Estimating urban flooding depth by integrating multimodal image-text data: A segment-level direct preference optimization-based multimodal large language model

Accurately estimating urban floodwater depth is a critical step in enhancing urban resilience and strengthening disaster prevention and mitigation capabilities. Traditional methods relying on hydrological monitoring stations and numerical simulations suffer from limitations such as sparse spatial coverage, insufficient validation data, limited accuracy, and delayed fast performance. In contrast, social media data—characterized by its vast volume and fast availability, can effectively compensate for these shortcomings. When processed using artificial intelligence (AI) algorithms, such data can significantly improve credibility, disaster perception speed, and water depth estimation accuracy. To address these challenges, this paper proposes a robust and widely applicable method for rapid urban flood depth perception. The approach integrates AI technology and social media data to construct an AI framework capable of perceiving urban physical parameters through multimodal big data fusion without costly model training. By leveraging the near real-time and widespread nature of social media, an automated web crawler collects flood images and their textual descriptions (including reference objects), eliminating the need for additional hardware investments. The framework uses predefined prompts and pre-trained models to automatically perform relevance verification, duplicate filtering, object detection, and feature extraction, requiring no manual data annotation or model training. With only a minimal amount of water depth annotated data and compressed cross-modal feature vectors as training input, a lightweight Multilayer Perceptron (MLP) achieves high-precision depth estimation based on reference objects. This method avoids the need for large-scale model fine-tuning, allowing rapid training even on devices without GPUs. Experiments demonstrate that the proposed method reduces the Mean Square Error (MSE) by over 80%, processes each image in less than 0.5 s (more than 20 times faster than existing large-model approaches), and exhibits strong robustness to changes in perspective and image quality. The solution is fully compatible with existing infrastructure such as surveillance cameras, offering an efficient and reliable approach for fast flood monitoring in urban hydrology and water engineering applications.

Improved flood depth estimation with SAR image, digital elevation model, and machine learning schemes

Increasing frequency and severity of urban flooding, driven by climate change and urban population growth, present major challenges. Traditional flood control infrastructure alone cannot fully prevent flood damage, highlighting the need for a comprehensive and multi-dimensional disaster management approach. This study proposes the Flood Risk Index for Building (FRIB)—a building-level assessment framework that integrates vulnerability, hazard, and exposure. FRIB assigns customized risk levels to individual buildings and evaluates the effectiveness of a multi-dimensional method. Compared to traditional indicators like flood depth, FRIB more accurately identifies high-risk areas by incorporating diverse risk factors. It also enables efficient resource allocation by excluding low-risk buildings, focusing efforts on high-risk zones. For example, in a case where 5124 buildings were targeted based on 1 m flood depth, applying FRIB excluded 24 buildings with “low” risk and up to 530 with “high” risk, reducing unnecessary interventions. Moreover, quantitative metrics like entropy and variance showed that as FRIB levels rise, flood depth distributions become more balanced—demonstrating that depth alone does not determine risk. In conclusion, while qualitative labels such as “very low” to “very high” aid intuitive understanding, FRIB’s quantitative, multi-dimensional approach enhances precision in urban flood management. Future research may expand FRIB’s application to varied regions, supporting tailored flood response strategies.

Introduction Urban flood modeling requires accurately representing building structures, as they influence flow paths, flood depths, and hydraulic dynamics in urban basins. However, the comparative performance of commonly used building representation techniques within rain-on-grid applications remains insufficiently quantified. This study assesses the effectiveness of three building representation techniques—Stubby Building (SB), Building Block (BB), and Building Resistance (BR)—in the urbanized Podoniftis basin, Attica, Greece. Methods Using a 2 m resolution digital elevation model within a rain-on-grid framework, buildings are represented with BB using 5 m and 12 m building footprint elevations, SB using a 0.5 m elevation, and BR using increased Manning’s roughness coefficients of 1 and 10 over building footprints. Results The different methods substantially influence simulated flood extent, water depth, and velocity fields. The BB method yields the highest mean depths (0.62 m) but tends to concentrate high-water depths adjacent to buildings in densely built-up areas. The BR method produces a larger flooded area with the lowest mean velocity (0.27 m/s), reflecting higher hydraulic resistance and a more distributed water storage. The SB method provides an intermediate behavior between the three approaches, while not explicitly resolving runoff generation and redistribution associated with building ingress and drainage systems. A Flood Hazard Rating assessment further shows that building representation affects the spatial pattern of hazard classes. Discussion The findings highlight that building representation is a key modeling choice in rain-on-grid applications, with each technique producing distinct flood hazard patterns that can influence risk management decisions. These results underline the need for more refined techniques that explicitly incorporate drainage networks and building ingress processes for urban flood hazard assessment.

Abstract. Rapid and accurate flood assessment is crucial for effective disaster response, rehabilitation, and mitigation strategies. This study presents a fully automated framework for floodwater delineation and depth estimation using the Earth Observation Satellite 4 (EOS-04) (Radar Imaging Satellite, RISAT-1A) synthetic aperture radar (SAR) imagery and a digital elevation model (DEM). This is the first study to apply the established automatic-tile-based segmentation method and the height above the nearest drainage (HAND) tool to EOS-04 data for flood extent delineation. For flood depth estimation, this study introduces a novel application of the trend surface analysis (TSA) technique, enabling rapid and data-efficient assessment. Unlike traditional hydrodynamic models that demand extensive datasets and computational resources, TSA operates using only the inundated water layer and DEM, providing a highly data-efficient solution. The methodology is applied to flood-prone regions in Andhra Pradesh, Assam, Bihar, and Uttar Pradesh, India. Validation of flood extent against optical data demonstrates accuracy greater than 90 %. Flood depth estimation using TSA is validated by comparing water depths derived from river gauge stations with real-time field measurements and results from the floodwater depth estimation tool (FwDET). The TSA method achieves a root-mean-square error (RMSE) of 0.805, significantly outperforming FwDET's RMSE of 5.23. This integration of high-resolution SAR imagery and DEM represents a transformative, automated solution for real-time flood monitoring and depth estimation, enhancing disaster management capabilities.

Glasshouse experiments were conducted from March to July 2003 to determine the effects of sowing depth, and the time, duration, and depth of flooding on the emergence, survival, and growth of Fimbristylis miliacea (L.) Vahl. The treatments that were evaluated in the first experiment were three seeding depths (0, 0.5, and 1.0 cm), while in the second experiment, three flooding depths (saturated soil with no standing water and soil with water depths of 5 and 10 cm) and three flooding durations (7, 14, and 21 days from sowing) were evaluated. In the third experiment, three flooding depths (saturated soil with no standing water, and soil with water depths of 5 and 10 cm) and four flooding regimes (7, 14, 21 or 28 days after sowing [DAS]) were evaluated. Surface seeding gave the highest emergence rate compared to the 0.5 cm and 1.0 cm soil depths. A significantly higher emergence rate was recorded with the saturated conditions than with the flooded conditions. A flooding duration of ≥ 14 days showed a clear trend of reduced emergence with increasing flooding depth. A significantly higher survival rate, plant height, root length, number of leaves, and dry matter were recorded at soil saturation followed by the 5 cm and 10 cm flooding depths when flooding was simulated at 7 and 14 DAS. When the flooding was delayed to 21 and 28 DAS, the 10 cm flooding depth was required to suppress this weed. The results provide sufficient evidence to confirm that from deeper seed burial (1 cm sowing depth), flooding depths of ≥ 5 cm of durations of 14 and 21 days and at the onset of flooding within 14 DAS were effective in suppressing the emergence and growth of F. miliacea.

PurposeDespite the extensive academic interest in social media sentiment for financial fields, multimodal data in the stock market has been neglected. The purpose of this paper is to explore the influence of multimodal social media data on stock performance, and investigate the underlying mechanism of two forms of social media data, i.e. text and pictures.Design/methodology/approachThis research employs panel vector autoregressive models to quantify the effect of the sentiment derived from two modalities in social media, i.e. text information and picture information. Through the models, the authors examine the short-term and long-term associations between social media sentiment and stock performance, measured by three metrics. Specifically, the authors design an enhanced sentiment analysis method, integrating random walk and word embeddings through Global Vectors for Word Representation (GloVe), to construct a domain-specific lexicon and apply it to textual sentiment analysis. Secondly, the authors exploit a deep learning framework based on convolutional neural networks to analyze the sentiment in picture data.FindingsThe empirical results derived from vector autoregressive models reveal that both measures of the sentiment extracted from textual information and pictorial information in social media are significant leading indicators of stock performance. Moreover, pictorial information and textual information have similar relationships with stock performance.Originality/valueTo the best of the authors’ knowledge, this is the first study that incorporates multimodal social media data for sentiment analysis, which is valuable in understanding pictures of social media data. The study offers significant implications for researchers and practitioners. This research informs researchers on the attention of multimodal social media data. The study’s findings provide some managerial recommendations, e.g. watching not only words but also pictures in social media.

Water-related disasters in urban area, especially urban floods have become more frequent and severe. This leads to loss of life, infrastructure damage, business interruption as well as difficulties in conducting daily activities. Phnom Penh, the capital city of Cambodia, has frequently experienced significant rainfall-flood events during rainy season. Without proper prevention and mitigation and management of urban drainage system, Phnom Penh is expected to confront with the current and future challenge of water-related disaster. To address this urban flood issue, flood modeling could be used to assess flood severity and make a visual representation of the urban flood hazard as basic information for land-use planning and limiting development in flood-prone areas. This study, therefore, aims to simulate flood events in a downtown area with the complex storm drainage system in Phnom Penh using FLO-2D model. The model was successfully calibrated and validated with time series data of water level in 2018 and estimated flood depth based on photo visualization through social media in 2015. The values of NSE, RMSE, and R2 were 0.76, 0.13, and 0.91 respectively. After model calibration, the model was used to simulate the flood flow through the channels, and floodplain within the study area in terms of flooding depth for 2-, 5-, 10-, and 25-year return periods of rainfall. The flood duration was determined by overlaying three flooding maps considered based on hydrograph components: rising flood, peak flood, and receding flood map. The result was delineated into flood hazard map defined by five categories, ranging from very low to very high. The result of flood hazard assessment shows that the floodplain under very low hazard category is constantly the highest one among all categories. However, with higher return period the flooding area under very low hazard category decreases while other hazard categories increase, especially medium hazard category increases significantly. The areas were found to have highest level of hazard are areas around Olympic stadium, areas around Royal Palace (along the river side), and areas in Beoung Trabek district. Besides, the simulated results show that the drainage system improved by JICA is capable to deal with the flood problem from the rainfall of 2-year return period in the study area.

<p>Urban flood mapping using SAR<strong> </strong>is an important tool for emergency flood incident management and improved flood forecasting. We have recently developed a method for detecting urban flooding using Sentinel-1 and WorldDEM data<sup>1</sup>. This is a change detection technique that estimates flood levels using pre- and post-flood images. It searches for increased backscatter in the post-flood image due to double scattering between water and adjacent buildings, compared to that in the pre-flood image where double scattering is between unflooded ground and buildings. If <em>φ</em> is the angle between building and satellite direction of travel, double scattering is strongest for low <em>φ</em>, and falls off as <em>φ</em> increases. It also depends on the building height and length, the depth of flooding, the roughness of the ground surface, and the complex dielectric constants of the building wall and ground surface.</p><p>Ref. 2, modelling X-band data, concluded that the increase of double scattering was only high if buildings were roughly parallel to the flight direction. The modelling assumed isolated buildings, and in a complex urban environment any increase would be further masked due to adjacent buildings. This implies a limitation in our method, since if the falloff with <em>φ</em> is very rapid, this could reduce the number of flooded double scatterers detected.</p><p>We used the model of ref. [3] to estimate the post- to pre-flood radar cross section (RCS) ratio for double scatterers in Sentinel-1 C-band images. In agreement with ref. [2], this predicted that high ratios would only be obtained from building walls with <em>φ</em> < 10°.</p><p>However, there are limitations in the models, and as a result it was decided to carry out an empirical study to examine the relationship between the RCS ratio and <em>φ</em>. This was based on S-1 data from the UK floods of winter 2019/2020, using flooding in Fishlake as an example of flooding in moderate housing density, and flooding in Pontypridd as an example of flooding in dense housing. A LiDAR DSM was used to allow accurate measurement of <em>φ.</em></p><p>Our results showed that, as well as flooded double scatterers (DSs) with <em>φ </em>< 10°, a significant number of flooded DSs with 10° < <em>φ </em>< 30° also produced a high RCS ratio. Our method also benefited from the predilection for building houses facing south in the northern hemisphere. As the S-1 sensor is in polar orbit, descending/ascending passes image the east/west walls of a house at low <em>φ </em>values<em>.</em> Similar arguments hold in the southern hemisphere and tropics. These effects combined to provide a sufficient density of high ratio DSs from flooded buildings to estimate an accurate average flood height for a local region. In areas of high housing density, the density of high ratio DSs from flooded buildings did fall, probably due to adjacent buildings, but was still sufficient to estimate an accurate local flood height.</p><p>1 Mason et al., <em>JARS</em> 15(3), 032003, (2021).</p><p>2 Pulvirenti et al., <em>IEEE TGRS, </em>54(30), 1532-1544. (2016).</p><p>3 Franceshetti et al., <em>IEEE </em><em>TGRS</em>, 40(8), 1787, 1801. (2002).</p><p> </p><p><strong> </strong></p>

Integration of an improved transformer with physical models for the spatiotemporal simulation of urban flooding depths

Growth and resource allocation of whitetop ( Scolochloa festucacea) along a water depth gradient

A framework for urban pluvial flood resilient spatial planning through blue-green infrastructure

In April 2024, Dubai recorded 118.9 mm of rain in 24 h during a multi-day storm (April 15–17), overwhelming drainage systems across a hyper-arid, densely built landscape. We map urban flood hazard by coupling a 2D HEC-RAS rain-on-grid simulation with a GIS-based MCDA weighted by AHP. Six drivers elevation, slope, aspect, land use/land cover, simulated flood depth, and surface flow velocity are standardized via functional rescaling and fused into a Hydrological Hazard Index (HHI). Inputs are harmonized at 12.5 m (ALOS PALSAR), while Landsat-9 supports LULC mapping and NDWI-based post-event validation (IoU > 0.86; ĸ = 0.81), confirming high model accuracy. Approximately 43 % of Dubai’s urban area falls within the High Hazard class, with localized flood depths reaching 2.5–3.0 m and peak surface velocities exceeding 2.5–3.0 m/s, particularly in low-lying underpasses and constrained flow corridors. High-hazard polygons (Classes 4/5) cluster over flat, impervious districts, offering actionable targets for detention storage, inlet retrofits, tidal backflow control at outfalls, and blue–green corridor integration. The results demonstrate that incorporating dynamic hydraulic parameters is essential for future urban planning and drainage design in Dubai. The combined 2D HEC-RAS and GIS–MCDA–AHP framework provides a reliable, transferable method for flood hazard assessment in hyper-arid cities. Future work should integrate real-time hydrological sensors, non-stationary climate-adjusted rainfall inputs, and socio-economic exposure layers to progress from hazard mapping to full urban flood risk assessment, supporting climate-resilient infrastructure and long-term adaptation planning.

Pneumothorax requires rapid recognition and accurate interpretation of chest X-rays (CXRs), particularly in acute settings where delays can have serious consequences. With the emergence of advanced image interpretation models capable of visual analysis, their diagnostic reliability in radiology practice remains to be determined. This study aimed to assess the diagnostic performance of three state-of-the-art systems in detecting pneumothorax using a large, well-annotated dataset. A total of 10,675 CXRs from the publicly available SIIM-ACR Pneumothorax Segmentation dataset were analyzed. Three multimodal models (GPT-4o, Gemini 2 Pro, and Claude 4 Sonnet) were evaluated using a uniform, image-based approach. Each model's binary outputs (presence: 1, absence: 0) were compared with reference results to determine accuracy, sensitivity, specificity, precision, and F1 scores. Additional subgroup analyses were conducted across pneumothorax size categories: small, medium, and large. Pairwise statistical comparisons were performed using McNemar's test. Sensitivity, specificity, and overall accuracy are reported with corresponding 95% confidence intervals. The prevalence of pneumothorax in the dataset was 22.3% (n = 2,379). All models demonstrated high specificity (above 0.90) but consistently low sensitivity (0.16-0.36). The best overall performance was observed with Gemini 2, which achieved an accuracy of 0.79 and specificity of 0.95, whereas Claude 4 showed greater sensitivity (0.20-0.34) across lesion-size categories. Diagnostic performance improved with increasing pneumothorax size, but smaller lesions remained difficult to identify. Pairwise comparisons confirmed statistically significant differences among all evaluated systems (P < 0.050). In this large-scale evaluation, the tested models exhibited strong reliability in identifying normal examinations but limited ability to detect subtle or small pneumothoraxes. Despite high specificity, low sensitivity limits the use of current Multimodal large language models as rule-out tools for pneumothorax. With continued refinement, these models may eventually support radiologists by improving workflow efficiency and diagnostic confidence. Automated systems capable of high specificity but low sensitivity should not be relied upon to exclude pneumothorax. However, they may serve as valuable assistants for confirming positive findings and prioritizing urgent cases in busy clinical workflows.

The Rapid Mapping (RM) component of the Copernicus Emergency Management Service (CEMS) delivers crisis information within hours or days, during and in the immediate aftermath of a disaster. Users of the service have requested the incorporation of flood depth estimates for inundations and related disasters, such as storms and tsunamis. This demand required the development of a quick and operational approach for estimating floodwater depth distribution, which can effectively adapt to a wide range of flood scenarios, remotely sensed data and Digital Terrain Models (DTMs). INFLOS is a tool developed to answer these requirements, within an expedient 30-minute timeframe at most. It uses sample points collected along the edges of flooded areas derived from remotely sensed imagery, and integrates high resolution DTMs and hydrography data to enhance accuracy. Flood depth is then estimated through a multi-step interpolation and filtering process. INFLOS underwent thorough testing across fourteen regions. As it transitioned into production, INFLOS successfully generated a wealth of flood depth products accessible through the CEMS RM portal: https://emergency.copernicus.eu/mapping/list-of-activations-rapid. Currently, INFLOS is operational and its computation time remains consistently within the order of minutes, making it a staple of the RM tooling and portfolio.