Rapid constraints on earthquake rupture geometry and shaking intensity are critical for early hazard assessment, but remain difficult to obtain in the first tens of minutes after large earthquakes, particularly in regions with sparse seismic instrumentation. Here we present a globally scalable framework that combines real-time dense-array seismic observations, high-frequency (0.5–10 Hz) multi-array back-projection, known fault geometries, global site-condition models (Vs30), and empirical ground-motion prediction equations to derive rupture extent and physically consistent seismic-intensity fields within ~30 min of origin time. Applications to the 2025 Mw 7.7 Myanmar strike-slip earthquake and a separate Mw 8.8 Kamchatka megathrust event recover rupture dimensions across contrasting tectonic settings that are consistent with InSAR measurements, early aftershock distributions, and macroseismic observations typically available only days later. Because the framework relies only on existing global seismic networks, it enables rapid impact characterization without dense near-source observations. By bridging rapid rupture imaging and physics-informed shaking assessment, this framework represents a methodological advance in global earthquake disaster science, enabling actionable impact estimates within the critical early response window and offering a scalable pathway to reduce seismic losses across vulnerable and under-instrumented regions worldwide.
Retrogressive thaw slumps, triggered by thawing ice-rich permafrost, have become increasingly widespread across the central Qinghai-Tibet Plateau. However, the spatiotemporal evolution of slumps remains poorly understood. Here we compiled a 1964–2024 slump inventory for the Beiluhe region using multi-temporal satellite imagery with deep learning-assisted manual interpretation. By tracking the evolution of hundreds of slumps, we found that 90% stabilized within 12 years, with an average longevity of six years. Among all of the stabilized slumps, only 13.1% reactivated, yet over 75% of these reactivations occurred after 2000. The newly initiated slumps shifted to higher elevations and well-drained areas, reflecting increased frequency of permafrost degradation in previously stable zones. Results suggest that high annual precipitation promotes new RTS initiation, while the permafrost thermal regime exerts a strong control on the magnitude of annual RTS activity (RTSN = e3.198×T+4.923 − 1, p<0.05). The spatial and temporal patterns in slump dynamics on the Qinghai-Tibet Plateau reveal compelling evidence of accelerated permafrost degradation under a warmer and wetter climate.
—Rapid identification of building damage distribution during the critical rescue phase is essential for efficient emergency response. However, most existing studies focus on earthquake-like structural hazards, limiting their generalization across different disaster scenarios. In addition, current methods often struggle to effectively integrate cross-temporal features from pre- and post-disaster imagery, while effective links between pixel-level classification and regional decision-making information remain limited. To address these issues, this study proposes an improved deep learning framework, NBDANet, developed based on BDANet for building damage assessment. NBDANet adopts a two-stage architecture with Spatial Pyramid Pooling (SPP) modules embedded in both stages to enhance multi-scale contextual representation. The first stage performs building localization, while the second stage introduces a Dual-input Triplet Attention (DTA) module to improve feature interaction between pre- and post-disaster images and enhance damage classification accuracy. The model is trained and evaluated on the xBD dataset and further applied to the Derna case using high-resolution GeoEye-1 imagery. Validation is supplemented by Sentinel-derived Normalized Difference Built-up Index (NDBI) and Damage Proxy (DP) maps, confirming spatial consistency. In addition, Global Human Settlement Layer (GHSL) data are integrated to estimate affected populations and delineate rescue-priority zones. Results show that 22% of buildings were damaged, including approximately 20% of all buildings classified as Major damage or Destroyed. Estimated affected populations in Major damage and Destroyed zones were approximately 5,200 and 2,650, respectively. The proposed approach demonstrates strong performance in building localization and damage classification, providing a scalable solution for multi-source disaster assessment and recovery planning.
In assessing future climate change risks the IPCC makes use of 58 standardized geographical reference regions. Global human population numbers and water-related risks in terms of economic and non-economic loss and damage are unevenly distributed across these reference regions. Here we show how the global population exposed to water-related risks is disproportionately contained in two reference regions and reflect on the implications for future IPCC assessments.
Storm surge from Atlantic tropical cyclones (TCs) is a large driver of fatalities and damage, accounting for 32% of all continental US (CONUS) TC fatalities from 1963 to 2024. Previous work has shown that landfalling minimum sea level pressure (MSLP) predicts normalized TC damage better than maximum sustained wind (Vmax), in part reflecting the robust relationship between storm size and MSLP. Given that storm size is an important driver of storm surge, we find a significantly stronger relationship between peak storm surge and CONUS hurricane landfalling MSLP (r = −0.78) than landfalling Vmax (r = 0.68). Peak storm surge also significantly correlates with both normalized damage (rrank = 0.72) and direct fatalities (rrank = 0.51). ADCIRC-simulated peak storm surges show consistent relationships with MSLP, Vmax, normalized damage, and direct fatalities, reinforcing the observational results. These findings highlight landfalling MSLP as a more skillful predictor of hurricane hazards and impacts than Vmax.
The April 2020 Siberian heatwave was among the most extreme early-spring warming events on record, with widespread environmental and societal impacts, yet it was missed by most subseasonal-to-seasonal (S2S) forecast systems. Understanding the sources of predictability and their limitations remains a key challenge in extended-range forecasting of land heat extremes. Here we show that Atlantic sea surface temperature (SST) anomalies constitute a key remote contributor, modulating a downstream Rossby wave response and upper-level ridging over Siberia, while local land–atmosphere feedbacks, including snow–albedo and radiative processes, amplified the near-surface warming. Regional replay experiments demonstrate that Atlantic SST conditions can provide effective forcing for the development of large-scale circulation anomalies consistent with the observed event, whereas Siberian land conditions further enhance their persistence. Forecast failures are associated with systematic SST biases in the Western North Atlantic, which weaken downstream wave propagation along the mid-latitude waveguide. As background conditions such as Atlantic warming, Eurasian ridging, and declining snow cover continue to evolve, these results highlight that reducing SST biases is essential for improving subseasonal prediction skill for high-impact Eurasian heat extremes.
Agricultural drought (AD), driven by root-zone soil moisture deficits, poses a major threat to food security. However, its future risk is commonly assessed by treating land-use and land-cover change (LULCC) and atmospheric shifts as independent drivers, overlooking their interacting and compounding effects. To address this gap, we use an integrated, multi-scale, multi-sector modeling framework to project AD risk for corn and soybean across the contiguous United States (CONUS) through 2055 under a range of plausible futures that link thermodynamic changes and LULCC pathways through shared socioeconomic and emissions scenarios. Model projections reveal sharp increases in drought-driven crop production losses. These losses are estimated as the drought-affected cultivated area multiplied by the yield deficit between the non-drought reference yield and simulated yield. Projected future losses rise by nearly 60% for corn and 135% for soybean relative to historical levels. LULCC acts as a driver of similar magnitude to atmospheric change, with the strongest amplification occurring where cropland expansion overlaps with drought-prone areas, such as the Great Plains. These findings highlight that interactions between LULCC and atmospheric shifts shape future agricultural drought risk and should be jointly considered to support effective adaptation and food-system planning.
Tropical cyclones are among the most destructive natural hazards, necessitating reliable uncertainty quantification in track forecasting. However, traditional probabilistic forecasting methods suffer from isotropic assumptions and fail to capture geometric distortion. This paper proposes an adaptive uncertainty quantification framework based on distribution-free conformal prediction. This framework utilizes deep learning models to construct anisotropic probability ellipses and integrates an online calibration mechanism to achieve joint dynamic adjustment of the ellipse scale and principal axis direction. Results demonstrate that the proposed framework outperforms conventional methods, offering a more expressive and operationally relevant characterization of forecast uncertainty.
Tropical cyclones (TCs) can elevate diarrheal disease risk yet inconsistent definitions and types of diarrheal data being used can obscure important heterogeneity. Here, we used harmonized weekly diarrheal mortality and morbidity data from 10 Asian regions between 2000 and 2021. We applied two-way fixed effects models to estimate TC-diarrhea associations across regions and compare associations across multiple TC definitions. These definitions include wind-based (wind intensity), rainfall-based (percentile threshold), and combined wind-rainfall metrics. We then assessed differences across a range of diarrheal outcomes. We found that mortality associations were generally not statistically significant and had wide confidence intervals across regions. For morbidity, we observed substantial heterogeneity across regions and definitions, with the greatest TC-attributable burden observed in Taiwan. These findings suggest that tailored TC definitions, calibrated to local health burden profiles, represent a promising strategy to improve early warning systems and guide interventions.
Snow avalanches pose a growing hazard in High Mountain Asia (HMA), yet their regional patterns are strongly governed by snow-climate regimes that are sensitive to both long-term warming and large-scale circulation variability. Using reanalysis-based meteorological and snow datasets (1980–2018), we develop a spatially continuous snow-climate zonation for HMA and classify the region into maritime (11.9%), transitional (4.1%), and continental (73.0%) snow-climate regimes. We detect a systematic shift toward warmer and wetter snow-climate characteristics, with the most pronounced changes along the southeastern HMA, where the transitional regime expands markedly. Variance decomposition further reveals non-stationary controls on zonation variability: temperature dominates temporal variability in the maritime and transitional regimes (explaining ~80% of their variance), whereas the continental regime is jointly regulated by temperature and snowfall, with a substantial contribution from the North Atlantic Oscillation (NAO) through its dynamical modulation of circulation and moisture transport. These findings provide a mechanistic, regime-aware framework for stratified avalanche-susceptibility modeling and for differentiated monitoring and risk management strategies across HMA under continued climate warming.
Tropical cyclones (TCs) have caused substantial mortality and economic losses worldwide, yet existing studies have primarily focused on individual risk components, specific regions, or single loss types. It therefore remains unclear whether TC-related mortality and direct economic losses are shaped by the same or different combinations of hazard, exposure, and vulnerability, and how these mechanisms vary across regions and income groups. This study develops an integrated assessment framework to quantify the relative contributions of these three factors to short-term directly recorded TC-related mortality and direct economic losses from 1990 to 2019. The results show that TC impacts are highly concentrated in a few countries and unevenly distributed across income groups: low- and lower-middle-income countries, including Bangladesh, Myanmar, and the Philippines, accounted for over 80% of TC-related mortality, whereas high- and upper-middle-income countries, including the United States, Japan, and China, bore 77% of direct economic losses. During the study period, TC hazard intensity increased, while both population exposure and GDP exposure exhibited pronounced upward trends, while vulnerability declined overall but remained high in many low-income countries. Attribution analysis indicates that GDP exposure is the dominant driver of global economic losses, with a 192.82% increase per 100% rise, whereas population vulnerability is the primary determinant of TC-related mortality, with a 63.58% increase per 100% rise. The dominant drivers also differ across regions and income groups. These findings provide quantifiable evidence for designing differentiated TC risk-reduction strategies based on exposure, vulnerability, and adaptive capacity.
As wildfires increasingly impact urbanized areas, modernizing early warning systems is vital for public safety. This research evaluates the operational utility of Fire Warnings—tactical National Weather Service (NWS) alerts issued at the request of local emergency management agencies—by examining how exposure and training influence practitioner adoption. Through focus groups with 85 experienced NWS fire partners, the study identifies a baseline of skepticism among those without Fire Warning training. Findings reveal a critical exposure gap: while untrained practitioners fear jurisdictional friction and bureaucratic delays, these “training-sensitive” barriers contrast with immersive simulations performed at National Oceanic and Atmospheric Administration’s Fire Weather Testbed where hands-on experience with a select group of fire partners fostered professional optimism towards Fire Warnings. Conversely, “structural barriers,” such as alert redundancies and the rapid lead-time collapse of fast-moving fires, remain persistent regardless of training. Here, we propose a flexible socio-technical framework to address these challenges. Recommendations focus on implementing modular training, encouraging Tactical Integrated Warning Teams (IWT), and ensuring Fire Warnings can be disseminated via the Wireless Emergency Alert (WEA) dissemination system. By triangulating focus group data with simulation insights, this research provides a roadmap for refining alerts during rapidly escalating, dynamic hazard events.
Didicas volcano in the northern Philippines is one of the most active volcanoes along the Luzon volcanic arc and a potential source of non-seismic tsunamis in the Luzon Strait region. Here we assess its conditional tsunamigenic potential using seven prescribed scenarios involving underwater explosions and flank collapses. Four explosion scenarios use equivalent source diameters of 3.5–6.0 km at depths of 100–400 m, whereas three collapse scenarios involve volumes of 0.5, 1.7, and 13.37 km³. These scenarios are not deterministic forecasts but are designed to bracket mechanism-dependent tsunami responses, from analogue-supported collapses to low-probability, high-impact upper-bound cases. The simulations reveal clear contrasts between the two mechanisms. Explosion-generated waves are short-period, rapidly dispersive, and largely confined to the near field, with amplitudes decreasing below 1 m beyond approximately 100 km. In contrast, flank-collapse tsunamis are longer-period, more persistent, and more efficient in transmitting energy across the Luzon Strait. In the largest collapse scenario, near-field amplitudes exceed 20 m, waves of approximately 1–2 m reach southern Taiwan island and the southern Ryukyus within 1–1.5 h, and amplitudes of approximately 0.8–1.0 m reach the southeastern China within about 3 h. Regional bathymetry strongly modulates these impacts: the Hengchun–Luzon–Gagua ridge–arc system guides energy northward, whereas the broad shelf delays and attenuates westward arrivals. These results show that flank collapses, particularly larger volume failures, should be explicitly considered in regional volcanic-tsunami hazard assessments and early-warning strategies.
Cities serve as major economic centers and experience significant socio-economic impacts from urban flooding, necessitating user-friendly decision-support tools. This study presents an R-based automated, web-messaging-based quasi-operational near real-time framework for rainfall and flood information dissemination. The framework integrates automated data acquisition and develops rainfall and consequence-specific flood risk maps using a probabilistic hazard-vulnerability approach; offering a scalable tool in resource-constrained cities, with potential extension to finer temporal scales.
Wildfire evacuation depends not only on hazard exposure but on whether residents can reach essential resources under stress. To capture this dimension of evacuation risk across California, the proposed Wildfire Evacuation Resource Index (WERI) quantifies census-tract accessibility to emergency shelters, hospitals, lodging, gas stations, and EV charging stations within a three-step floating catchment area framework, combining proximity, capacity, demand, and competition. A Monte Carlo disruption analysis stress-tests the index under fire-driven road and facility outages, and high-threat low-access tracts are classified by terrain and population density to separate addressable infrastructure gaps from structural geographic constraints. Bivariate Moran’s I links the resulting accessibility patterns to wildfire threat from California’s FRAP dataset. Wildfire threat coincides with high neighboring accessibility for shelter, lodging, gas, and charging, while hospital accessibility shows the inverse pattern. The sharpest mismatches concentrate in Los Angeles for gas and charging, where population outpaces capacity. Between 77 and 86 percent of high-threat low-access tracts for lodging, gas, and charging are in populated communities and are addressable through facility expansion, whereas hospital gaps are largely structural. Population exposure further shows that residents in high-threat areas disproportionately face limited rather than strong access to hospitals, gas stations, and EV charging.
Flood disasters pose persistent challenges to urban safety and emergency management. Traditional remote sensing technologies remain limited in real-time capability and spatial coverage. In surveillance-based urban flood monitoring, direct physical measurement of flood depth is often unavailable; therefore, flood risk level detection based on visual reference objects provides a feasible alternative for rapid flood assessment. However, existing lightweight YOLO models still struggle to accurately identify submerged targets under partial submergence, large-scale variation, and complex urban conditions. To address these issues, this study proposes UFD-YOLO, a lightweight framework for urban flood risk level detection using pedestrians and vehicles as multi-scale visual references. First, a reference-based flood risk level criterion was established according to the submergence states of pedestrians and vehicles, and a dataset covering diverse urban scenarios was constructed. Second, the DySnakeConv, AIFI, and DetectAux were progressively integrated into YOLO11n to improve local structural perception, strengthen feature interaction, and enhance multi-scale supervisory learning. Compared with YOLO11n, UFD-YOLO improves Precision, Recall, mAP50, and mAP50:95 by 2.5%, 5.1%, 5.8%, and 2.6%, respectively, while maintaining low computational cost. Additional evaluations further demonstrate its accurate and stable performance in challenging scenarios. Overall, this study provides an effective technical approach for urban flood risk assessment.
Rapid and reliable earthquake magnitude estimation in the first seconds after rupture initiation is critical for effective earthquake early warning and emergency response. However, the limited information contained in the earliest seismic signals makes this task challenging. Here we develop a deep-learning model, A2MAG, to estimate earthquake magnitude using only the first 3 s of strong-motion acceleration records from the dense K-NET network in Japan. The model is trained and tested on more than 130,000 waveforms from over 12,000 earthquakes recorded between 1996 and 2024. It achieves a mean absolute error of 0.33 magnitude units and performs consistently across different tectonic regions. The model systematically underestimates large earthquakes (M > 6), reflecting the longer rupture duration and limited early energy released by large events. Using the same input waveforms, the approach also provides highly accurate P-wave arrival times with a mean absolute error of 0.07 s. These results demonstrate both the potential and the fundamental limitations of estimating earthquake magnitude from the earliest seismic signals and provide preliminary evidence for the feasibility of the proposed approach in earthquake early warning applications.
Flood risk assessments underpin flood management and resilience efforts worldwide, including land-use planning, infrastructure design, and insurance requirements. Many of these assessments rely on design storms, which assume a one-to-one relationship between the frequency of storms, flooding, and damage and neglect the spatial and temporal structure of rainfall. Here, we show that these assumptions can lead to systematic misrepresentation of flood hazard and risk. Using a coastal watershed in North Carolina, we compare design storm-based estimates with those from stochastic storm transposition, a probabilistic framework used to generate synthetic events with realistic rainfall fields. Though both methods produce similar basin-averaged rainfall statistics, we find design storms underestimate flood inundation frequency by 31–35% and expected annual damage by 93% relative to SST. These results reveal how complex storm-flood-damage relationships amplify risk from smaller, more frequent storms and illustrate that accounting for spatiotemporal rainfall variability is crucial to risk assessment.
Tropical cyclones (TCs) and subtropical cyclones (STCs) can bring very severe disasters to coastal areas and nearby inland regions when they move along the coastlines or make landfalls. Moreover, their cyclone phase transitions (PTs) may lead to substantial changes in their structures, intensities and tracks. Hence, near-coast PTs pose additional challenges in disaster preparedness. The current study shows that the North American Subseasonal Variability (NASV) have substantial modulation of landfalls and near-coast PTs at all Atlantic coasts of the U.S. and Canada. When the anomalies associated with strong northward propagating NASV are located at specific regions, the Atlantic coasts of interest experience significantly different landfall rates, landfall intensities or PT percentages. Thus, our findings imply possibilities of improving predictions of landfalls as well as near-coast PTs on both subseasonal and synoptic timescales. The results also suggest that NASV may have contributions from the convection anomalies over the western tropical Pacific.
