rapid mass movements block river channels, are strongly influenced by hydrological conditions of the upstream catchment. The formation and long-term stability of landslide dams are closely linked to hydrological factors such as streamflow, lake area, water level and precipitation. Conversely, the formation of these dams changes upstream and downstream hydrological regimes such as water storage, sediment transport and flooding risk. Since the previous reviews by Costa and Schuster, (1988) and Korup, (2002), several review articles have focused on the formation and failure mechanism of landslide dams (e.g., Fan et al. 2020; Zheng et al. 2021). These reviews have mainly focused on geomorphological factors of dams and their formation. However, the hydrological processes around and within the landslide dam are crucial for the behavior of landslide dams, the downstream hydrology and potential flash flood risks. Therefore, there is an urgent need to study hydrological processes upstream and within landslide dams in order to better predict landslide dam stability, up- and downstream consequences. Streamflow (inflow and outflow), precipitation and snowmelt, water level, dammed lake area, lake volume, water balance, surface runoff, erosion, sedimentation, and water storage capacity are key factors for assessing the behavior of landslide dams. This review analyzes the current state of knowledge on hydrological processes influencing the evolution, and failure of landslide dams. It further examines the role of remote sensing and modelling techniques in assessing hydrological factors where field data are scarce.
Rapid urban growth can intensify flood hazard when land-cover change alters runoff production on slopes and concentrates drainage toward occupied lowlands. We assessed this process in the Port-au-Prince Arrondissement, Haiti, by combining a 30 m digital elevation model and geomorphon classification with Landsat 8 land-use and land-cover maps for 2015, 2020, and 2025, a 10 m Sentinel-2 map for 2025, weighted flood-susceptibility modelling, and five daily precipitation products for 1983–2024. The 1983–2013 interval was used only to estimate the fixed R99p percentile threshold; every trend test covered 1983–2024. Landsat 8 classifications yielded overall accuracies of 98.44, 95.92, and 94.13
The Hindu Kush region of eastern Afghanistan is an active intracontinental deformation zone where ongoing India-Eurasia convergence is accommodated through interacting strike-slip, reverse and transpressional fault systems. On 31st August 2025, an Mw 6.0 earthquake occurred within the Kunar-Spin Ghar fault system, causing widespread infrastructure damage, landslides and surface deformation across a rugged and poorly accessible mountainous region. Field reconnaissance at accessible sites documented structural damage, ground cracks, slope failures, rock avalanches and newly emerged springs along the Kunar River valley. To characterize the earthquake-period deformation and examine its spatial relationship with the mapped fault network, we integrated field observations with Persistent Scatterer Interferometric Synthetic Aperture Radar (PSInSAR) time-series analysis of Sentinel-1 A data acquired between April and October 2025. The ascending and descending line-of-sight velocity fields reveal a laterally continuous deformation pattern extending for several tens of kilometres along strike. Decomposition into east-west and vertical components indicates dominant horizontal motion accompanied by broader and spatially variable uplift-subsidence patterns. Swath-profile analysis shows that the strongest east-west velocity gradients occur south of the mapped Kunar Fault and within or adjacent to the Spin Ghar mountain-front structural corridor. These observations indicate that the earthquake-period deformation signal was distributed across multiple interacting structures rather than confined to a single mapped fault trace. The observed pattern is consistent with distributed transpressional deformation within the broader Kunar-Spin Ghar system and suggests that seismic-hazard assessments based only on mapped fault traces may underestimate the width of the active deformation corridor and the spatial extent of associated ground deformation and secondary hazards.
This study proposes a comprehensive multi-hazard risk assessment framework for the Low-Elevation Coastal Zone (LECZ) of China, integrating four key dimensions: hazard, exposure, vulnerability, and adaptive capacity. By employing spatiotemporal models and regression analysis methods, the study identifies major risk hotspots and analyzes the driving factors behind the evolution of multi-hazard risk from 2000 to 2020. The results show that, overall, multi-hazard risk has decreased, with the average risk index dropping from 0.51 in 2000 to 0.38 in 2020, a reduction of 25.3
Landslide detection and mapping are essential for disaster risk management, particularly in regions with complex terrain and persistent cloud cover. This study presents a deep learning approach for automatic landslide detection in the Chilean Patagonia using Synthetic Aperture Radar (SAR) imagery from Sentinel-1 combined with topographic derivatives. We developed a modified U-Net architecture trained on the Patagonian Andes Landslide Inventory (PALDI), comprising 722 manually delineated landslides. The model uses eight input channels: SAR polarizations (VV, VH), incidence angle, and five topographic features derived from the Copernicus DEM (elevation, slope, aspect, Topographic Position Index, and Terrain Ruggedness Index). We present a spatially-validated benchmark of SAR-based deep learning landslide detection, reporting performance under both conventional random cross-validation and spatially-independent block cross-validation. Within the region of training, the model attains an F1-score of 58.2
Climatic and environmental stressors influence urban traffic accidents, yet their combined and spatial effects remain incompletely understood. While previous studies have explored the relationships between climatic factors and urban transportation, only a limited number have focused specifically on urban traffic accidents, and most of these have analyzed single indicators. To address this gap, the present study proposes a comprehensive, data-driven framework for both linear and nonlinear analysis and spatial mapping of the impact of heat stress on urban traffic accident risk in Austin, TX. To this end, six machine learning algorithms were evaluated to assess the influence of climatic stressors derived from remote sensing data on urban traffic accidents. Among these, the random forest algorithm achieved the highest prediction accuracy. Interpretation using SHAP analysis revealed that NDVI acts as the strongest protective factor, while relative humidity showed a complex effect. Additionally, high terrain slope, elevated land surface temperature, aerosol concentration, and wind strength are other factors that increase vulnerability. Furthermore, to quantify urban traffic vulnerability, two complementary indices, accident risk ratio (ARR) and the climate stress index (CSI), were combined to create a novel Climate-Stressed Urban Traffic Climate Vulnerability Index (UTCVI), enabling the simultaneous assessment of predicted and observed crash risk. Spatial analysis of UTCVI indicated that 65
Hurricanes are a major hydrological disturbance in the southern United States (US), yet their impacts on different streamflow components and their linkage to forest damage across different forest types remain poorly understood. This study examines the impacts of Hurricane Michael (2018) on streamflow dynamics across multiple watersheds (n = 21) in South Carolina, a state located in the southeastern US. Streamflow was separated into its components using a digital filter, and recovery times were estimated using anomaly-based metrics. In addition, we analyzed whether variations in forest damage among forest types are associated with changes in streamflow components across watersheds. Our results show a statistically significant difference in recovery behavior between the two components. Quickflow recovered rapidly (median of 8 days), and baseflow exhibited a substantially delayed recovery (median of 114 days) post-hurricane. On average, the post-hurricane quickflow slope increased by 3.5 times relative to pre-hurricane conditions, compared to 2.5 times for baseflow, with greater variability observed in quickflow (SD = 1.69 vs. 1.21). Finally, watersheds with significant changes in quickflow and baseflow also had significantly higher forest damage in the Oak-Gum-Cypress and Elm-Ash-Cottonwood forests (p < 0.05). This study shows that forest cover, particularly bottomland forests, plays a critical role in buffering streamflow responses. The findings of this research will help water resources managers to understand watershed-specific streamflow responses following hurricanes and other severe storms.
Assessing the vulnerability of shallow foundation masonry buildings to landslide disasters is critical for risk evaluation. Current assessments employ a combination of qualitative and quantitative methods based on various building attribute indices, resulting in diverse outcomes. To improve this process, shallow foundation masonry buildings in the tension region of the Sifangbei landslide were examined. In such buildings, landslide-induced tensile deformation commonly first affects the shallow foundation and then propagates to the superstructure, undermining overall stability. Physical experiments using scaled-down models were conducted, applying horizontal tensile forces to investigate the relationship between tensile strength (Rm), disaster load (F), and shallow foundation damage (D). Empirical formulas linking Rm, F, and D were established through logistic models and orthogonal distance regression parameters, demonstrating a high degree of correlation with historical disaster data (R2 > 0.8). Furthermore, the degree of shallow foundation damage was incorporated into evaluation indices. The fuzzy comprehensive evaluation entropy weight method was employed to calculate the vulnerability of masonry buildings in the region. Findings revealed a strong correlation between the empirical formula for shallow foundation damage and experimental data, facilitating reliable predictions of damage to shallow foundations. Incorporating this index into the evaluation framework improved the identification of low-vulnerability structures, addressed gaps in existing methods, and contributed to a more holistic approach. This research highlights the significance of including shallow foundation damage in vulnerability assessments of masonry buildings, particularly in areas prone to landslides, thereby improving the precision and reliability of risk evaluations.
Hurricane induced sediment transport and flooding can cause catastrophic damage to coastal communities. Forecasts test our understanding of the drivers of sediment transport and flooding, as well as quantify potential storm effects on coastlines. Real-time availability of forecasts allows emergency response and management decisions to be made with additional information. This study analyzed the skill of probabilistic forecasts of overwash and inundation along the Louisiana coastline from back-to-back Hurricanes Laura and Delta in 2020. To test skill, forecasts were compared with observations of mean water level relative to beach elevations and evidence of overwash and inundation identified in post-storm aerial imagery. We found that probabilistic forecasts accurately identified areas where overwash and inundation were likely to occur. In some cases, forecasts also produced a conservative estimation of overwash and inundation. For the first time, we explored the role of exceedance value in probabilistic forecasts of overwash and inundation and found that choice of exceedance value can influence forecasts. Additionally, we found that waves contributed up to 27
Quantifying the spatiotemporal uncertainty and temporal evolution of landslide hazard under potential seismic loading poses a significant challenge for risk reduction in mountain settlements. Taking the Zagunao River basin in Sichuan Province, China, as a case study, this study first constructs slope-unit-based dataset comprising seven continuous and four discrete assessment factors based on 2034 landslide events recorded from 2013 to 2022. Subsequently, an integrated model combining convolutional neural networks with a Bayesian-optimized gradient boosting tree is developed for feature extraction and susceptibility prediction. Furthermore, a probabilistic seismic hazard analysis (PSHA) method is employed to simulate peak ground acceleration (PGA) corresponding to a 10
Coastal urban road assets are increasingly exposed to storm flooding under climate change and sea level rise. Taking Shanghai as an example, this study assesses the exposure and risk of road assets to flooding using a probabilistic risk model, based on storm flood inundation data under different climate scenarios (Current, RCP4.5, RCP8.5, RCP8.5 High-end) with return periods (200-, 500-, 1000-, and 5000-year), along with high-resolution road distribution data. The results show that: (1) The expected annual exposure (EAE) of road assets to flooding in Shanghai will rise from the current 306.5 × 103 m2 to 705.0 × 103 m2 by 2050 under the worst-case scenario (RCP8.5 High-end). Notably, the EAE for inundation depths exceeding 0.3 m will account for between 47.07 and 53.96
Rainstorm and flood disasters often trigger a series of secondary hazards, forming complex cascading processes. This study constructs a cascading effect analysis framework by integrating complex network with Bayesian network, and proposes a chain-level risk assessment model for rainstorm and flood disaster chains. Taking the Pearl River Delta region as a case study, this study identifies typical evolution paths and cascading effects of disaster chains, assesses chain-level risks, and proposes chain-breaking mitigation strategies. The results show that: (1) The evolution of disaster chains presents a typical structural pattern of “Rainstorm → Flooding and urban waterlogging → Infrastructure damage → Social impacts”, with urban waterlogging serving as a key hub node. (2) With increasing rainstorm intensity, the cascading effects are significantly amplified, as evidenced by a sharp rise in the probability and severity of secondary hazards such as urban waterlogging and landslides and debris flows, which further intensify social impacts including transportation disruption and casualties. (3) Urban waterlogging is identified as the highest-risk node, while the disaster chain “Gale-force winds → Storm surge → Rising water levels → Flooding → Urban waterlogging → Casualties” exhibits the highest overall risk. Additionally, four-node disaster chains are the most frequent and demonstrate the strongest cumulative risk effects. (4) Implementing interventions at upstream high-risk nodes within the disaster chain can significantly reduce overall risk. Specifically, enhancing urban waterlogging prevention measures reduces total disaster chain risk by 37.96
Landslides induced by subsurface water infiltration represent a growing hazard in rapidly urbanizing regions, particularly where infrastructure is deficient. Leakage from potable water distribution systems can trigger slope instabilities that often remain undetected by conventional rainfall-based early warning systems. This study employs electrical resistivity tomography and three-dimensional resistivity imaging to identify leakage pathways and assess subsurface moisture conditions in a high-risk sector of Recife municipality, Northeastern Brazil. High-resolution 2D and 3D multigrad array surveys were conducted and converted into volumetric soil moisture estimates using empirical relationships. The results reveal a major infiltration zone and several shallow, spatially distributed leaks, with saturation levels frequently exceeding 30–40
This study uses an agent-based model (ABM) to assess how household and government actions influence flood risk in an urban area under climate change. Seventeen regional climate models are evaluated, and projections from the best-performing model are used in hydrologic and coupled 1D/2D hydraulic models to simulate hydrographs and corresponding inundation maps for different return period events. A survey is conducted to more accurately represent households’ initial flood risk and coping perceptions. A flood event is randomly assigned to each year within the simulation period (i.e., 2025–2100) in accordance with the exceedance probabilities associated with different return period events to generate a realization. An ensemble of 100 realizations is generated to capture stochastic variability in flood occurrences over 2025–2100. Results show that using survey data leads to higher estimated economic damages than those obtained with randomly initialized risk and coping perceptions, highlighting the importance of site-specific data. Government actions are generally found to be more effective than household actions. For extreme realizations, such as the occurrence of two 500-year flood events within the simulation period, only proactive government actions can significantly reduce economic damages by up to two-thirds or more. Similarly, prevention-based household measures are more effective than response-based ones triggered by extreme floods, especially when the triggering extreme flood(s) occur later in the simulation period.
