The development of tsunami fragility curves and vulnerability functions has expanded in recent years, including those specifically for port environments. Given the importance of ports in global trade and supply chains, there is a need for operational tools that can support the assessment of potential tsunami impacts on port structures. In this study, we contribute to bridging the gap between published tsunami fragility curves and vulnerability functions and their practical application through the development of RAFT–Ports (Risk Assessment For Tsunamis in Ports). RAFT–Ports is an open-source, user-friendly application that applies fragility and vulnerability functions developed for port structures to estimate structural-damage exceedance probabilities, expected physical loss, and production-capacity recovery probabilities for specified tsunami scenarios. Its output is a report for scenario-based tsunami impact assessment, summarising damage, loss, and recovery metrics through inundation maps, bar charts, summary tables, and the complete set of raw analysis files. It is designed to be adaptable, allowing users to modify fragility-curve and vulnerability-function values as new empirical data become available. We demonstrate its functionality through a case study at Tanjung Gerem Port, Indonesia, using hypothetical Mw 8.9, Mw 8.7, and Mw 8.5 earthquake scenarios along the Sunda Megathrust. This work contributes to technology transfer between scientific research and end users by showing how fragility-curve and vulnerability-function data can be transformed into actionable assessments of tsunami impacts that can support risk-informed mitigation, preparedness, and recovery strategies.
Abstract Probabilistic tsunami hazard assessment (PTHA) at the inundation level over large spatial scales remains computationally challenging, requiring many scenarios, fine-resolution modelling, and high-quality digital elevation models (DEMs). Previous studies have addressed either regional offshore hazards or local-scale inundation, leaving comprehensive regional inundation assessments largely unexplored. We applied a two-stage framework to perform the first inundation-level PTHA covering over 1000 km of the southern coast of Java, Indonesia, for Java megathrust tsunamis. The framework couples coarse-resolution (450 m) offshore propagation modelling using JAGURS (Stage 1) with high-resolution (30 m) onshore inundation using SFINCS (Stage 2). We simulated 4400 earthquake scenarios ( $$M_w$$ 7.0–9.2) using a space- and time-dependent Epidemic Type Aftershock Sequence (ETAS) model with heterogeneous stochastic slip models. Maximum offshore tsunami heights reach 10.3 m, 18.9 m, 23.0 m, and 29.5 m at annual exceedance rates of 1/500, 1/2500, 1/5000, and 1/10000, respectively, with inundation extending several kilometres inland depending on local topography. Sensitivity analyses show that hazard estimates depend strongly on the number of simulated events and that linear shallow water equations in Stage 1 can generate unrealistic large onshore hazards. Open-source DEMs (BATNAS and DeltaDTM) proved suitable for regional PTHA once known dataset limitations are accounted for. The onshore estimates represent a lower bound, and the $$M_w$$ 9.2 scenario a conservative upper bound given structural constraints on the megathrust. This modular framework provides a transferable approach for large-scale inundation-level PTHA in other tsunami prone regions.
Detrital zircon U–Pb ages from a Middle Siwalik Nagri Formation sandstone ( 11–9 Ma) in the Kohala section, Hazara–Kashmir Syntaxis (western Himalaya, Pakistan), are used to assess sediment provenance and drainage connectivity in the western Himalayan foreland basin. The 98 concordant ages are dominated by Neoproterozoic–early Paleozoic grains ( 1000–400 Ma; 59
Abstract Flooding is a persistent environmental challenge in Bayelsa State, Nigeria, driven by low-lying terrain, high rainfall, and inadequate drainage systems. This study evaluates flood susceptibility in the Lower Niger Basin using a geospatial modelling approach. Ten flood-conditioning factors; rainfall, drainage density, slope, elevation, soil texture, land use/land cover (LULC), Normalized Difference Vegetation Index (NDVI), and proximity to rivers were integrated within a Random Forest (RF) framework. Spatial datasets were processed using ArcGIS and Google Earth Engine, and the model was trained with historical flood and non-flood locations to produce a high-resolution susceptibility map. Results indicate that areas receiving 2500 to 3199 mm of annual rainfall, particularly Southern Ijaw, Ekeremor, and Brass, exhibit the highest flood susceptibility. Over 82% of the region is characterized by low slopes (0 to 3°) and elevations ranging from 16 to 41 m, increasing flood vulnerability. High drainage density zones (up to 0.677 km/km²) contribute to flash flooding, while clay and sandy clay loam soils (covering > 60% of the area) reduce infiltration and enhance runoff. Built-up areas and regions with NDVI < 0.1 are also associated with higher flood risk. The RF model achieved an accuracy of 91%, with high-risk zones predominantly located within 1 km of major rivers. This study demonstrates the effectiveness of integrating machine learning with geospatial analysis for flood susceptibility mapping in data-scarce, low-lying environments. The resulting map provides a reliable decision-support tool for flood risk management, land-use planning, and targeted mitigation strategies in Bayelsa State.
Abstract Accurate short-term wind speed forecasting is critical for wind power integration. Although deep learning models exhibit strong predictive capability, their performance is often limited by systematic underestimation when optimized with conventional symmetric loss functions. This issue arises from the right-skewed Weibull distribution of wind speed, where the scarcity of high-wind-speed samples induces pronounced data imbalance and leads to biased learning, particularly under longer forecast lead times. To address this problem, we propose a lightweight Time-Dependent Weighted Quantile Loss (TDWQLoss), which introduces a time-dependent quantile weighting mechanism that adjusts penalty strength according to forecast horizon, enabling a more appropriate representation of increasing uncertainty with lead time. We evaluate the proposed method using three representative spatiotemporal models (Conv3D, SimVP, and MFWPN). Experimental results show that TDWQLoss consistently reduces systematic negative bias and improves overall forecasting accuracy across all models and lead times. In addition, it improves performance in event-based metrics, including Probability of Detection (POD) and Critical Success Index (CSI), without requiring any modification to model architectures. These results demonstrate that TDWQLoss provides a simple, architecture-agnostic, and effective strategy for mitigating bias in wind speed forecasting.
