The Neretva River rises southeast of the Zelengora Mountain in Bosnia and Herzegovina at an elevation of 1,095 meters above sea level and flows approximately 220 kilometers to the Adriatic, of which only the final 22 kilometers lie within Croatia. Its topographic catchment spans roughly 10,975 square kilometers, mostly across the Dinaric Mountain massif, with only 692 square kilometers in Croatian territory. The delta itself developed over an alluvial area of about 190 square kilometers, formed during the last ice age when rising sea levels flooded karst areas along the river. Gravel and sand carried from upstream deposited in the valley, gradually raising the ground level. The delta existed in a largely natural state until the end of the 19th century, when the first regulatory works began, initially focused on maintaining the waterway from the sea to Metković for commercial transport. During the 20th century, numerous embankments, dams, and other hydrotechnical facilities were constructed and wetland reclamation transformed much of the former wetland into agricultural land, significantly altering the natural character of the area.

A defining feature of the Neretva system is its hydropower infrastructure. In the second half of the 20th century, a series of hydroelectric power plants with reservoirs were built in the upstream basin, including Jablanica in 1954 with a dam height of 85 meters and a reservoir volume of 318 million cubic meters, Rama in 1969, Grabovica and Salakovac in 1981, and Mostar in 1985. At the beginning of the 21st century, additional plants such as Peć Mlini on the Tihaljina tributary and Mostarsko Blato were commissioned, while the construction of the Ulog plant has begun and Konjic is planned. The Trebišnjica River basin, interconnected with the Neretva through deep and underexplored underground karst, carries its own substantial hydropower developments, including the large Bileća reservoir with a volume of 1,277 million cubic meters. These structures exert a dominant influence on the frequency, extent, and duration of floods in the downstream delta, and they also reduce sediment delivery to the valley, deepening the riverbed and worsening problems of water and soil salinity.

The hydrological regime of the basin reflects a transition between moderate continental and maritime influences. Average annual precipitation ranges from 1,000 millimeters on the coast to 1,900 millimeters in the mountainous parts of the catchment, with maxima exceeding 2,900 millimeters in elevated areas. The rainiest quarter of the year is November through January, which is also the period when high waters in watercourses are most frequent; the highest recorded monthly precipitation in this quarter at Metković was 400 millimeters in November 2021. Most precipitation in the Dinaric karst quickly sinks underground, accumulating in the subterranean system before reemerging through numerous springs along the edges of the valley, feeding the Neretva, Mala Neretva, and Norin watercourses. This interaction of surface and underground water produces stretched runoff hydrographs and means that floods in the delta are caused primarily by inflow of large waters from the upstream basin, with smaller contributions from karst springs and the sea, and the least impact from precipitation falling directly on the delta area.

Historical measurements at Metković, the furthest upstream hydrological station in Croatia, located approximately 21 kilometers from the river mouth, illustrate how the hydropower construction changed the flood regime. The highest water levels, exceeding 430 centimeters, were recorded in the 1950s before the construction of the Jablanica and Rama plants. After 1961, the highest recorded water level was 414 centimeters in December 2010, an event that flooded a large part of the valley. Because tidal influence prevents unambiguous determination of flows at Croatian stations, the study relied on statistical analysis of maximum annual water levels. Five theoretical distribution functions were tested, and the log-normal function was selected for Metković and Prud, while log-Pearson was chosen for the sea boundary. Calculated high waters for the 1,000-year return period reached 548 centimeters at Metković, 163 centimeters at the sea station, and 182 centimeters at Prud on the Norin River.

For the hydraulic simulation, the researchers built one-dimensional and coupled one-dimensional/two-dimensional hydrodynamic models using the MIKE 11 and MIKE FLOOD program packages from DHI, covering the Neretva River and its banks from the Croatian state border to the mouth. Calibration and verification were carried out separately for four areas due to the complexity of the system, using registered water levels from boundary stations and treating central stations as control points. Manning roughness coefficients ranged from 0.033 for the riverbed to 0.050 for inundation areas. The difference between modelled and measured water levels was mostly within an interval of plus or minus 5 centimeters, with larger deviations occurring only rarely during rapid water level changes. The authors attribute residual inaccuracies to insufficiently precise terrain geometry and to marginal karst springs whose variable inflows were not defined as boundary conditions but were indirectly incorporated through calibration against measured levels.

The simulations classified flood extent into three statistical probability levels. A high-probability flood, corresponding to a 25-year return period, would cover 51 square kilometers; a medium-probability 100-year event would inundate 62 square kilometers; and a low-probability 1000-year flood would spread across 110 square kilometers. The most endangered zone is the Vid–Norin area, where the city of Metković has expanded significantly and many residential, economic, and infrastructure facilities are at risk. The mouth of the Norin is a particularly sensitive location, allowing Neretva flood waters to enter that area. The Kuti area, a swampy zone intersected by numerous canals and influenced by karst springs fed from the underground hinterland, also has a high probability of flooding, though the risk there is moderated by low population density. The Opuzen-Ušće area, protected by numerous embankments following the course of the Neretva and Mala Neretva, carries the lowest risk.

Damage calculations considered seven categories, including populated areas, economic areas, agricultural areas, permanent plantations, green areas, water management facilities, and potential human victims. For the 1,000-year flood, total damage amounted to 250 million Euros, while a 100-year flood would cause 93 million Euros in damage. Populated areas would account for more than 90 percent of the total, with the greatest territorial losses expected in the wider area of Metković on both banks. The estimated delta damage represents approximately 1 percent of the total potential flood damage for the entire Republic of Croatia and about 6 percent of the damage for the Adriatic basin area. Flood risk, defined as the product of probability and adverse consequences, was expressed as an average annual potential number of victims of 0.67 per year and an average annual damage of 11.63 million Euros under conditions of 2015, before the construction of the Jerkovac embankment in Metković.

Climate change simulations modified both boundary conditions, assuming sea level rise at the downstream edge and increased inflow at Metković. The greatest predicted mean sea level increase was 87 centimeters, or 61 percent, for the end of the 21st century at the 1,000-year return period, accompanied by a 23 percent increase in discharge. Under these conditions, the average annual number of potential victims would rise to 0.88 and total potential average damage to 16.5 million Euros, increases of 31 and 41 percent respectively compared to the 2015 baseline. Based on a multi-criteria analysis, the researchers proposed optimal protective measures, including construction of an embankment on the right bank near Metković, a concrete defensive wall with a mobile barrier in the town, embankments on Mislina and near Krvavac, rehabilitation of riverbanks in critical sections, and a self-regulating gate at the junction of the Norin and Neretva rivers. Implementation would reduce the average annual potential number of victims by 70 percent, to 0.2 per year, and average annual damage by 83 percent, to approximately 2 million Euros, with a benefit-cost ratio of 18.04.

The study acknowledges limitations, including reliance on hydrological series ending in 2013 and the use of digital terrain models in certain areas instead of geodetic measurements, which introduces topographic inaccuracy. Extending the input series and defining hydrographs for marginal karst springs are identified as priorities for future work. Nevertheless, the researchers argue that the methodology is universal and can be applied to similar estuary and delta systems worldwide. Beyond flood defense, the quantified hydrological parameters such as water depth, flow rate, flood duration, and salinity can support ecological studies, which is particularly important given recommendations to preserve natural wetlands in the Mediterranean region. Because the Neretva basin spans two countries with separate protocols and professional services, transboundary cooperation in water management remains essential for reducing flood risks and improving conditions across the entire delta.

Subject of Research: Fluvial flood risk assessment and damage modelling in the Neretva River Delta, Croatia

Article Title: Delta flood risk analysis: case study from the Neretva River (Croatia)

Article References: Ljubenkov, I., Kvesić, D., & Erceg, J. (2024). Delta flood risk analysis: case study from the Neretva River (Croatia). Estuarine Management and Technologies, 1, 69-93. https://doi.org/10.3897/emt.1.137829

Image Credits: AI Generated

DOI: 10.3897/emt.1.137829

Keywords: Neretva River Delta, flood risk, hydrodynamic modelling, MIKE FLOOD, return period, climate change, sea level rise, Mediterranean deltas, karst hydrology, flood protection, damage assessment, Croatia

The research, carried out by Onur Onat of the Department of Civil Engineering at Inonu University in Malatya, rests on a deceptively simple idea. Every structure, from a guitar string to a skyscraper, has a set of natural frequencies at which it prefers to oscillate. These frequencies depend on the structure’s mass and stiffness, so when an earthquake cracks masonry, opens bed joints, or otherwise softens a building, the natural frequencies drop. By measuring the minarets’ vibration signatures before and after the Kahramanmaraş sequence, engineers can effectively ask the towers themselves how much damage they sustained, even in places the eye cannot see.

Ambient vibration measurement, the technique at the heart of the study, requires no artificial shaking. Wind, distant traffic, and micro-tremors in the ground provide a continuous, gentle excitation that is more than enough to coax a tall minaret into swaying at its natural frequencies. Sensitive seismometers recorded these barely perceptible motions at both towers, and signal processing identified the natural frequencies, damping ratios, and mode shapes, the characteristic patterns of deformation associated with each frequency, for the first four vibration modes of each minaret. The pre-earthquake and post-earthquake modal vectors were then compared using the Modal Assurance Criterion, a standard statistical measure of how similar two vibration patterns are.

The results tell a consistent story of measurable but not catastrophic degradation. The first four identified frequencies decreased by an average of 5.46 percent for the Yusuf Ziya Pasha minaret and 5.98 percent for the Sunullah Melik minaret. A frequency drop of roughly six percent may sound modest, but in structural terms it corresponds to a meaningful loss of lateral stiffness, the property that allows a slender tower to resist horizontal earthquake forces. Intriguingly, the damping ratios, which describe how quickly vibrations die away, showed no uniform trend, a reminder that energy dissipation in cracked masonry is a far messier phenomenon than stiffness loss.

Field inspections corroborated what the vibration data implied. At the Yusuf Ziya Pasha minaret, observers documented bending-induced tensile cracking in the lower body of the tower, exactly where the flexural demands of earthquake shaking would be expected to concentrate in a cantilever-like masonry shaft. At the Sunullah Melik minaret, the damage took a different form: sliding and separation along the bed joints, the horizontal mortar lines between stone courses, around the pulpit region. These distinct crack patterns illustrate a crucial point about masonry behavior, namely that the same earthquake can produce qualitatively different failure mechanisms in structurally similar towers depending on their geometry, construction details, and local ground conditions.

To translate these observations into a quantitative assessment of seismic performance, the study employed finite element models, numerical representations of each minaret built from thousands of interconnected elements. Rather than assuming material properties from textbooks, the researcher independently calibrated separate pre-earthquake and post-earthquake models using the measured fundamental frequencies of each tower. The models were then verified against the higher-mode frequencies and the modal vectors, ensuring that the numerical replicas matched reality not just in their simplest vibration pattern but across the full measured spectrum. This two-stage approach, calibration followed by independent verification, guards against the common pitfall of a model that fits one number by accident while being wrong everywhere else.

With trustworthy models in hand, the study moved from diagnosis to prognosis. Nonlinear time-domain analyses subjected both the pre-earthquake and post-earthquake models to five unscaled ground-motion records, letting the computer simulate how the towers would respond to real earthquake shaking in their intact and damaged states. Nonlinear analysis is essential here because masonry does not behave elastically under strong shaking; cracks open and close, joints slide, and stiffness degrades progressively as the motion intensifies. The models incorporated material behavior consistent with established plasticity formulations for masonry, allowing the simulations to capture the initiation and propagation of the very cracking patterns observed in the field.

The comparison between intact and damaged simulations produced one of the study’s most consequential findings. The post-earthquake stiffness reductions were reflected far more clearly in the deformation demands than in the global force responses, particularly under the stronger ground motions. In other words, a damaged minaret may carry roughly similar peak forces as an intact one, yet it must bend, sway, and distort substantially more to do so. Since deformation, not force, is what ultimately tears masonry apart, this result carries a sobering implication for heritage managers: towers that appear to have weathered an earthquake without collapse may nonetheless have crossed into a regime where the next strong event produces disproportionately larger displacements and more severe cracking.

Perhaps the most methodologically valuable conclusion concerns what vibration monitoring can and cannot detect. The combined measurement and modeling framework demonstrated that mode-dependent dynamic changes, the frequency shifts distributed across the first four modes, can identify earthquake-related stiffness loss even when the principal global mode shapes remain broadly preserved. This is good news for the growing field of structural health monitoring of cultural heritage, because mode shapes are notoriously difficult to measure accurately in the field, while frequencies can be tracked reliably with modest instrumentation. The study suggests that a monitoring program need not capture perfectly resolved shape changes to flag damage; a careful reading of frequency trends across multiple modes, interpreted through a calibrated model, suffices.

The implications extend well beyond two minarets in Malatya. Türkiye holds an extraordinary inventory of historical masonry minarets, and the 2023 Kahramanmaraş sequence damaged or destroyed many mosques and their towers across the affected region, prompting widespread post-earthquake assessments. The approach demonstrated here, pairing ambient vibration surveys with independently calibrated nonlinear finite element models, offers a repeatable template for triaging that inventory: identifying which towers have silently lost stiffness, quantifying how much seismic reserve remains, and prioritizing retrofit resources accordingly. As climate extremes and seismic activity continue to threaten unreinforced masonry heritage worldwide, the quiet hum of a minaret vibrating in the wind may prove to be one of the most eloquent damage reports a historic structure can give.

Subject of Research: Structural health monitoring and seismic performance assessment of historical masonry minarets damaged by the 2023 Kahramanmaraş earthquake sequence

Article Title: Structural health monitoring and seismic performance assessment of two historical masonry minarets before and after the 2023 kahramanmaraş earthquake sequence

Article References: Onat, O. (2026). Structural health monitoring and seismic performance assessment of two historical masonry minarets before and after the 2023 kahramanmaraş earthquake sequence. Bulletin of Earthquake Engineering. https://doi.org/10.1007/s10518-026-02721-4

Image Credits: AI Generated

DOI: 10.1007/s10518-026-02721-4

Keywords: structural health monitoring, masonry minarets, Kahramanmaraş earthquakes, ambient vibration, modal analysis, finite element model updating, nonlinear seismic analysis, natural frequencies, damage assessment, cultural heritage, earthquake engineering, Türkiye

The research, led by Alessandro Borre of the University of Genoa and the CIMA Research Foundation, together with Silvia De Angeli of the Université de Lorraine and colleagues, addresses a long-standing gap in disaster science. Scientists have long recognised conceptually that hazards interact, but the literature has lacked standardised, reproducible methods for quantifying those interactions. Most existing quantitative models of multi-hazard physical damage are confined to seismic scenarios, such as earthquakes followed by aftershocks or tsunamis, and often focus on single infrastructure types like bridges. Recovery, meanwhile, is typically treated simplistically or ignored altogether, even though it determines how much vulnerability persists when the next hazard arrives.

The team’s methodology distinguishes between two fundamental impact dynamics. Concurrent impacts arise when two or more hazards overlap in space and time, as when Hurricane Maria’s winds and floods struck the same buildings simultaneously. Consecutive impacts occur when hazards strike close enough together that assets have not fully recovered before the second event begins. The researchers translated these dynamics, previously described only in qualitative diagrams and conceptual frameworks, into a compact piecewise mathematical formulation controlled by two classification parameters, which they implemented as modular, open-source Python code.

The mathematics is elegant in its economy. For concurrent hazards, damage is computed through a bivariate vulnerability function that takes the maximum intensity of each hazard as input, capturing the amplification that arises when multiple stressors act on the same asset at once. For consecutive hazards, the model tracks each asset’s physical integrity through discrete time steps. After the first event, a recovery function, which can be linear, exponential, logistic, or user-defined, gradually restores the asset’s condition. When the second hazard arrives, the model calculates the residual damage remaining at that moment and applies the second event’s damage only to the portion of the asset still undamaged, preventing double-counting. Crucially, the vulnerability function applied to the second event is state-dependent: it shifts to reflect the fact that a partially damaged building is more fragile than an intact one.

To probe the internal mechanics of this formulation, the team conducted a systematic model behaviour analysis, varying three parameters: the time interval between events, the shape of the recovery trajectory, and the degree of state-dependent vulnerability modification. The results show that cumulative damage is strongly path-dependent. When the second event strikes during early recovery, losses substantially exceed the sum of two independent events. As the interval between events grows, additional cumulative loss declines and converges toward independent behaviour, but the rate of convergence depends heavily on the recovery function. Exponential recovery, with rapid early restoration, limits amplification; logistic recovery, with delayed early restoration, prolongs vulnerability and sustains elevated losses across a much wider range of inter-event timings.

Perhaps the most striking insight from the behaviour analysis concerns the fragility profile itself. When the researchers shifted the second event’s vulnerability function upward by 10, 20, and 30 percent to represent increasing fragility from incomplete recovery, they observed convex, non-linear amplification of cumulative damage. Even a modest 10 percent shift produced measurable increases in additional loss, while 30 percent shifts substantially raised both median outcomes and the upper tail of the loss distribution. The amplification, in other words, does not arise merely from the magnitude of residual damage but from a structural reconfiguration of vulnerability between events. Consecutive multi-hazard interaction, the authors conclude, is not simply additive accumulation but a transformation of the system’s response function.

The framework’s real test came in Puerto Rico, an island that experienced an extraordinary concentration of disasters in a short window. Hurricane Maria alone caused nearly 90 billion US dollars in damage, and reconstruction progressed unevenly; by the end of 2019, significant portions of the building stock remained unrestored. A post-hurricane assessment identified nearly 138,000 buildings still damaged more than a year after the storm, many concentrated in coastal and southern municipalities such as Ponce and Guayanilla, precisely the regions that later experienced the strongest ground shaking, exceeding 0.5 g peak ground acceleration, during the January 2020 earthquake. The team modelled the hurricane phase using a compound wind-flood damage model built on standard FEMA Hazus vulnerability functions, and their simulated compound loss of approximately 90 billion dollars matched official assessments without any post-event parameter fitting.

For the seismic phase, the researchers applied their consecutive damage model, transforming standard FEMA fragility curves into state-dependent curves to reflect residual hurricane damage. Because building-level recovery data were unavailable, they explored a predefined range of vulnerability modifications from 5 to 30 percent, applied uniformly across the building portfolio. The results revealed a pattern with profound implications: when the earthquake was modelled as an independent single-hazard event, assuming full recovery of building vulnerability after Maria, the simulated loss came to 1,205,525 dollars in 2020 prices, underestimating the official figure of roughly 1.7 million dollars by about 29 percent. But when the fragility medians for both structural and non-structural components were shifted by 15 percent, the simulated total loss rose to 1,742,296 dollars, closely matching the official estimate. The 15 percent value falls within the predefined exploration range and represents one plausible residual vulnerability scenario rather than a calibrated constant, yet the improved agreement demonstrates the structural importance of incorporating incomplete recovery.

The authors are careful to delineate the framework’s boundaries. The methodology targets sudden-onset hazards such as hurricanes and earthquakes, excluding slow-onset processes like droughts and sea-level rise. It captures only direct physical damage, not the indirect socioeconomic consequences that proved so devastating in Puerto Rico, where prolonged power outages triggered health crises, migration, and economic decline. Recovery is approximated with simple functional forms and assumed uniform across structural and non-structural components, even though non-structural elements typically recover faster. The portfolio-based approach groups buildings into categories, smoothing over localised variations in topography, construction quality, and exposure. These limitations, the researchers argue, define the scope of validity rather than undermining the approach, and they chart clear paths for future work, including agent-based recovery modelling, dynamic exposure tracking, functionality-based indirect impact assessment, and integration with probabilistic hazard occurrence models for predictive applications.

The broader significance of the study extends well beyond Puerto Rico. By providing a transparent, reproducible analytical template, the framework enables forensic reconstruction of past multi-hazard events, systematic exploration of the drivers of cumulative damage, and scenario-based assessment of alternative hazard timings and recovery pathways. Planners could use it to ask what-if questions: how much would losses shrink if recovery accelerated by a year, or how much would they grow if the next earthquake arrived sooner? The Puerto Rico case also demonstrates that post-disaster analysis can feed back into vulnerability modelling, allowing state-dependent fragility functions to be derived from observed cumulative impacts. As climate change intensifies hurricanes and urbanisation concentrates exposure in hazard-prone regions, the assumption that disasters strike fully recovered, fully intact systems becomes increasingly untenable. This study provides the quantitative machinery to retire that assumption, and with it, a more honest accounting of what consecutive disasters really cost.

Subject of Research: Quantitative assessment of physical damage and recovery dynamics from concurrent and consecutive natural hazards

Article Title: A quantitative methodology for analysing physical damage and recovery dynamics from concurrent and consecutive hazards: forensic insights from Puerto Rico

Article References: Borre, A., Ottonelli, D., Trasforini, E., Ghizzoni, T., Rudari, R., Zoppi, G., Boni, G., & De Angeli, S. (2026). A quantitative methodology for analysing physical damage and recovery dynamics from concurrent and consecutive hazards: forensic insights from Puerto Rico. Natural Hazards and Earth System Sciences, 26(10), 4763-4783. https://doi.org/10.5194/nhess-26-4763-2026

Image Credits: AI Generated

DOI: 10.5194/nhess-26-4763-2026

Keywords: multi-hazard risk, Puerto Rico, Hurricane Maria, earthquake sequence, recovery dynamics, fragility curves, damage assessment, disaster forensics, vulnerability, compound events, risk modelling, Python

The review, authored by Neha Sen, Swarup Chattopadhyay, and Rudra Mohan Tripathy of XIM University in Bhubaneswar, India, together with Dilip K. Prasad and Arif Ahmed Sekh of UiT The Arctic University of Norway in Tromsø, offers an unusually broad synthesis of how deep learning models have been applied to disaster data analysis across both domains. Rather than cataloging techniques in isolation, the authors trace the evolution of research trends, compare the architectures that have dominated different phases of the field, and assemble a rich archive of existing disaster datasets that researchers can draw upon for future work. The result is both a status report and a diagnosis of why so much promising laboratory work has failed to reach the field.

Technically, the field’s trajectory mirrors the broader history of computer vision. Early disaster-focused studies leaned heavily on convolutional neural networks for image classification, sorting satellite scenes into categories such as flooded or intact, damaged or undamaged. As datasets grew and architectures matured, attention shifted toward semantic segmentation, in which models label every pixel of an image, producing detailed damage maps that can guide aerial assessment. More recently, the authors observe a pronounced pivot toward object detection and real-time human monitoring, tasks that produce directly actionable outputs: the coordinates of stranded people, the footprint of a collapsed structure, the extent of an oil spill or wildfire front. This shift reflects a hard-earned lesson about deployment, because a classifier that reports aggregate statistics is far less useful to a rescue coordinator than a detector that points to individuals in need.

On the social media side, deep learning has transformed what is possible with text, images, and video posted by people in the middle of a disaster. Natural language processing models can triage millions of tweets and posts in minutes, extracting requests for help, identifying locations mentioned in messages, and filtering out the misinformation that predictably floods platforms during emergencies. Multimodal models that jointly process the text and attached images of a post can distinguish, for example, a genuine photograph of a flooded street from a recycled image circulating with false context. These streams provide something satellites cannot: the ground-level perspective of the people actually experiencing the event, often within seconds of impact.

Geospatial data offers the complementary virtue of synoptic coverage and physical precision. Satellite platforms deliver wide-area views that reveal the overall extent of flooding, fire, or structural destruction, while unoccupied aerial vehicles can be dispatched quickly to capture high-resolution imagery of specific sites. Deep learning pipelines built on this imagery support damage assessment, road network analysis for routing relief convoys, and change detection that compares pre- and post-disaster scenes to isolate what has been altered. Techniques such as deep fusion, in which information from multiple sensors or multiple moments in time is combined within a single learned representation, have become central to squeezing maximum value from these heterogeneous sources.

The core argument of the review is that these two domains, social media and geospatial sensing, have largely been studied by separate research communities with separate datasets, benchmarks, and processing pipelines. Yet during an actual disaster, responders must reason about both simultaneously. A flooded highway visible in satellite imagery gains operational meaning when combined with social media reports of vehicles trapped along it; a cluster of distress messages becomes verifiable when cross-checked against aerial imagery of the location. The authors propose a conceptual framework that consolidates the processing pipelines of both domains into a single, cohesive generative ecosystem, one in which inter-domain disaster data is used concurrently so that AI systems can develop a more comprehensive, human-like understanding of evolving disaster scenarios.

That ambition runs into sobering practical constraints, which the review catalogs candidly. Most existing implementations, despite substantial methodological advances, remain difficult to deploy in real-world disaster response settings. Timeliness is a recurring problem: models trained offline on curated datasets can be slow to adapt to an unfolding event whose imagery, vocabulary, and visual conditions differ from anything in their training data. Scalability is another, because processing the full firehose of crisis-related data at national scale demands infrastructure that many response agencies lack. Operational integration may be the deepest challenge of all, since research prototypes rarely connect to the software systems, workflows, and communication protocols that emergency managers actually use under pressure. The review identifies four categories of limitations spanning both domains, and its dataset archive is intended in part to help researchers benchmark against deployment-relevant conditions rather than convenient laboratory ones.

The implications for the disaster management community are considerable. As climate change intensifies floods, wildfires, and storms, and as urbanization concentrates populations in hazard-prone regions, the volume of both crisis data and crisis demand is rising together. A unified AI-driven pipeline that fuses citizen-generated signals with remote sensing could shorten the gap between an event’s onset and a coordinated response from days to hours or minutes. The authors’ emphasis on generative, ecosystem-level integration also points toward systems that do not merely classify incoming data but actively synthesize a coherent operational picture, prioritizing the locations, populations, and infrastructure most at risk and adapting as new information arrives from every source.

Published as open access with funding provided by UiT The Arctic University of Norway, the review is positioned as a resource for the entire field rather than a narrow technical contribution. By consolidating datasets, comparing model families, and mapping the research frontier from image classification toward real-time detection and human monitoring, it gives both newcomers and specialists a common map of where disaster informatics stands. The authors’ unifying vision is straightforward in principle: AI systems that see disasters the way skilled human responders do, from above and from the ground at once. Closing the gap between that vision and deployable practice, the review makes clear, is now the defining challenge for artificial intelligence in disaster management.

Subject of Research: Deep learning integration of social media and geospatial data for AI-driven disaster management

Article Title: Ai-driven disaster management: a comprehensive review of deep learning for social media and geospatial data integration

Article References: Sen, N., Chattopadhyay, S., Tripathy, R. M., Prasad, D. K., & Sekh, A. A. (2026). Ai-driven disaster management: a comprehensive review of deep learning for social media and geospatial data integration. Artificial Intelligence Review. https://doi.org/10.1007/s10462-026-11696-2

Image Credits: AI Generated

DOI: 10.1007/s10462-026-11696-2

Keywords: deep learning, disaster management, social media, satellite imagery, UAV, computer vision, remote sensing, damage assessment, object detection, real-time monitoring, data fusion, artificial intelligence