PICO
Models, input-, and output datasets range from raster-based descriptions of population distribution or built-up area, to complex datasets that describe people’s characteristics (e.g., gender, age and education) and detailed asset information (e.g., building material, number of floors, road types). Some models are local implementations, that are close to the ground truth and have a high-resolution, while others cover continents or even have a global reach. Some find their origins in grassroots activities, such as OpenStreetMap-based exposure models, while others rely on big data, through remote sensing and AI-driven methods, and are often created by larger agencies (e.g., the work of commercial parties like Google Open Buildings; WorldPop; or mixed organisations like Overture). The broad landscape of exposure is reflected in the wide variety of stakeholders, ranging from the insurance industry, to local and national governments, research institutes, the tourism sector and NGOs.
In this session we will welcome submissions addressing (1) geospatial methods and tools for the creation of exposure models, such as Volunteered Geographic Information or earth observation and AI models; (2) assessment of the quality or completeness of the data sources of exposure models, such as remote sensing, crowd-sourced, or official registry datasets; (3) exposure models for single hazards, for multi-hazard or hazard-independent contexts; (4) Comparison, validation and analysis of exposure models; (5) Cat model, insurance, government and financial exposure models and datasets; and (6) innovative applications of exposure models.
PICO: Tue, 5 May, 08:30–10:15 | PICO spot 1a
In disaster risk assessments, spatial population data are fundamental for determining exposure and social vulnerability. Traditionally, these analyses rely on static datasets, such as census records or population density maps. While accurate for representing nighttime distribution, these static models fail to capture the high levels of human mobility during the day. This can lead to significant under- or overestimations of the affected population, particularly in the event of sudden-onset disasters (e.g., flash floods, earthquakes, or critical infrastructure failure).
To address this, high-resolution temporal data are required. However, existing approaches often rely on real-world measurements, such as cell phone data or GPS positions. Such data is usually purchased from data companies, limiting its utility for both research and emergency management.
This study presents a spatio-temporal population model, initially developed for Germany's rich open data landscape, but with potential to be transferred to similar regions in the future. Using the city of Cologne as a primary test site, the model generates population maps for different time intervals throughout a typical weekday. The methodology employs iterative dasymetric mapping, integrating publicly available socio-demographic data, detailed building footprints, and the "Mobility in Germany" study. This approach ensures high transferability to other German metropolitan areas without requiring proprietary data.
The model estimates the number of individuals in each building across seven distinct time intervals (e.g., 8 am – 10 am, 10 am – 1 pm) and further disaggregates the population into socioeconomic groups (e.g., students, elderly). The results were validated against three independent datasets: emergency call volumes, the ENACT POP dataset, and mobile phone positioning data.
The model results enable refined disaster risk analysis by incorporating the temporal component of hazards and the corresponding population exposure. In the case of Cologne, this results in areas, such as the inner City at midday, with up to 5 times more exposed citizens than exposure analyses that rely on static data. Following the principles of Open Science, both the model code and the resulting datasets will be made publicly accessible to facilitate dynamic population assessments in other contexts.
How to cite: Priesmeier, P., Fekete, A., Haberl, M., Geiß, C., Baumhauer, R., and Taubenböck, H.: Spatio-temporal population exposure modeling for German cities, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8298, https://doi.org/10.5194/egusphere-egu26-8298, 2026.
Knowing people’s locations is crucial for successful disaster risk management. Within disaster risk research, geolocalized mobile phone data are increasingly recognized in recent years for providing an efficient and cost-effective way to quickly and precisely assess population movement patterns. Alongside location information, additional sociodemographic data such as age and gender are often provided, enabling valuable insights into the susceptibility of individuals to natural hazards. The widespread use of mobile phones, which continuously generate large amounts of real-time data, allows the study of time-dependent differences in human exposure, vulnerability, and consequently risk for large parts of the world’s population.
In this research, geolocalized mobile phone data, capturing exposure and vulnerability represented by user age, were examined as dynamic risk-related components to assess the spatiotemporal variations of landslide risk in the Wipp Valley, Tyrol, Austria. As a representative example, day- and week time dependent spatial differences in risk were analysed for several days in May and June 2024. Since landslide susceptibility values remained static, daily and weekly variations of population movements and the spatial distribution of vulnerable groups caused specific risk patterns.
This knowledge may contribute significantly to disaster risk analysis in the future, underlining the potential of geolocalized mobile phone data for disaster risk management.
How to cite: Kussegg, J., Wenzel, T., and Glade, T.: Assessing spatiotemporal dynamics of exposure and vulnerability using geolocalized mobile phone data – explored for landslide risks in the Wipp Valley, Austria, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7456, https://doi.org/10.5194/egusphere-egu26-7456, 2026.
Fine-scale exposure information is essential for natural hazard risk assessment, particularly in urban environments where vulnerability and hazard intensity can vary substantially within short distances. Building-level exposure data support a range of applications, including identification of priority areas for emergency response, estimation of shelter demand, and the development of more targeted early-warning and preparedness strategies. Although Brazil’s national census datasets are robust for demographic analysis, their spatial resolution, typically 200-meter grids or coarser, limits their use in detailed exposure assessments. In addition, high-resolution building datasets remain limited or unaffordable in many developing regions, especially outside major metropolitan centers, underscoring the need for reproducible methods based on openly available geospatial information. This study presents a methodology to disaggregate residential population from census grids to individual building footprints by integrating several complementary open datasets: (1) OpenBuildingsMap and GlobalBuildingAtlas footprints to obtain building geometry and attributes; (2) OpenStreetMap (OSM) for road network geometry and attributes; (3) the National Registry of Addresses for Statistical Purposes from the Brazilian Institute of Geography and Statistics (IBGE), used as georeferenced Points of Interest (POIs) classified by establishment type; and (4) population counts at 200-meter resolution from the 2020 IBGE Statistical Grid. Together, these datasets yield a scalable, transparent, and replicable exposure model tailored to Brazilian urban contexts. The proposed method adapted a weighted scoring framework in which residential building-level population allocation is driven by both physical building characteristics (floor area and height) and a POI-based residential attractiveness index. POI relevance weights were computed using Term Frequency–Inverse Document Frequency metrics and Pearson correlation between POI categories and population totals within each census grid. We applied a Gaussian Network Kernel Density Estimation along OSM road segments to propagate POI influence and derive an attractiveness score for each segment. Buildings were then linked to the nearest road segment, and their attractiveness scores were multiplied by their physical attributes to obtain a composite allocation weight. Two population distribution strategies were evaluated: the proposed POI-integrated method and a baseline model relying solely on building physical characteristics. The distribution was assessed through a flood-exposure analysis for a potential dam-breach scenario downstream of the Ibirité Dam, located in Minas Gerais, Brazil. We focused on the population potentially affected within the Self-Rescue Zone (SRZ), an area requiring immediate evacuation in the event of a failure. Across 169 directly affected 200-meter resolution census grids, a total of 8,578 residents were identified. Within the SRZ, the POI-integrated method estimated 3,124 residents, compared with 3,212 residents under the baseline approach. Results indicate that incorporating POI-based attractiveness produces more realistic spatial population patterns, particularly in mixed-use neighborhoods and areas with heterogeneous building typologies, enabling more accurate classification of flood hazard exposure. Future work includes sensitivity analysis, field validation, comparison with alternative disaggregation approaches, incorporation of demographic attributes, expansion to other occupancy types, and evaluation of methodologies to improve building-footprint geometry and attribute accuracy.
How to cite: Rocha da Silva, A. F., Cardoso Eleutério, J., Krause Camilo, B., and Ferreira Rodrigues, A.: Integrating open data for high-resolution residential population disaggregation and flood exposure assessment in Brazil , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-870, https://doi.org/10.5194/egusphere-egu26-870, 2026.
The Indian territory coastlines, which are some of the most populated areas in the world, are prone to tsunamis generated from subduction zones such as Makran, the Northern part of the Sunda trench and submarine landslides. Within the project “People-centered tsunami early warning for the Indian Coastlines (PCTWIN)”, we are striving to forecast not only the hazard but also the impact, focusing on impact on the population through modelling of human exposure in different spatio-temporal scales.
Human exposure maps transform coastal risk into actionable information, enabling smarter planning, appropriate decision-making, and stronger resilience against tsunami hazards. In PCTWIN, human exposure maps are mainly required for developing (1) site-specific Probabilistic Tsunami Risk Analysis (PTRA) maps to estimate the distribution of population at risk for different return periods; (2) Impact Forecasting to define the number of people being affected by the tsunami.
To this end, the Human exposure maps for the whole Indian Coastlines are being developed. The national coastal human exposure maps will map the population at domicile, with a resolution of 100 meters. These maps are developed through a top-down census-based approach using the Python-based software Popcorn https://popcorn-population.github.io/. It is a population mapping workflow that employs the globally available satellite images from Sentinel-1 and Sentinel-2, and the number of aggregate population counts over coarse census districts for calibration. The building occupancy is trained through Deep Learning algorithms with coarse census counts. Herein, we have employed the 2011 Indian census data, while the population is projected based on growth rates estimated by the UN World Urbanization Prospects Database (UNPD). Preliminary comparison with the surveyed population data for selected coastal areas by INCOIS (Indian National Centre for Ocean Information Services) are promising. We are also going to compare human exposure maps developed at national scale with other open-source exposure databases.
How to cite: Ebrahimian, H., Ghaffarian, S., Jalayer, F., Ranganalli Somashekharappa, M., Metzger, N., Kim, G., Galasso, C., and Khairnar, G.: Human exposure maps for Indian coastlines, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15067, https://doi.org/10.5194/egusphere-egu26-15067, 2026.
How to cite: Dimasaka, J., Geiß, C., and So, E.: METEOR 2.5D: An Open Geospatial Dataset of the Spatiotemporal Evolution of Physical Vulnerability in UN-recognized Least Developed Countries (as of 2020) at Five-year Intervals, 1975-2030, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1057, https://doi.org/10.5194/egusphere-egu26-1057, 2026.
Tourism is one of the largest economic sectors globally contributing to 10% of the world’s GDP and also 1 in 10 jobs, however, there is comparatively little standardized data spatially for tourism globally. Around the world, there exist many approaches to collecting statistics for tourism across a country and many disparate sources: some countries have subnational data collection yearly, others collect certain parameters, others at a national level, but no standardized way globally to aggregate the statistics appropriately.
Tourist accommodation stats from open data sources include region/province and even district-level cuts in some countries, in others there are unique tourism regions different to administrative level boundaries within the country. A key part of this work is the collection of the GIS layer associated with these boundaries in order to use the collected statistics within each country.
Although there exist a lot of products using raster inputs like nighttime lights, population proxies, global vector inputs of hotel points (where available) and partial data such as OSM globally, as well as aggregated statistics at a national level via UNWTO, WTTC etc., this work is the first known global subnational level set of official country-by-country, region-by-region tourism statistics using tourism boundaries for use in risk modelling.
The analytics allow for checks of global datasets, as well as vice versa with the statistics coming from each country office given the spatial consistency.
Over 3,000 tourism regions are characterized as part of this work, with many more destinations globally saved. Millions of hotels, overnight stays and other statistics have been and continue to be added to the databank. This database forms the basis for risk modelling across regions and destinations speaking the same language as the tourism industry.
This work builds upon Daniell et al. (2025) with a Europe-wide tourism destination socioeconomic risk model for tourism and is a companion abstract to Schaefer et al. (2026) characterizing the development of a 1km global tourism hazard and risk screening classification.
How to cite: Daniell, J., Schaefer, A., Brand, J., Romero, R., Maier, A., Girard, T., Khazai, B., Michalke, S., Claassen, J., and Daniell, J.: The first Global Tourism Region statistics database for risk and exposure modelling, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8853, https://doi.org/10.5194/egusphere-egu26-8853, 2026.
The built environment is, globally speaking, the largest unknown in the understanding of the effects of disasters and in assessing their risk. This includes not only the location of buildings but also their size, occupancy, structural type, vulnerability and value. Detailed knowledge of it is necessary for many tasks in disaster risk reduction but also in other fields, e.g. climate-related sustainability, urban planning and management, insurance and re-insurance. While in well-regulated countries cadastral data is available that provides various details about the buildings, in most parts of the world such information is lacking. In some areas not even the locations of buildings and settlements are known to the authorities. Buildings, the core part of the built environment, can be strongly mixed within small areas in their structural types, sizes, shapes and number of people in them and the socio-economic structure can vary highly on these scales. This heterogeneity cannot adequately be described by classical exposure models that provide aggregated building data over larger areas.
A global model describing the built environment at the scale of individual buildings has never been achieved, nor has such a model been dynamic, with continuous updates reflecting changes in input data. Here, we present a global, building-level resolution, open, reproducible and dynamic exposure model with the aim to provide global exposure data on the building level. This model is based on volunteered geographic information, predominantly OpenStreetMap and open data that is created with earth observation and machine learning, e.g. the building footprints of the Google Open Buildings and Microsoft ML Building Footprints, and the Global Human Settlement Layer to estimate the extent of built area. Further datasets like EUBucco and full 3D building geometries are added where available and the height information covering approx. 70% of all buildings is used to further create 3D models at the Level-of-Detail 1. The distribution of different structural types of buildings per region are taken from open aggregated exposure models or developed from cadastral data. Every building is assessed separately and its exposure indicators are computed deterministically, where possible, or probabilistically. This level of detail is necessary when it comes to localized hazards, such as strong shaking of earthquakes, floods or tsunamis due to local site conditions. In particular 3D buildings are now becoming part of the next-generation seismic risk framework. The model covers every country and territory globally and is to a large degree building complete with approx. 3 billion buildings described in detail.
How to cite: Schorlemmer, D., Oostwegel, L. J. N., Calliku, D., de la Mora Lobaton, P., Evaz Zadeh, T., Lingner, L., and Rao, C.: Every building on Earth – The Global Dynamic Exposure model, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5681, https://doi.org/10.5194/egusphere-egu26-5681, 2026.
Exposure models such as ESRM20 provide building exposure to seismic hazards per district or geocell. These kinds of models hold a great amount of information about building types and construction in different countries, but remain primarily relevant to fields within seismology. While some of the data they hold could be used interdisciplinarily, incorporation of other data points could improve its relevancy for climate-change related hazards such as heat waves and cold fronts. For countries within Europe, which belong to the ESRM20 models, there are two datasets which provide the necessary data and the link to tie them together, TABULA and the European Building Stock Observatory (EUBSO). The building typologies of these datasets were mapped to match the taxonomy of the exposure models in order to include the relevant data.
TABULA is a European dataset for residential buildings’ energy efficiency capabilities and characteristics. The dataset lacks cohesion and homogeneity across the 21 countries it covers and for some countries, there is no clear distribution for the total number of buildings per class. Without a probability distribution of building types, it is not possible to combine TABULA with the seismic exposure models, as the definition of the classes in each taxonomy does not overlap. As an example, TABULA has classes for years of construction while the seismic exposure models have none; if the number of buildings in each class is not known, the models cannot be combined. The EUBSO can bridge that gap. This dataset has similar building type classifications as TABULA but with the number of buildings in each of its classes and some data on the energy efficiency of buildings, including for non-residential ones, filling the gap left by TABULA.
We combined the three datasets in two stages. First, the exposure models are improved using the more detailed occupancy and construction year descriptions from the EUBSO. Afterwards, the building types in the models match with those from TABULA, and each feature in the exposure model can be linked to a TABULA class. Finally, these enriched models are used, along with other sources such as OpenStreetMap, in the creation of the Global Dynamic Exposure model (GDE). This is a building by building model of the entire world with seismic and climate exposure data. This includes buildings' resilience to seismic activity and several points concerning energy efficiency such as CO2 emissions, energy required for heating, and how much different building components resist the flow of heat. While this dataset may be used for exposure, it can also serve as a digital twin of the building stock in projects for urban development, and improve understanding of cities and how they are built. This dataset is being used for the Local Digital Twins Toolbox initiative by the European Union which provides cities with urban development tools. With an increase in smart cities initiatives and a search for ways to improve the sustainability in European municipalities on the rise, this dataset provides a detailed base understanding of the buildings that are in them.
How to cite: de la Mora Lobaton, P., Schorlemmer, D., Oostwegel, L., Çalliku, D., Rao, C., Evaz Zadeh, T., and Lingner, L.: Enriching Seismic Exposure Models to Create a Multipurpose Building Model, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20847, https://doi.org/10.5194/egusphere-egu26-20847, 2026.
Recent urban earthquakes and rapid urbanization have intensified the demand for fine-scale building exposure information in disaster risk assessment. However, existing approaches for high-resolution building exposure extraction often suffer from limited data completeness, insufficient semantic detail, and weak update capability, particularly at detailed spatial scales. Moreover, traditional methods relying on homogeneous data sources and static classifications struggle to represent the heterogeneity of urban building exposure.
To address these limitations, we propose a multi-source data-driven framework combined with machine learning to extract high-resolution building exposure information, focusing on building function and building height. Building function types are inferred by integrating OpenStreetMap building footprints with time-series mobile signaling data, exploiting differences in population activity patterns across day-night and workday-non-workday periods. Machine learning techniques are then applied to identify clusters of buildings with similar population dynamic characteristics, enabling the inference of building function types. Building height is extracted from bi-temporal Sentinel-2 imagery by capturing variations in image brightness induced by seasonal differences in building shadow length, and a random forest model is employed to learn the nonlinear relationship between image features and building height, thereby reducing reliance on very high-resolution imagery and manual interpretation.
Case studies in representative Chinese cities indicate that the integration of multi-source data and machine learning enables more effective use of data for different building exposure attributes, resulting in improvements in spatial detail, attribute completeness, and data timeliness. Population-dynamic-based building function identification provides an activity-oriented characterization of building use, while building height estimation based on freely available Sentinel-2 imagery offers a cost-efficient and scalable approach. Overall, these findings suggest that multi-source data integration and machine learning can support large-scale, high-resolution urban building exposure mapping.
How to cite: Nie, W., Fan, X., Wang, J., wang, L., Qi, Y., Liu, M., Lu, F., Oostwegel, L., and Schorlemmer, D.: Multi-source data and machine learning supporting high-resolution building exposure extraction, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9671, https://doi.org/10.5194/egusphere-egu26-9671, 2026.
Disaster risk is commonly represented as the interaction between hazard, exposure, and vulnerability. The accuracy of disaster risk assessments largely depends on the level of detail, diversity of attributes, and temporal dynamics represented in the exposure model. However, multimodal datasets—spanning crowd-sourced data like OpenStreetMap (OSM), official building registries, cadastral records, national statistics, AI-generated building data, and remote sensing—remain fragmented . They are heterogeneous in structures, scales, and resolutions creating challenges for seamless integration and consistent interpretation. The proposed method incorporates the high-resolution UAV mapping results into the Global Dynamic Exposure Model (GDE), leveraging diverse data sources for more robust disaster management. Unlike conventional data sources, UAV mapping technologies and derived building information can capture rapid spatial and temporal changes, significantly enhance the completeness, accuracy of multimodal exposure datasets. These benefits are most evident in a high-resolution, local-scale exposure modeling.
UAV mapping provides high-resolution orthophoto imagery and dense 3D point clouds as primary data sources. The orthophoto imagery enables the extraction of complete and accurate building footprints, which are used to improve and update existing building geometries, and identify newly constructed buildings that are absent from these sources. The 3D point clouds capture detailed building heights and geometric forms, allowing the generation of Level of Detail (LoD) 2+ 3D building models that serve as geometric enrichment for GDE. Furthermore, building attributes such as roof shape, number of stories, volume, and construction materials can be derived deterministically, rather than estimated as is commonly required in open datasets. By substantially reducing uncertainties in building asset representation, the proposed approach significantly enhances the accuracy and reliability of disaster risk assessments. The approach can further extend to post-disaster UAV surveys which allow rapid assessment of damaged areas and direct comparison with the most updated model before the disaster. Changes in height, volume, façades or roof condition can capture structural deformation and collapse indicators for loss evaluation and recovery planning.
Beyond geometries characteristics, UAV-derived orthophotos and point clouds provide detailed information on building geometry, height, roof form, and signs of recent modification, which characterize exposure-relevant attributes. For example, irregular roof shapes may indicate building extensions or mixed use, while large footprints with multiple entrances suggest functional subdivision or vertical complexity. Targeted field surveys, supported by tools such as StreetComplete, Field Tasking Manager from the Humanitarian OpenStreetMap Team (HOT), and KartaView (street-level photography), are conducted to augment these UAV-derived indicators in the datasets. The resulting semantic building information combined with the 2D and 3D geometries serve as the up-to-date representation, which is the essential core of a local digital twin. By integrating UAV-mapped builidng geometries with the on-site observations in OSM, together with the other datasets, the exposure modeling framework embeds local knowledge into building entities, establishing a full-scale and transferable workflow from data acquisition to exposure model enrichment. Case studies in Glückstadt, Germany, and Nairobi, Kenya, demonstrate its applicability for high-resolution, dynamic exposure modeling.
How to cite: Rao, C., Schorlemmer, D., J.N. Oostwegel, L., Çalliku, D., and de la Mora Lobaton, P.: UAV-Enhanced Multimodal Exposure Modeling for the Global Dynamic Exposure Model, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21400, https://doi.org/10.5194/egusphere-egu26-21400, 2026.
Geospatial data is essential for disaster risk assessment but it is often fragmented across independent taxonomies such as PAGER, GEM, OASIS, and the Building Stock Observatory. Each of these frameworks provides structured and detailed information, yet differences in schemas and terminology limit integration and broader reuse. OpenStreetMap (OSM), as a widely adopted open geospatial platform, offers a practical baseline for integration. Its surrounding ecosystem includes a rich set of tools that are critical for humanitarian mapping such as JOSM and Tasking Manager that can integrate the exposure-related features.
Aligning diverse building taxonomies with OSM enables structured datasets to be compared and cross-referenced within a common framework, but it also requires balancing different levels of detail. Not all information used in exposure or risk modeling is useful for the mapping community, as they concentrate on visible features. Attributes such as population or structural value are critical for exposure analysis, but often are based on estimates derived from regional statistics and based not on mapping in the ground. So, we use the OSM tools and tagging standards to provide the semantic backbone, while exposure-related information is integrated through controlled, range-based tags that remain compatible with OSM practices and reflect inherent uncertainty.
This is done through tagging presets that are defined for both physical building characteristics and exposure-related attributes. Observable features such as material, height class, and occupancy follow established OSM conventions, while complementary exposure presets allow contributors to assign population and structural value ranges based on reference values from the Global Dynamic Exposure project. These exposure-relevant values provide a consistent starting point but can be refined using local statistics or expert judgment. For example, a residential masonry building mapped in OSM can be tagged with its material and height class, and additionally assigned a population range and a structural value class derived from regional reference estimates. The OSM-relevant information is pushed to the open dataset, and the refined exposure information can be used to estimate risk or damage in a specific area.
By embedding this workflow into existing OSM editors, humanitarian organizantions and institutions can use familiar tools to efficiently map areas and characterize exposure, improving data consistency and supporting disaster risk assessment, humanitarian response, and resilience planning.
How to cite: Calliku, D., Schorlemmer, D., Oostwegel, L. J. N., de la Mora Lobaton, P., Rao, C., Evaz Zadeh, T., and Lingner, L.: Simplifying Mapping for Building Exposure using OpenStreetMap Tools, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21319, https://doi.org/10.5194/egusphere-egu26-21319, 2026.
Exposure models that provide up-to-date, spatially explicit information on buildings’ vulnerability-relevant characteristics are key to effective disaster mitigation and risk management. As (i) different building attributes influence vulnerability to different natural hazards, and (ii) natural hazards vary in spatial scale and exhibit distinct spatial patterns, holistic multi-risk assessments place particularly high demands on thematic detail and spatial resolution. A generic yet detailed representation of the building stock enhances the flexibility of risk models to consistently address diverse hazard scenarios. However, given the vast number of buildings, their structural heterogeneity, and high spatio-temporal dynamics, maintaining a comprehensive inventory across large areas remains a complex challenge. The rapid transformation of disaster risk regimes due to global change, coupled with limited exposure data, necessitates automated, data-driven approaches to efficiently infer building vulnerability at scale. This work investigates the potential of heterogeneous, multimodal geospatial image data—including street-level imagery (SLI), very high-resolution optical remote sensing data, and a normalized digital surface model—for generic building characterization using deep learning. To infer multiple building attributes from multimodal inputs, we introduce a deep multimodal multitask classification framework. It incorporates a feature-level fusion module designed to optimally exploit synergies among data modalities within a multitask learning setting. The common challenge of missing SLI is addressed through a dedicated submodel that learns spatio-contextual representations from available SLI as substitutes. Using the earthquake-prone metropolis of Santiago de Chile as a case study, we evaluate the contribution of the employed geo-image modalities and the proposed methods to the reliable inference of five structural target variables: building height, lateral load-resisting system material, seismic building structural type, roof shape, and block position. Our results demonstrate that integrating ground-based and top-view geo-image data with tailored deep learning models offers a promising path toward the automated generation of detailed, area-wide exposure models.
How to cite: Aravena Pelizari, P., Geiß, C., and Taubenböck, H.: Advancing Building Exposure Modeling at Scale through Multimodal Geo-Imagery and AI, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1641, https://doi.org/10.5194/egusphere-egu26-1641, 2026.
Roof geometries, such as slopes, orientations and overhangs, play a key role in defining vulnerability to cyclonic winds as it directly informs the pressure and uplift forces applied to buildings. However, these parameters are not available at a large scale especially in French overseas territories (DOM-TOM) particularly exposed to cyclonic winds and where building databases often lack sufficient geometric details. The objective of this work is to establish a workflow to estimate building vulnerabilities to cyclonic winds at a territorial scale using three roof-related parameters.
Using freely available airborne LiDAR data acquired at a large scale and distributed by the Institut Géographique National (IGN), the proposed approach takes account of the current limitations of the equipment used for the description of buildings. Limitations include: acquisition angle and point density leading to incomplete wall sampling and planimetric uncertainties on the order of fifty centimeters that reduces the object discrimination capacities.
LiDAR point clouds are used to describe buildings through three classes, walls, roofs and rooftop objects (chimney, technical equipment). LiDAR points are spatially associated with database buildings' polygons IDs, and they are used to reconstruct buildings' footprints to avoid spatial issues during analysis. Wall and roof points are then used to compute parameters.
- Orientation and slope can be defined by removing walls and rooftop objects using elevation within buildings’ footprints. Statistical analysis can be finally used to describe the roof into categories such as dominant roof orientation, the number of distinct roof orientations, and slope gradient. LiDAR intensity may also provide coarse information on roof material type.
- Roof overhang estimation remains more sensitive to wall point density and precision of the equipment used. Walls are reconstructed using density-based clustering (DBSCAN) combined with line-fitting (RANSAC and Hough transform) enabling the extraction of geometric features from heterogeneous LiDAR data distribution. Using airborne LiDAR compared to terrestrial increases the number of faces that can be detected but also lowers the global quality of the result.
Resulting indicators are intended to improve and to complete existing databases at a large scale with relevant details on wind vulnerability. The proposed workflow is meant to be reproducible, scalable to large areas, it is intentionally data-driven and designed to benefit from ongoing improvements in LiDAR acquisition and classification. Current limitations primarily arise from point density and classification quality. Improvements in these parameters would enable more accurate wall reconstruction, roof object discrimination (chimneys, technical equipment), and roof–façade separation, ultimately leading to more reliable vulnerability estimates.
How to cite: Ancian, P. and Pugnet, L.: Large-scale estimation of roof geometry indicators for wind vulnerability assessment using airborne LiDAR , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10016, https://doi.org/10.5194/egusphere-egu26-10016, 2026.
Europe’s critical infrastructure (CI), including energy, transport, communication, waste, and water infrastructure, is increasingly exposed to climate extremes, such as coastal flooding. Despite progress in addressing projections of extreme sea levels under climate change, future exposed population or exposed gross domestic product (GDP), limited research has attempted to address future risks on projected CI exposure. This study develops spatially explicit future scenarios for CI across five Shared Socioeconomic Pathways (SSPs) for Europe and the United Kingdom in 2030, 2050, and 2100. Future infrastructure density (FutureCISI) is estimated using covariates associated with infrastructure development, including land cover, GDP, population, elevation, and inland water. The study evaluates four modelling approaches: a regression, a random forest, a convolutional neural network and a vision transformer. The projections are then used to assess changes in infrastructure exposure within the coastal floodplains across the SSPs and time frames.
How to cite: De Plaen, J.: FutureCISI: Spatially Explicit Projections of Critical Infrastructure Consistent with the Five SSPs, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7575, https://doi.org/10.5194/egusphere-egu26-7575, 2026.
