Landslide susceptibility mapping (LSM) is important for hazard prevention and land-use planning. Current machine learning-based LSM studies often lack integrated interpretation of spatial explanation, model contribution and factor interactions. To address this issue, this study developed a GeoDetector–SHAP coupled explainable machine learning framework for LSM and applied it to Jiangxi Province, China. The modeling dataset comprised 156 verified damaging landslides documented during 2019–2020, 468 non-landslide samples, and nine conditioning factors. Four machine learning models were constructed and compared to identify the optimal model for subsequent interpretation and mapping. Based on the optimal model, GeoDetector and SHAP were first used to interpret landslide-controlling factors from spatial explanatory and model-prediction perspectives, respectively, and were then integrated through a coupled explanatory index for comprehensive analysis. To further clarify the interaction mechanisms among these dominant controls, the GeoDetector interaction detector was applied. The coupled results showed that, given the available inventory and analytical design, distance to roads, DEM, and lithology ranked highest. The proposed framework provides a more comprehensive interpretation of landslide-controlling factors and offers useful support for regional landslide susceptibility assessment.
Climate change and human activities jointly affect the hydrological cycle of rivers. Quantifying the contributions of climate change and human activities to future runoff changes and assessing the impact of uncertainty sources on future runoff changes are crucial for water resource management. Taking the source area of the Lancang River (SALR) as the research area, this study constructed a framework that integrated a global climate model (GCM), shared socioeconomic pathway (SSP), hydrological model (HM), seasonal Budyko model, and variance analysis (VAAN). This framework was adopted to explore the contributions of climate change and human activities to the future seasonal runoff changes in the SALR, and to quantify the influence degree of uncertainty sources on the attribution results of future seasonal runoff changes. The main conclusions were as follows: (1) The mutation year of runoff depth in SALR was 2008. (2) Both the ABCD and DWBM hydrological models can accurately simulate runoff variation processes in SALR. (3) Human activities are the dominant factor influencing the future runoff changes in the spring, summer and autumn of SALR, contributing 5.63–18.96, 3.10–16.18, and 4.26–16.51 mm, respectively, to runoff increases under different SSP scenarios, respectively; while climate change is the dominant factor influencing the future winter runoff changes of SALR, contributing 2.30–8.79 mm to runoff increases. (4) The interactions among HM, GCM, and SSP are the dominant factors influencing the uncertainty of the attribution analysis results of future runoff changes, accounting for over 68.64% of the total variance (within the ensemble framework comprising 2 HMs, 3 GCMs, and 3 SSPs).
This study analyzes microwave brightness temperature (MBT) anomalies associated with the Mw 7.7 earthquake in Myanmar on March 28, 2025. Satellite remote sensing identified four localized positive MBT anomalies along the Sagaing Fault. The eastern anomaly persisted since February 2024, while the southern anomaly exhibited intermittent enhancement, peaking in March 2025. The southeastern anomalies were less pronounced and post-seismic. For the central zone, based on threshold, confidence interval, and p‑value analyses, the anomaly is negligible or absent. The spatial correlation between these anomalies and the fault is attributed to prolonged compressional stress from the Indian–Sunda Plate interaction. Integrated multi-frequency and multi-source data analysis excluded confounding factors such as vegetation and soil moisture. These results suggest that changes in surface dielectric properties are the most probable mechanism for the observed MBT enhancements. These findings provide valuable references for future seismic monitoring and prediction in the region.
This study presents a Bayesian smoothed-seismicity probabilistic seismic hazard assessment (PSHA) for Baish Dam and the surrounding Wadi Baish area, southwestern Saudi Arabia. The workflow combines catalogue preparation, Gardner-Knopoff declustering, kernel-smoothed spatial seismicity modelling, Bayesian posterior sampling of source-node and recurrence parameters, GMPE logic-tree treatment, hazard-curve integration, uniform hazard spectrum calculation, and magnitude-distance/spatial deaggregation. The approach is used as a reproducible catalogue-based baseline because the local active-fault slip rates, segmentation, paleoseismic history, and recurrence parameters required for a complete fault-source model are not yet sufficiently constrained. The results show that the dam is mainly influenced by distributed seismicity along the southern Red Sea, Jizan-Yemen Hot Spot, southwestern Arabian-Yemeni, and Gulf of Aden-Afar domains. Mean rock-site PGA values are approximately 0.040 g and 0.069 g for the 475-year and 2475-year return periods, respectively, while SA(0.1 s) reaches about 0.096 g and 0.170 g. Deaggregation indicates that Mw 5.1–5.9 earthquakes at about 225-325 km dominate the catalogue-based hazard. These rock-site estimates provide a baseline for dam-safety evaluation, while future work should integrate active-fault characterization, site-response analysis, reservoir-foundation interaction, and fragility assessment to translate the probabilistic ground-motion levels into engineering risk metrics.
Previous studies have not precisely quantified the relative contributions of different drought types during propagation. This study used lagged feature importance derived from a random forest model to quantify the contributions of meteorological and surface water droughts to groundwater drought. Conditional probability and spatial analyses are combined with Copula functions to assess joint exceedance probabilities under different scenarios and reveal the conditional and multidimensional responses of groundwater drought. The results show that: (1) The optimal propagation time from meteorological drought to groundwater drought is 3 months, compared with 24 months for surface water drought. Under their combined effects, the groundwater drought index (GDI) is best explained by standardized precipitation index (SPI) lags of 0–3 months and standardized runoff index (SRI) lags of 0–9 months (R2=0.600). (2) Meteorological drought propagated to surface water mainly within 12 months, whereas groundwater responded more slowly. Groundwater drought areas generally overlapped with earlier surface water deficits, and longer lags occurred at higher elevations. (3) Under high joint probability scenarios (0.66–1.00), the average duration, severity, and affected area of groundwater drought are markedly greater than those under low and medium probability levels. These findings improve the understanding of drought propagation and support watershed water management and drought risk assessment.
Slope unit (SU) extraction is essential for landslide susceptibility assessment. Traditional methods neglect mechanical properties, yielding coarse units with poor stability and uniformity. We propose YaMSS, a novel method integrating topographic and mechanical properties via multi-scale segmentation. It first segments DEM into units with uniform slope aspect, then subdivides them based on yield acceleration (Ya) – a key seismic stability indicator. When applied to Wenchuan County, Sichuan, China, an area heavily impacted by the 2008 Mw 7.9 Wenchuan earthquake, YaMSS generates SUs with high uniformity in both slope aspect and stability, reducing the average size by approximately 97% compared to that with the hydrologic method, and is similar to that with the Huang method. Validation via logistic regression demonstrates that YaMSS captures the spatial heterogeneity of regional landslide susceptibility, achieving the landslide area density in extremely high, and the landslide susceptibility levels of high-position slope was 43.58% for YaMSS – markedly higher than the other two methods. In addition, the permanent displacement analysis verifies that the yield acceleration adopted herein enables rapid seismic landslide hazard assessment under various scenarios. By incorporating mechanical and aspect details, YaMSS offers an enhanced framework for seismic landslide susceptibility assessments in complex earthquake-prone mountain regions.
Flood disasters impair transportation networks, reducing accessibility of shelters and emergency supply routes. Existing emergency plannings assume road reliability and seldom account for flood-induced disruptions in shelter allocation and supply distribution. To address this gap, this study develops an integrated framework for emergency shelter allocation and supply distribution planning under flood scenarios. For road reliability assessment, betweenness centrality is incorporated as a key indicator to identify critical road segments and capture their topological importance within the road network. For shelter allocation, both surrounding flood inundation conditions and road reliability of candidate shelters are incorporated into the allocation process. For supply distribution, K-means clustering is employed to group shelters into distribution zones. A multi-objective routing optimization model is built with minimizing travel time and maximizing route reliability using NSGA-II algorithm. The framework is applied to Fengxian District, Shanghai, China. The results indicate the improved shelter allocation model enhances safety and accessibility of selected shelters. The Pareto-optimal solution set reveals trade-offs between travel time and route reliability, providing diversified decision support for route planning. The proposed framework provides scientific decision support for municipalities and emergency management agencies in flood preparedness and emergency logistics management, and is transferable to other flood-prone urban areas.
Deep-seated Gravitational Slope Deformations (DGSDs) are large-scale slowly progressing mass movements that can have a prominent influence on long-term landscape evolution and slope stability in mountainous regions. However, despite their geomorphological relevance, they have only been partially mapped in the Friuli Venezia Giulia (FVG) region of northeast Italy. In this study, we compiled the first comprehensive inventory of DGSDs in this region based on the integration of high-resolution LiDAR-derived products and orthophotos, remote sensing datasets, and existing landslide inventories. A total of 21 DGSDs, covering areas ranging from 0.63 to 6.48 km2, were identified, 14 of which were previously unreported. Their spatial distribution shows a clear clustering in the Western and Central Carnic Alps, and is controlled by lithological variability, tectonic inheritance, glacial debuttressing, and relief energy. We selected four representative DGSDs for detailed geomorphological analysis using field surveys supported by Uncrewed Aerial Vehicle Digital Photogrammetry (UAV-DP), thereby enabling their high-resolution mapping and characterisation of their internal kinematics. Overall, the inventory provides a robust geomorphological framework for understanding the distribution of DGSDs in the FVG Alps and will contribute to future hazard assessments and land-use planning. Moreover, our findings highlight the value of UAV-DP for detailed landform detection.
Understanding the relationships between flooding and potential factors is fundamental to effective risk management; however, progress is often impeded by the scarcity of spatially explicit flood data and the limited capacity of conventional models to decipher complex nonlinear relationships. To address these limitations, this study develops a regional-scale explanatory framework using remote sensing-derived flood records, multi-source environmental variables, and an interpretable machine learning model. Taking China’s Yangtze River Delta (YRD) as a case study, the results demonstrate that the XGBoost-based SHAP model outperforms random forests and logistic regression in regional flood prediction, achieving an accuracy of 0.970. The model characterizes nonlinear relationships between flood occurrences and environmental factors, revealing opposite associations of specific drivers across identified thresholds. These relationships vary across terrains and land-use types. Floods in mountainous areas are primarily associated with topographic wetness and surface texture indices, whereas those on plains are jointly influenced by river network density and surface texture. Notably, the flood effects of the largest patch index shift from negative in forests to positive in urban areas. These findings advance the understanding of regional flood occurrences and support site-specific flood mitigation strategies.
Landslide susceptibility mapping (LSM) commonly relies on the assumption of geographical similarity. However, existing sample augmentation methods assume global similarity and often overlook regional differences in landslide-forming mechanisms, limiting model generalization in spatially heterogeneous mountainous regions. To address this gap, we proposed a Geo-domain Constrained Similarity (GCS) framework and validated it using 336 historical landslides in Zhenxiong County. The framework first partitions the study area into homogeneous geo-domains through spatial clustering and then develops two complementary similarity metrics: SWCFD, which characterizes environmental background similarity, and CF-BE-HMD, which quantifies landslide-forming mechanism similarity. Based on these metrics, three sample augmentation strategies were designed to identify representative training samples from real geographical units rather than synthetic feature-space samples. The results showed that GCS successfully identified geo-domains with distinct controlling factors, supporting the necessity of zonal modeling. The combined strategy increased the Accuracy of both Random Forest and Support Vector Machine models from approximately 73% to over 83%, while improving the AUC from about 0.80 to above 0.92. It also enhanced prediction stability by reducing uncertainty associated with spatially heterogeneous samples. The proposed framework provides an effective sample optimization strategy for LSM, geohazard investigation, and regional landslide risk management in complex mountainous areas.
The southeastern Qattara Depression is a geomorphologically active region of Egypt’s Western Desert, where longitudinal dune fields intersect major development corridors, including the New Delta Project and petroleum concessions. This study presents the first integrated multi-temporal assessment of longitudinal dune morphodynamics over 35 years (1990–2025) using satellite imagery, digital elevation data, geological maps, and climatic records. Fifty-two dunes were analyzed using morphometric and kinematic indicators. Simple dunes dominate (78.85%), while complex forms account for 21.15%. Dune volumes range from 6.42 × 106 to 4.98 × 109 m3. Strong correlations between dune dimensions and volume (r = 0.73–0.89), indicate that lateral accretion as the primary growth mechanism. Dune activity has accelerated through increasing lateral migration, longitudinal growth, and vertical accretion driven by high-energy winds and prolonged drought. An Analytic Hierarchy Process (AHP) framework integrated the Sand Mobility Index, Normalized Difference Sand Index, and Normalized Difference Vegetation Index to assess geomorphological hazard and land-use vulnerability. The resulting dune hazard, vulnerability, and sand-drift risk maps spatially classify risk. ROC–AUC validation showed excellent predictive performance (AUC ≈ 0.93). High and very high hazard zones cover 22.96% of the study area, revealing substantial threats to ongoing development.
With the advancement of urbanization in China, human activities have gradually intensified, further exacerbating geological and ecological damage. This study aims to develop a comprehensive vulnerability assessment model to enhance the accuracy of risk evaluation and devise effective prevention strategies. To advance existing FAHP frameworks, this study extends the defuzzification process into the triangular fuzzy number domain using a possibility degree relation matrix, and combines expert preference information to derive a crisp and reusable indicator weight system. This optimized FAHP is then integrated with a projection model to construct the FAHP projection model, which captures both the magnitude and directional closeness of alternatives to positive and negative ideal states, thereby overcoming the information loss typical in standard fuzzy evaluations. Using Luding County in Sichuan Province as a case study, we established a multi-level vulnerability evaluation indicator system integrating geographical and social factors. The indicator weights and comprehensive projection values were calculated for nine townships and mapped using ArcGIS with the Natural Breaks method. The quantitative results indicate that Luqiao and Lengqi Town exhibit extreme vulnerability (accounting for 11.79% of the county area), primarily driven by dominant socio-economic exposures such as population density (weight: 0.0968) and education level (weight: 0.0885). Conversely, remote areas like Pengba fall into the low vulnerability category (accounting for 68.72% of the county area). This study demonstrates the superior practicality of the proposed FAHP-projection model in geological disaster vulnerability evaluations and provides a quantitative scientific basis for regional security planning.SYNOPSIS Advances fuzzy analytic hierarchy process with triangular possibility degrees for vulnerability modeling in China’s urbanizing regions. Applied in Luding County, our integrated geographical-socioeconomic system quantifies township-level risks via projection weights. ArcGIS mapping validates superior accuracy for disaster prevention strategies.
Taiwan is situated on the Circum-Pacific Seismic Belt and experiences frequent, destructive earthquakes, such as the 2024 Hualien event. Large-scale post-disaster assessment remains reliant on manual on-site investigation, which is labor-intensive, poses safety risks to personnel, and makes rapid quantitative damage measurement across extensive infrastructure difficult to achieve. To address these limitations, this study proposes a Geo-AI workflow integrating high-resolution UAV photogrammetry with deep learning for automated damage quantification. U-Net is utilized for macroscopic land cover classification to isolate pavement areas as primary regions of interest. Microscopic crack identification and mask segmentation are subsequently performed using YOLOv11. This process enables the automated extraction of 3D geometric features including area, length, and width by overlaying identified crack masks onto high-precision Digital Surface Models. Validation at Hualien Port shows that YOLOv11 achieved a mean average precision of 93.4% for crack detection, with a crack area RMSE of 0.0065 m2. The automated workflow generates a comprehensive Pavement Condition Index map for the entire 55-hectare infrastructure within a single day, representing a time saving of over 90% compared to traditional reconnaissance at the validated site. This approach provides a safe, objective, and efficient solution for rapid post-earthquake damage assessment and reconstruction planning.
Climate change and anthropogenic activities pose significant threats to terrestrial ecosystem functions and processes, which greatly increase ecological risk. Therefore, investigating the spatiotemporal dynamics of ecological risk and its driving factors in the Beshilo River Watershed is essential for evaluating ecological conditions and supporting sustainable ecosystem management. This study integrates the composite ecological risk index (ERI) with machine learning (ML) and ensemble learning (EL) approaches to assess ecological risk. Ten driving factors were analyzed using the Geodetector and Shapley Additive Explanations (SHAP) to explore their influence on ecological risk. The results revealed that high ecological risk was primarily concentrated in the western and central parts of the watershed. Furthermore, the ML and EL models outperform the ERI in predictive accuracy. Among the ensemble models, the Random Forest -Extreme Gradient Boosting (XGBoost) model achieved the best performance (R2 = 0.976, RMSE = 0.0028, MAE = 0.0024). Trend analysis using the Mann-Kendall test and Sen’s slope indicates a statistically significant declining trend in ecological risk, despite noticeable temporal fluctuations during the study period. The normalized vegetation index, leaf area index, elevation, soil and precipitation were identified as the dominant drivers of ecological risk. This study provides valuable insights for developing effective ecological restoration and conservation strategies in the watershed.
Unpaved rural roads are critical lifelines in mountainous developing nations yet acutely vulnerable to monsoon-driven degradation, leaving infrastructure managers in data-sparse, cloud-prone regions without tools to identify vulnerable segments or detect pre-failure signals. This study presents a data-independent Sentinel-1 SAR time-series framework to characterise seasonal road surface vulnerability and identify hydrological anomalies preceding documented slope failures in Nepal, applied across four physiographically diverse sites over January 2020–January 2026. A seasonal hydrological sensitivity model classifies road segments into relative Low, Moderate, and High risk using monsoon–non-monsoon backscatter contrasts, while a complementary anomaly detection model identifies segments deviating abnormally from a site-specific pre-incident baseline. Monsoon-season VV backscatter peaks reach 1.7–2.0 dB above dry-season baselines across all six observed monsoon cycles, attributable to moisture-driven dielectric changes in road surface materials. The framework is validated against two independently documented geohazard events: the August 2020 Guthitar deep-seated road failure and multiple landslides along the Daklang–Listi corridor. The anomaly model localises pre-failure hydrological precursors four to six weeks before surface failure. The framework requires no field data, operates under all-weather conditions via Google Earth Engine, and is transferable to other data-sparse, monsoon-affected regions.
Natural landslide dams generally exhibit strong vertical heterogeneity in soil composition due to the complex source materials and emplacement processes of landslides. There is insufficient study on the failure evolution and mechanism of vertically heterogeneous landslide dams under coupled seepage and overtopping. In this study, the failure evolution process, breaching characteristics, and failure mechanism of landslide dams with different vertical heterogeneous structures were investigated by a series of physical model tests. The test results showed that the vertical heterogeneous structure significantly changed the seepage-erosion pattern of landslide dams before overtopping. The degree of slope erosion and crest deformation was quantified, and that of the dam with a sandwich structure was the largest. The breach erosion rate of heterogeneous dams was controlled by the erodibility of soils, the difference in soil properties of the contact zone and the seepage erosion. Compared with the homogeneous dam, the peak flood discharge of the dams with sandwich structure and reverse grain sequence were significantly higher, with the peak discharge increased by 11.1% and 6.9%, respectively. The breaching duration of the dams with sandwich and reverse grain sequence was significantly shortened, and the emerging time of the flood peak was advanced, whereas the situation was opposite for the dam with positive grain sequence. Interlayer seepage shear stress dominated contact erosion initiation and overall failure evolution of heterogeneous dams. The typical longitudinal evolution patterns of heterogeneous dams were stepped profile, multiple scarps, and inflection point in the erosion surface compared with a homogeneous dam.
Understanding the correlation and propagation between hydrological and meteorological droughts is crucial for drought early warning and water resource management. However, non-stationary behaviors under changing environments complicate accurate characterization. This study develops a non-stationary framework using climatic and anthropogenic factors within GAMLSS to construct non-stationary meteorological (NRDI) and hydrological (NSRI) drought indices. Combined with nonlinear response models, the propagation relationship is analyzed. Results indicate: (1) Non-stationary indices reveal more reliable drought propagation patterns and scale-dependent lag responses; (2) They capture drought events more accurately, particularly under climate change and human activities; (3) Stationary indices overestimate the hydrological response thresholds, underscoring the necessity of non-stationary approaches.
Landslide susceptibility mapping (LSM) is essential for disaster risk reduction in mountainous regions, but its reliability is often limited by spatial biases and incompleteness of landslide inventories. These reduce landslide inventory representativeness and influence the generalization of LSM models at the regional scale. To overcome this challenge, we developed a multi-source inventory integration framework that combines three typical complementary datasets: publicly available inventories, reliable field survey observations, and deformation hotspots derived from Multi-Temporal Interferometric Synthetic Aperture Radar (MT-InSAR) analysis. Four benchmark machine learning algorithms, including Random Forest, Support Vector Machine, Extreme Gradient Boosting, and Categorical Boosting, were applied to evaluate the incremental effects of different inventory integration strategies on the LSM performance. Results indicate that adding field survey data improves model reliability (AUC = 0.85), while incorporating InSAR-derived hotspots substantially enhances spatial coverage of landslides (AUC = 0.87), mitigating the geographical bias of LSM. The highest predictive accuracy (AUC = 0.89) was achieved by combining all three landslide inventories, capitalizing on the high fidelity of field data and the broad spatial reach of InSAR. This synergistic integration improves both sensitivity and precision in identifying landslide-prone slope units. The proposed framework offers a transferable approach for generating LSM in data-scarce regions.
Accurate extraction of damaged roads from post-earthquake remote sensing imagery is essential for emergency response and transportation recovery. However, post-earthquake roads are often affected by occlusion, structural fragmentation, irregular damage patterns, and complex background interference, which makes reliable extraction difficult for existing deep learning methods. To address these challenges, we propose HSGANet, a deep learning network for post-earthquake damaged-road extraction. HSGANet incorporates a Hybrid Strip Context Aggregation (HSCA) block to strengthen long-range dependency modeling and preserve road continuity, and a Multi-Scale Large-Strip Gated Attention (MS-LSGA) module to enhance the representation of irregular damaged roads while suppressing background interference during feature fusion. To evaluate the model under both synthetic and real post-disaster conditions, we construct and use two datasets: a DeepGlobe-based Hybrid Noise Synthetic Post-Earthquake Damaged Road Dataset and a Türkiye Post-Earthquake Damaged Road Dataset derived from post-event UAV imagery. Experimental results show that HSGANet achieves overall macro-averaged F1-scores of 90.03% and 96.46% and IoU scores of 82.71% and 93.31% on the two datasets, respectively, demonstrating its potential for more reliable and structurally consistent damaged-road extraction in post-earthquake accessibility assessment and rescue-route identification.
Floods are a frequent and destructive hazard in southern Bangladesh, particularly in low-lying regions like the Feni district. This study developed flood susceptibility maps for the coastal frontier of Feni using three machine learning models—Support Vector Machine (SVM), Logistic Regression (LR), and K-Nearest Neighbours (KNN). Fourteen topographic and environmental factors, including elevation, slope, MNWDI, MSAVI, rainfall, land use, and proximity to streams and roads, were used as inputs. A total of 199 flood and 132 non-flood points were identified through field surveys and satellite imagery. Multicollinearity checks confirmed variable independence. SVM achieved the highest train AUC (0.958) and test AUC (0.951), followed by LR and KNN. Susceptibility maps varied across models, with SVM predicting the greatest extent of high-risk zones. Feature importance analysis highlighted MSAVI, MNDWI, rainfall, road proximity, and Geology type as key predictors. The findings support flood planning, infrastructure development, and early warning strategies in Feni District.
