The integration of AI with digital twin improves the analytical and operational capabilities of geospatial systems, which through the analysis of historical data and the integration of real-time information (IoT) are able to highlight even “hidden patterns” in the data, identifying new models capable of improving forecasts with greater control over the quantification of uncertainty and the variability of the phenomenon analysed.
This session aims to focus on flood hazard and risk assessment, monitoring, and management. This Topic invites the submission of articles focused on, but not limited to, the following areas:
• Monitoring of extreme rainfall events and flood hazards for risk assessment and communication.
• Digital twins (DTs)/prototypes of DTs in flood hazard forecasting, early warning, monitoring, and supporting tools for urban governance.
• DSSs to extract meaningful information in the artificial intelligence era, eventually serving to reduce risk and provide support tools to mitigate flood hazards.
• The role of AI and digital twins to assess the economic impacts of flood hazards and the cost-effectiveness of various mitigation strategies.
• Novel techniques to analyse big data coming from Earth observation platforms, drones, and other geospatial data in order to provide timely information related to the extend, exposure, and impacts of flood hazards.
Urban flooding and water pollution have become increasingly severe challenges worldwide as a result of climate change and rapid urbanization, posing substantial risks to public safety, urban infrastructure, and environmental quality (Mark et al., 2004; Andrade et al., 2018). Intense rainfall events frequently exceed the capacity of urban drainage systems, leading to surface inundation and the transport of pollutants into receiving water bodies. To address these issues, numerical hydrological and hydraulic models have been widely applied to simulate urban runoff processes, sewer network performance, and water quality dynamics. Among these models, the Storm Water Management Model (SWMM) is one of the most commonly used tools for analyzing urban drainage systems and pollutant transport under various rainfall scenarios (Gironás et al., 2010). Despite its widespread adoption and robust modeling capabilities, SWMM primarily presents simulation outputs in the form of numerical tables and two-dimensional graphs. This conventional output format limits intuitive interpretation and restricts the ability to analyze spatial and temporal flood dynamics within complex urban environments (Zhang et al., 2016). This study proposes a virtual reality (VR)–based visualization framework that integrates SWMM simulation results with the Unity game engine to enhance the interpretability of urban flooding and water quality simulations. In the proposed framework, rainfall–runoff processes, inundation depth, and pollutant diffusion are first simulated using SWMM for a selected urban catchment. The resulting hydrological and hydraulic outputs are then converted into data formats compatible with the Unity environment. A three-dimensional urban model is constructed to represent surface topography and drainage infrastructure, enabling the visualization of flooding processes in a spatially explicit manner. Flood extent and water depth are visualized dynamically within the virtual environment, allowing users to observe flood propagation over time. In addition, pollutant transport is represented using color-based visualization techniques, where variations in color indicate changes in pollutant concentration. This approach provides an intuitive representation of water quality degradation during flood events. The VR system supports interactive exploration through the use of head-mounted displays and motion interfaces, enabling users to navigate the virtual urban space and examine flooding and pollution patterns from multiple perspectives. The immersive nature of the VR environment enhances spatial perception and facilitates a more comprehensive understanding of complex flood processes compared to traditional two-dimensional visualization methods. By allowing users to directly experience simulated flood scenarios, the proposed framework supports more effective interpretation of model results and improves communication of flood risk information. The results of this study demonstrate that VR-based visualization has significant potential as a decision-support tool for urban flood risk assessment, emergency response planning, and disaster management training.
How to cite: Park, J., Cho, M., Bang, G., Jeung, M., Yun, D., and Baek, S.-S.: Virtual Reality–Based Visualization of Urban Flood Dynamics Using SWMM, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16153, https://doi.org/10.5194/egusphere-egu26-16153, 2026.
The accelerating impacts of climate change and subsequent impact on urban environments such as flooding risks, extreme heat and heavy rain, necessitate rapid and integrated planning strategies. Urban Digital Twins (UDT) have emerged as valuable tools, offering the ability to dynamically model, simulate, and visualize complex processes to support data-driven decision-making. However, a comprehensive strategy that supports the integration of the multitude of UDTs that is being developed specifically into climate adaptation measures, while ensuring interoperability, digital sovereignty and stakeholder participation, is still lacking.
This contribution introduces the collaborative project LINKUDT (“Coordination and Collaboration Platforms for the Synergetic Conception, Development, Interoperability, and Digital Sovereignty of Urban Digital Twins”). Funded by the German Federal Ministry of Research, Technology and Space for a duration of 48 months, LINKUDT serves as the overarching companion research project for six regional real-world laboratories across Germany. The primary objective of the project is to establish UDTs as central instruments for speeding up urban planning processes to improve climate adaptation and sustainable urban development by identifying synergies and supporting interoperability.
A core challenge addressed by LINKUDT is the creation of interoperable and sustainable data infrastructures. Following the FAIR principles (Findable, Accessible, Interoperable, Reusable), the project aims at advancing standards that allow for the efficient integration of heterogeneous data sources, such as sensor networks and environmental models. To prevent vendor lock-in and ensure long-term data portability, LINKUDT emphasizes digital sovereignty through the use and further development of open-source software modules and standards (e.g., OGC API Processes, SensorThings API, CityGML).
Further key outcomes of LINKUDT include training modules for stakeholders /e.g. public administration, developers), and policy recommendations for the nationwide application of digital twin technologies.
By linking the National Research Data Infrastructure for Earth System Sciences (NFDI4Earth) with administrative data infrastructures (GDI-DE), LINKUDT creates a scalable model for evidence-based urban governance.
With our contribution we aim to reach out to further digital twin initiatives related to climate change to initiate further exchange on interoperability, digital sovereignty and emerging technologies.
How to cite: Jirka, S., Radtke, J., and Reiß, J.: LINK Urban Digital Twinning (LINKUDT): Advancing Climate Adaptation and Planning Acceleration through Interoperable Digital Twin Ecosystems, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17814, https://doi.org/10.5194/egusphere-egu26-17814, 2026.
Hydrological forecasts are essential for the timely and accurate prediction of flooding events, which are among the most impactful natural hazards for both infrastructure and human life in Europe and many other regions worldwide. Most existing flood warning systems are supported by hydrological models. Their accuracy depends not only on the representativeness and proper calibration (when required) of the model itself, but also on the quality of its inputs. While static inputs, particularly soil parameters, are highly uncertain, weather forecasts are arguably the most influential drivers.
In this study, we recreate the entire operational modelling framework used in Switzerland. Weather forecasts are provided by ICON (MeteoSwiss) and are used as input for WaSiM (FOEN), which produces streamflow predictions and issues warnings when necessary. We focus on several case studies, including selected catchments (e.g., Thur) and historical events that exceeded national flood warning levels (e.g., 30 May–2 June 2024).
This setup allows us to experiment with different configurations of the numerical weather prediction (NWP) model and to assess their downstream impacts on hydrological forecasts. We test different lead times to evaluate how early flood peaks can be detected, varying ensemble sizes to determine how many members are required to capture “extreme” flooding scenarios, and different spatial resolutions (500m – 2km) to assess the impact of resolving small-scale processes (e.g., convection).
Model performance is evaluated using classical hydrological metrics (NSE, KGE, RMSE, etc.), as well as more operationally relevant metrics for warning systems, such as whether thresholds are exceeded, how early exceedances occur, and their duration. Finally, we test different products for initializing model runs, either interpolated station-based products or NWP analysis products and assess the influence of the hydrological model itself through a sensitivity analysis of its parameters.
The results of this study will shed light on how NWP model configurations affect flood forecasting and, in turn, improve flood early warning design and decision-making.
How to cite: Leonarduzzi, E., Ehlert, K., Leutwyler, D., and Zappa, M.: On the optimisation of numerical weather prediction model configuration for improved flood forecasting, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10725, https://doi.org/10.5194/egusphere-egu26-10725, 2026.
Floods are among the most devastating natural disasters, causing significant loss of life and economic damage. As extreme flood events become more frequent, rapid and accessible flood analysis tools are crucial in guiding early recovery efforts. This study presents the QGIS plugin ‘Floodalyzer’ developed to provide a quick and easy workflow for flood event analysis. By automating the processing and visualization of flood extent data from the Global Flood Monitoring System (GFM), derived from remote sensing, in combination with building footprints from various data sources, the plugin enables users to analyze past flood events without requiring expert knowledge or expensive proprietary software.
Floodalyzer operates within the widely used open-source GIS platform QGIS, making it highly accessible. Users manually download raster data and shapefiles from the web, which serve as inputs for automated analysis. The plugin then processes the data and generates output files, including a shapefile showing which buildings were flooded and for how long. Additionally, it compiles a HTML report including graphs that further describe the area of interest and summarize the plugin’s results (e.g. Building Footprint Heatmap, Observed Flood Extent Raster Calendar Display, Flooded Area Duration Bar Chart). The effectiveness of the tool was evaluated using case studies in Pakistan and Germany, where results were compared against CEMS’s Rapid Mapping Product. The CEMS product was not captured at the time of maximum flooding and therefore shows smaller inundated areas in many places compared to the plugin’s results. However, the locations and overall shapes of the flooded areas are generally consistent.
The case studies highlight the unique selling point of Floodalyzer – it’s ability to process flood extent data over extended time periods to analyze flood duration and damage, which enables a more comprehensive analysis of the available data. At the same time the results highlight uncertainties in flood extent, primarily originating from the GFM input data. Large exclusion mask areas indicate zones of high uncertainty, especially in urban environments where flood detection is more challenging. Temporal uncertainties also arise from gaps in satellite coverage, limiting data availability, especially in regions between the tropics.
Future improvements will focus on reducing runtime, and integrating statistical uncertainty assessments in the plugin’s output with human-readable explanations. Further, automated GFM data retrieval from the Global Flood Awareness System automating the download of the flood masks given an input AOI, would eliminate the need for manual downloads and thereby streamline the analysis process. By bridging the gap between high, complex data amounts and the need for a rapid response to flooding events, this tool provides decision-makers with a sound basis for dealing with the impacts of flooding in the response and recovery phase. Floodalyzer thus supports improved flood management through broader uptake of remotely sensed flood information, by lowering barriers to accessibility for flood extent data.
How to cite: Fuest, L. and Dasgupta, A.: Floodalyzer: A QGIS Plugin for Accessible and Rapid Flood Event Assessment, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1139, https://doi.org/10.5194/egusphere-egu26-1139, 2026.
LiDAR-derived digital elevation models (DEMs) are increasingly adopted in hydrodynamic flood modelling; however, their direct use, particularly in complex urban environments, remains problematic. Although LiDAR provides high-resolution surface information and supports the generation of bare-earth digital terrain models (DTMs), unresolved flow-permeable structures such as bridges, culverts, and elevated transport infrastructure, together with micro-scale urban features including narrow river channels, pathways, kerbs, and missing submerged channel bathymetry, systematically distort flow connectivity and channel conveyance. These deficiencies introduce structural biases into flood simulations, yet existing studies typically address individual features in isolation, limiting transferability and large-scale applicability.
This study reframes LiDAR DEM preprocessing as a process-based investigation into how unresolved terrain features bias flood hydraulics and introduces an automated, physically consistent terrain reconstruction framework that explicitly targets these bias mechanisms. The framework is implemented at the national scale using the 2 m LiDAR-derived DTM for England.
Three dominant sources of hydrodynamic bias are addressed. First, flow-permeable structures, including bridges, culverts, and elevated transport infrastructure, are systematically identified using observed water surface information and river network data, and the terrain beneath these structures is reconstructed using interpolation-based techniques to restore hydraulic connectivity. Second, impermeable urban features, such as buildings and kerbs, are selectively elevated while preserving longitudinal connectivity along roads and pathways, ensuring realistic overland flow routing. Third, submerged river bathymetry is reconstructed using empirical relationships between river width and water depth to recover channel conveyance absent from bare-earth DTMs.
The resulting terrain dataset is directly applicable to hydrodynamic flood modelling without manual intervention. Sensitivity analyses across multiple historical flood events demonstrate that restoring flow connectivity and reconstructing channel bathymetry exert distinct and flow-regime-dependent controls on simulated flood extent, water levels, and discharge. In particular, unresolved flow-permeable structures predominantly govern urban inundation patterns, whereas missing bathymetry represents the primary source of error in channel hydraulics.
By systematically isolating and correcting key terrain-induced bias mechanisms, this study provides generalisable insights into the process sensitivity of catchment and urban flood models to DEM representation and offers a scalable pathway for improving large-scale flood simulations using LiDAR data.
How to cite: Chen, H., Tong, X., and Liang, Q.: Reconstructing Flow Connectivity and Channel Conveyance in LiDAR-Derived Terrain for National-Scale High-Resolution Flood Modelling, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-22261, https://doi.org/10.5194/egusphere-egu26-22261, 2026.
Non-contact river monitoring is essential for understanding hydraulic phenomena and providing real-time disaster mitigation information during large-scale floods. Our previous research (Yorozuya et al., 2026) developed a method to inversely estimate riverbed elevation by integrating UAV-derived surface velocity (via PIV) and water surface geometry (via LiDAR) into a Physics-Informed Neural Networks (PINNs) framework using automatic differentiation of the governing equations. However, that approach relied on a uniform velocity correction factor across the entire reach, which led to significant underestimations of water depth in complex flow fields, such as those near spur dikes.
In this study, we propose an enhanced estimation algorithm that incorporates secondary flow effects into the momentum equations to improve bathymetric accuracy. Following the methodology of Iwasaki et al. (2013), we identify regions where surface velocity vectors exhibit curvature and account for the resultant increase in flow resistance. This approach aims to correctly identify water depth even in regions where surface velocities are low but hydraulic complexity is high.
Field experiments were conducted in a reach of the Kurobe River (bed slope ≈1/100, 20m wide by 50m long), characterized by a spur dike in the center of the domain. High-resolution water surface geometry and velocity fields were captured using a UAV-mounted LiDAR (DJI Zenmuse L2) and a photogrammetric camera (P1). These data were integrated into the PINNs loss functions, which were defined based on the continuity equation, the shallow water equations, and the conservation of discharge across cross-sections.
The results demonstrated a marked improvement in estimation reliability, particularly in the separation zones downstream of the spur dike. Without secondary flow considerations, the model estimated near-zero water depth in large wake vortices due to the low surface velocities. By incorporating secondary flow effects, the model correctly evaluated the increased apparent roughness due to flow curvature, yielding deeper and more accurate bathymetry consistent with ground-truth data obtained by boat-mounted ADCP. This study highlights the potential of using only UAV-based remote sensing to achieve high-precision bathymetric inversion in morphologically complex river environments.
Iwasaki, T., Shimizu, Y., and Kimura, I. (2013). An influence of modeling of secondary flows to simulation of free bars in rivers. Journal of Japan Society of Civil Engineers, Ser. B1 (Hydraulic Engineering), Vol. 69, No. 3, 147–163.
Yorozuya et al. (2016) Seeing the unseen, RiverFlow2026 (Under review)
How to cite: Yorozuya, A., Inaba, R., and Kudo, S.: Bathymetry Estimation in Complex River Morphology using UAV-based Remote Sensing and Physics-Informed Neural Networks Incorporating Secondary Flow Effects, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17859, https://doi.org/10.5194/egusphere-egu26-17859, 2026.
Flood risk management in urban floodplains strongly depends on the spatial resolution of digital elevation models (DEMs), which control floodplain connectivity, flow pathways, and surface storage. In many developing countries, flood-related studies rely predominantly on publicly available global DEM products, whose spatial resolution and vertical accuracy are often insufficient to represent subtle topographic gradients, densely vegetated floodplains, and complex urban microtopography. These limitations are particularly critical in low-relief environments, where small elevation differences exert a disproportionate control on inundation extent and flood dynamics. This issue has become increasingly evident in subtropical lowland regions of southern Brazil, where extreme flood events in 2023–2024 exposed shortcomings of commonly used global DEMs for urban floodplain applications. Therefore, the Piratini River watershed has been the focus of ongoing efforts to develop a real-time hydrological forecasting system to support decision-making during flood emergencies under data-scarce conditions. The urban areas of Pedro Osório and Cerrito along the main floodplain of the Piratini River constitute the core operational domain of this system and are recurrently affected by flooding. The watershed drains approximately 4,700 km² upstream of the municipalities and is characterized by low relief and wide floodplains. This study investigates the applicability of publicly available global DEMs and locally derived high-resolution elevation datasets for floodplain mapping and hydrological–hydrodynamic applications in these urban areas. A comparative assessment was conducted using two global DEM products - ALOS PALSAR (12.5 m) and ANADEM (30 m) - and three locally derived DEMs generated from high-resolution surveys. Local datasets include two Global Navigation Satellite System (GNSS) Real-Time Kinematic (RTK)–based surveys (static and kinematic) acquired with an Emlid Reach RS2+ receiver using real-time corrections via NTRIP (Networked Transport of RTCM via Internet Protocol), and an unmanned aerial vehicle (UAV)–based Light Detection and Ranging (LiDAR) survey acquired with a DJI Matrice 350 RTK platform equipped with a Zenmuse L2 sensor. The static GNSS survey comprised 2,921 points, while the kinematic survey yielded approximately 34,000 at a 1-s sampling interval. The UAV–LiDAR survey covered 21.5 km² of the urban floodplain. Raw elevation data from local surveys were converted from ellipsoidal to orthometric altitude using the hgeoHNOR2020 geoid model. GNSS-derived altitudes were interpolated using ordinary kriging in ArcGIS Pro. LiDAR data were processed in DJI Terra, resulting in a high-density point cloud (> 98 points m⁻²) and a terrain model with decimetric spatial resolution. Results reveal clear differences among datasets. Global DEMs show limited capability to represent floodplain connectivity and microtopography, particularly in vegetated areas. GNSS RTK–based DEMs provide intermediate performance but are constrained by survey logistics and GNSS signal degradation. In contrast, the UAV-based LiDAR DEM provides the most detailed and hydrologically meaningful representation of floodplain morphology, including vegetated and off-street areas, enabling improved delineation of flow paths and floodplain storage. These findings highlight the critical role of high-resolution elevation data for floodplain mapping and hydrological–hydrodynamic analyses in low-relief urban environments, reinforcing UAV-based LiDAR as a key remote sensing tool for risk assessment and climate adaptation in data-scarce regions.
How to cite: Luceiro Santana, E., Martins Bueno, L., Souza da Paz, G., De Oliveira Alves, R., Leitzke Caldeira, T., Beskow, S., Araujo Rodrigues, A., Angelo Borges, J. C., Leal Teixeira, D., Adolfo Karow Weber, G., and Leandro, D.: Remote sensing–based urban floodplain mapping: the added value of UAV-LiDAR compared to global and GNSS-derived DEMs, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15447, https://doi.org/10.5194/egusphere-egu26-15447, 2026.
Earthquakes can cause rapid changes in elevation and topographic relief, which, in turn, affect hydrologic regimes and modify flood risk in affected regions. The regulatory floodplain, an area of elevated flood hazard adjacent to water bodies, is critical for managing exposure and mitigating flood risk in nations. Shifts in the distribution of flood risk in regions impacted by seismic activity constitute a compound hazard. Tools are needed to enable reevaluation of regulatory flood maps after seismic events, minimizing exposure of affected populations to additional flood risk. In the United States, floodplain mapping is primarily implemented by the Federal Emergency Management Agency (FEMA), known as regulatory flood mapping. They are based on Hydraulic modeling and delineate the floodplain for areas representing a 1% annual chance of flooding. The floodplain map is not updated regularly by FEMA; it relies on manual, costly revision processes and does not consistently use current, high-resolution, and up-to-date elevation data. Therefore, these maps will struggle to detect recent flood behavior, thereby increasing flood risks and limiting the effectiveness of regulatory flood mapping management. This study presents a rapid, satelliteintegrated framework for updating regulatory flood maps in regions exposed to topographic shifts from earthquakes. Using the 2019 Ridgecrest earthquake sequence as a case study in the North and South Fork Kern River basin, California. Specifically, we used the U.S Geological Survey 3DEP/NED with 10-m resolution DEM, which represented the pre-earthquake topography, integrated with a vertical displacement data derived from InSAR time series analysis to generate a corrected post-earthquake DEM. Both DEMs were then used in the HEC-RAS model to quantify changes in floodplain extent and inundation patterns under multiple return-period scenarios. To assess model performance and quantify the accuracy improvements in regulatory flood mapping, observed flood inundation maps derived from high-resolution PlanetScope satellite imagery were used in the validation. Our integrated approach demonstrates how InSAR-updated topography improves floodplain mapping accuracy and enables rapid updates to regulatory flood maps. HEC-RAS modeling results across three reaches along the North and South Fork Kern River consistently showed larger flood extents in post-earthquake simulations relative to pre-earthquake conditions. Validation using PlanetScope-derived flood inundation maps demonstrates improved model performance for the post-earthquake DEM, with an F-score 84.52% compared to pre-earthquake simulations, using an optimal NDWI threshold of 0.35.
How to cite: Al-Amry, N. and Carter, E.: Assessing Fluvial Flood Risk Changes Using an Updated Digital Elevation Model Post-Earthquake: A Case Study, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15883, https://doi.org/10.5194/egusphere-egu26-15883, 2026.
Extreme rainfall events have become more frequent and intense under climate change, presenting increasing challenges for hydrological monitoring and flood risk management. High-resolution rainfall observations are essential for capturing the spatial and temporal variability of storm events, yet conventional rain-gauge networks suffer from limited spatial coverage and cannot resolve rapidly evolving convective structures. Moreover, high-intensity rainfall events are inherently rare in natural settings, resulting in data gaps in upper rainfall categories. To address this limitation, we integrate natural rainfall observations with controlled artificial rainfall experiments to construct a comprehensive and balanced multi-class dataset covering 0–70 mm/hr at 5 mm/hr intervals. We develop a multimodal deep learning framework that jointly leverages rainfall imagery and acoustic measurements for rainfall-intensity estimation. The two sensing modalities provide complementary physical information: imagery captures streak morphology, drop density, and spatial distribution patterns, while acoustics encode drop momentum, kinetic energy, and impact signatures. Neither modality alone fully characterizes rainfall processes across all intensity ranges; by combining them, the model benefits from richer and more discriminative features. Two-second audio segments are converted into log-mel spectrograms, and a Cross-Attention fusion mechanism enables the network to selectively emphasize the most informative cues from each modality for different rainfall categories. Image-based data augmentation such as horizontal flipping further expands the training space and improves model generalization.
Compared with previous studies that relied on single-modality inputs or coarse categorical schemes, our framework achieves a substantially finer classification resolution (0–70 mm/hr in 5-mm/hr bins) and exhibits improved discrimination between adjacent intensity levels. The multimodal architecture consistently outperforms single-modality baselines, with the performance gains being particularly notable in the moderate-to-heavy rainfall range, where the model achieves higher classification accuracy, highlighting the benefits of true cross-modal complementarity. The integration of artificial and natural rainfall further produces a balanced and physically representative dataset that captures both controlled high-intensity scenarios and real-world variability.Overall, this study demonstrates the potential of multimodal sensing and deep learning to advance rainfall monitoring capabilities. The proposed non-contact, low-cost, and high-resolution approach offers a promising pathway for enhancing rainfall observation in regions with sparse gauge coverage, strengthening flood early warning systems, and supporting real-time hydrological applications under a changing climate.
How to cite: Lin, C.-C. and Ho, H.-C.: Cross-Attention Multimodal Learning Using Image and Audio for Rainfall Intensity Estimation, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2660, https://doi.org/10.5194/egusphere-egu26-2660, 2026.
Irrigation water management is a critical factor that influences crop biomass, yield, and water usage, since irrigation makes the crop development independent of rainfall. Poor irrigation management can result in many problems on the farm and off the farm, such as waterlogging, erosion, and non-point source pollution. Therefore, improving irrigation water-use-efficiency is essential to reduce the amount of water needed without penalizing the yields. Considering the growing competition for water resources, there is a need to explore novel methods for quantifying and enhancing water use efficiency in irrigated fields, such as Unmanned Aerial Vehicle (UAV)-based remote sensing. This study integrates UAV-derived vegetation indices with machine-learning (ML) algorithms to quantify biomass and yield response of rice under alternate wetting and drying (AWD) and wheat under different irrigation methods (drip, sprinkler, and flood) with variable rates of crop evapotranspiration (100%, 75%, 50% and 0% rainfed treatment) across two seasons of the rice-wheat cropping system in Roorkee, India. The biomass and yield results obtained from the different ML algorithms were compared. During the training process of the ensemble random forest model, it performed better with a higher KGE (0.91) and a lower value of NRMSE (0.033), and a minimal PBIAS of 0.13%. The ensemble random forest model performed better during the testing process of the rice yield estimation (R2 = 0.60, KGE = 0.71, PBIAS = −2.26%, NRMSE = 0.136). For wheat yield estimation, training results were similar with strong model performance (R2 = 0.8137, KGE = 0.83, PBIAS = 1.36%, NRMSE = 0.470). The UAV-ML workflow captured both the fine-scale spatial variability needed for site-specific field decisions and the process understanding needed for generalization across the seasons. This integrated workflow supports the UN Sustainable Development Goals (SDGs), specifically SDG 2 (Zero Hunger) and SDG 6 (Clean Water and Sanitation).
How to cite: Kumar Vishwakarma, S., Kothari, K., and Pandey, A.: Spatial Mapping of Biomass and Yield of Rice-Wheat Cropping Systems across Different Irrigation Methods Using UAV Images and Machine Learning Algorithms , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-452, https://doi.org/10.5194/egusphere-egu26-452, 2026.
