The scientific community needs to move beyond natural hazard forecasting and towards impact- and action-based forecasting. This, in turn, requires commitment to the creation and dissemination of multi-hazard risk and multi-source impact data (including from social media) as well as the collaborative production of impact-based forecasting services and linked early action protocols.
However, much remains unknown and significant knowledge gaps persist. This session aims to offer valuable insights and share best practices on impact-based early warning systems from the perspective of both the knowledge producers and users. Such systems demand much knowledge about how hazards translate to impacts through exposure and vulnerability, novel impact-based forecasting technologies (including machine learning models), the costs and benefits of triggered actions, human decision-making and risk perception dynamics.
Topics of interest include, but are not limited to:
- Practical applications and operational use-cases of impact-based forecasts
- Novel physics-based, Artificial Intelligence (AI) and hybrid models for impact-based forecasting
- Innovative solutions to address challenges in impact-based forecasting effectively, including the application of AI, harnessing big data and earth observations
- Development of cost-efficient, evidence-based early action portfolios
- Impact and action-oriented forecast verification and post-processing techniques
- Triangulation of indigenous and scientific knowledge for leveraging forecasts, multi-hazard risk information and climate services to last-mile communities
- Bridging the gaps in risk and impact data to support impact-based forecasting
- Collecting and expanding datasets on interventions and adaptations to build an early action evidence base
Early warning systems (EWS) are widely recognized as one of the most effective tools for protecting lives and livelihoods from natural hazards. The Early Warnings for All initiative, launched by the United Nations Secretary-General in 2022, aims to ensure universal protection from hazardous hydrometeorological, climatological, and related environmental events through life-saving, multi-hazard early warning systems, anticipatory action, and strengthened resilience by 2027. However, despite substantial advances in forecasting capabilities over recent decades, the practical implementation of effective and actionable EWS remains challenging, with pronounced regional disparities, particularly in developing and fragile contexts.
This talk presents real-world experiences from the implementation of impact-based early warning systems in developing countries. It highlights key operational challenges across the early warning–early action chain, including gaps in risk and impact data, institutional coordination constraints, and difficulties in translating forecasts into timely and trusted decisions. The contribution also discusses opportunities offered by innovative approaches, such as the collaborative co-production of early warnings in transboundary river basins, impact-based forecasting frameworks, AI-supported forecasts, and the integration of local knowledge in operational EWS.
How to cite: Rossi, L., Mapelli, A., Libertino, A., Gabellani, S., Alfieri, L., Testa, N., Poletti, L., Panizza, E., Fiorucci, P., Trucchia, A., Perello, N., Meschi, G., D'Andrea, M., Cremonese, E., Isabellon, M., Trotter, L., and Masoero, A.: Early Warning Systems in the Global South: challenges and innovative approaches, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8039, https://doi.org/10.5194/egusphere-egu26-8039, 2026.
In Switzerland, surface water floods (SWF) account for approximately 23% of the financial losses to property caused by floods. Improving the understanding of these events is therefore essential to enhance prevention and risk mitigation efforts. However, SWF impacts are challenging to forecast, as they result from the interaction of multiple processes and are strongly influenced by local conditions, building exposure, and vulnerability.
We develop a data-driven model to predict potential damages, trained on damage data provided by the Swiss Mobiliar Insurance Company and the Building Insurance of the Canton of Zurich (GVZ). The objective is to predict the probability of damage to buildings caused by SWFs using gridded hourly precipitation data and morphological properties.
We compare several approaches, including a simple threshold-based method, logistic regression, random forests, and deep learning models such as Convolutional Neural Networks (CNNs) and Transformers. The relevance of spatio-temporal patterns in precipitation fields is assessed using 1-D, 2-D, and 3-D CNNs. Variants of Transformer architectures are also evaluated.
How to cite: Horton, P., Mosimann, M., Kaderli, S., Zischg, A. P., and Martius, O.: Impact-based prediction of building damage from surface water floods using machine learning., EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7705, https://doi.org/10.5194/egusphere-egu26-7705, 2026.
As weather-related disasters become more frequent and severe, there is a growing global push toward impact-based early warning systems, exemplified by initiatives such as EW4All. This transition positions machine learning (ML) and artificial intelligence (AI) as powerful tools for integrating meteorological hazard data with information on vulnerability and exposure into data-driven forecasting systems. In this work, we explore the use of 112 emergency calls as high-resolution impact proxies for an ML-based prediction problem. Specifically, we develop a model that combines rainfall-related weather data and static vulnerability-exposure layers to predict, at a municipal and hourly resolution, whether flood-related impacts will occur in the next hour. This study spans a period of over six years (October 2018 to February 2025) in Catalonia, northeastern Spain.
To address the severe temporal class imbalance and uncertainty characteristics of emergency calls data, we define a custom walk-forward evaluation scheme that ensures the same number of positive samples across comparable time periods. We then distribute municipalities into three distinct population density groups (low, medium, and high) and train one model for each one. This stratification enables us to evaluate performance across diverse population dynamics and varying data availability. The resulting models are compared against operational methodologies, such as climatology-based weather warnings issued by meteorological agencies. Our results show that the ML approach represents a substantial improvement in two of the three groups. The model for the lowest-density group, however, struggles due to a substantial lack of impact data, highlighting a key roadblock for data-driven algorithm development in sparsely populated regions.
To gain a more complete understanding and improve model trust and explainability, we perform a series of experiments: a feature importance analysis using SHAP (SHapley Additive exPlanations), ablation studies over different feature groups, and training models on individual feature sets. From these results, we can ascertain how the combination of varied data sources (such as weather radar, station sensors, or call history) can result in more powerful predictions than using single sources in isolation.
Finally, we present a methodology for characterizing the different stages of a rainfall event, as performance is expected to vary throughout its evolution. We distinguish five stages based on observed rain in the previous and following hours: The first hour with rain, intermediate hours, the last hour with rain, the hours immediately after the event, and hours without rain. Evaluating all approaches following this framework adds a valuable dimension to the performance analysis and further improves explainability. The results demonstrate that our models outperform the baselines across all event stages, from the initial onset of rain to the hours after precipitation has stopped. This highlights the strong potential of even relatively simple ML pipelines to deliver timely, localized anticipation of weather-related impacts.
How to cite: Morales Casas, J., Lapedriza, A., Kaltenbrunner, A., and Llort, X.: Using ML for the prediction of flood-related emergency calls, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9458, https://doi.org/10.5194/egusphere-egu26-9458, 2026.
Floods are increasing in frequency and severity. Flood forecasting is ever improving and is already of high quality at national to regional level. However, there are fundamental limits in flood forecasting, especially at sub-regional to building level, making observations indispensable. Unfortunately, observations are often hampered by limitations in frequency, accuracy and the covered area. It is crucial to bridge the gap between modeling and observations to obtain situational awareness and to guide rescue forces and further data acquisitions. One possible approach is the use of focus maps, which combine multiple proxy layers into one common proxy of risk. These have been successfully applied to identify hotspots of areas affected by earthquakes or floods. This work uses the concept of focus maps and applies it to Ahr valley and Vesdre valley, two of the main affected areas of the European floods in July 2021. The work presents a thorough survey of static and observational proxy layers, such as flood hazard maps, satellite derived flood maps and Facebook user activity data, with various coverage (global, European, national). It tests how well individual layers and their combinations approximate the areas affected by the floods and finds that already few data layers suffice to obtain a strong approximation. Furthermore, it shows that Facebook user activity data provides a valuable source to identify the onset time of the flood event and to identify the affected regions. However, the user activity data is too coarse and noisy to obtain accurate predictions. By combining the dynamic data with readily available static proxy layers of higher spatial resolution a risk proxy is obtained, which could potentially scale to other areas of interest.
How to cite: Skuppin, N., Gottschling, N. M., Dujardin, S., Camero, A., Martinis, S., Palmaerts, B., and Taubenböck, H.: Flood hotspot mapping using static and dynamic data: A case study of the European Floods in 2021, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17171, https://doi.org/10.5194/egusphere-egu26-17171, 2026.
Impact-based forecasting uses hydrometeorological information to trigger timely early actions, but real-world events show a gap between early warning and action. This paper addresses this bottleneck by examining how people interpret uncertainty in warnings and the impact this has on their actions.
We introduce the Uncertainty Lens Framework (ULF), which analyses how perceived uncertainty shapes threat, ownership, and coping appraisals in flood risk contexts. The ULF combines Protection Motivation Theory, decision heuristics and the Safe Development Paradox to explain why uncertainty can trigger protective action in some settings but lead to delay, denial or delegation in others. In this study, we apply the ULF to the 2021 flood in Germany, using quotations from newspapers and open-ended survey responses that capture the reasoning of affected residents during the event.
Three 'illusions of safety' that suppress early action emerge: (1) experience-based normalisation ('we've seen floods before'), (2) responsibility delegation ('someone else will handle this'), and (3) overconfidence in systems and protection ('the infrastructure/authorities will protect us'). These illusions are reinforced when uncertainty is implicit, inconsistently acknowledged or communicated without stable anchors to help people contextualise unprecedented escalation.
We therefore advocate proactive uncertainty management also for impact-oriented services and warning systems. Rather than trying to eliminate uncertainty, services should incorporate it into risk communication and policy design by deliberately establishing anchors and availabilities that help people understand residual risk from immediate and potential future exacerbation. Crucially, uncertainty communication must be embedded in sustained community-level engagement and long-term risk awareness so that warnings issued during an event are interpreted in the context of shared mental models, established trust relationships and preparedness measures.
How to cite: Höllermann, B. and Heidenreich, A.: Why Accurate Flood Warnings Still Fail: Behavioural Mechanisms of Uncertainty Interpretation and Implications for Impact-Oriented Services, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17685, https://doi.org/10.5194/egusphere-egu26-17685, 2026.
Flood Forecasting and Early Warning Systems (FFEWS) are key to reduce flood impacts by providing timely information to individuals, communities, and authorities. However, during the July 2021 floods in Europe, major gaps were observed between forecasts, warnings, and protective actions. In the impacted region of Limburg in the Netherlands, only 55% of people in flood-prone areas received an evacuation warning, and just 41% took emergency measures. This highlights a critical weakness in the FFEWS chain: the translation of forecasts into actionable warnings that effectively trigger response. Impact-based forecasting (IbF) has been promoted as an important step in bridging this gap, shifting the focus from hazard forecasting to forecasting societal consequences of potential flooding. Despite increasing interest in IbF, most FFEWS still focus mainly on hazards and are not tailored to forecast users and the specific actions they can trigger. Moreover, FFEWS effectiveness is often only assessed by the skill of flood hazard warnings, while there is little research on whether warnings lead to effective responses. To address this issue, we developed an impact-based flood forecasting, early warning, and response system (IbF-FEWS) using the Geographical, Environmental, and Behavioral (GEB) platform. This system consists of three novel interconnected components: (i) a flood forecast module, in which probablistic ensemble rainfall forecasts force a combined hydrological-hydrodynamic model to generate ensemble forecasted flood maps; (ii) a warning module, in which these flood maps are transformed into lead-time–dependent flood probability maps and evaluated against two action-based hazard thresholds: damaging water-level ranges and exposure of critical infrastructure. Each threshold is associated with recommended emergency measures (e.g. placing sandbags). Then, for each postal code, flood probabilities are filtered using a predefined probability threshold to identify flooded areas, after which the fraction of affected buildings or flooded area within the postal code area is evaluated to determine whether a warning is issued; and (iii) a decision-making module, in which households decide whether to implement the recommended measures based on their responsiveness to warnings, modeled as a binary state classifying households as either responsive or non-responsive. We demonstrate the system for the July 2021 flood event in the Geul catchment in the South of the Netherlands, showing how probabilistic, impact-based, and action-oriented warnings can lead to earlier and more effective early action. The results demonstrate the potential reduction in flood damage had such a system been operational during the 2021 event.
How to cite: Oliveira, R., Busker, T., de Bruijn, J., de Moel, H., Pontman, R., Botzen, W., and Aerts, J.: From forecasts to action: testing a new impact-based flood early warning system, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-13437, https://doi.org/10.5194/egusphere-egu26-13437, 2026.
Accurate drought impact forecasting is fundamental for effective decision-making, yet forecasting drought impacts rather than hazards remain difficult due to the complex, non-linear relationship in which they can materialise. Impact-based drought forecasting at seasonal timescales is particularly challenging, thereby benefitting from methods that ensure reliability and transparency. Here, we present a novel machine-learning (ML) framework for drought impact-based forecasting that explicitly evaluates model performance, actionability, and explainability.
The study is structured as a sequence of experiments. First, an autoencoder and several ML models—Gradient Boosting (XGB), Random Forest (RF), and Support Vector Machine (SVM) are trained with observed drought hazard indicators at multiple aggregations (e.g. SPI-12, SPI-24, SPEI-1, SPEI-3, SPEI-6, FAPAR-1, FAPAR-3, SMA-1, SMA-3) up to the current date to understand the data and how ML models perform to predict water scarcity levels in Italy, chosen as the drought impact indicator. The U-Net and ConvLSTM models were chosen as baseline models, as they directly predict gridded water scarcity levels. The framework is then extended by incorporating seasonal climate forecasts (precipitation and temperature) up to six months ahead to enable real-time impact prediction. Model sensitivity to spatial resolution is evaluated by testing inputs at 1 km and 25 km scales. To ensure that results are physically meaningful, explainable AI (xAI) techniques are applied to quantify predictor importance using SHAP, identify spatial hotspots using Integrated Gradients, and determine the most informative periods of the year using Partial Dependence Plots.
Results show clear performance differences among models. Tree-based approaches, particularly Gradient Boosting and Random Forest, consistently outperformed deep learning baselines at both spatial resolutions. At 1 km resolution, xAI identifies SPEI-6 and SMA-1 as the most influential predictors, while at 25 km resolution SPEI-6 and FAPAR-3 emerge as the dominant drivers. Model performance improves at coarser resolution, with tree-based models providing the most accurate and robust predictions. Overall, the study (i) presents a workflow for assessing the effectiveness of ML in enhancing the seasonal prediction of drought impacts, (ii) leverages xAI to evaluate the relationship between the drought hazard indicators and drought impact data, including the most informative periods of the year and the spatial hotspots; and (iii) enabling real-time drought impact-based forecasting at seasonal scale.
How to cite: Azas, K., Cremonese, E., Rossi, L., Hrast Essenfelder, A., Trotter, L., Provenzale, A., and Ficchì, A.: Towards the Operational Implementation of Seasonal Drought Impact-based Forecasting with Explainable Machine Learning, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8138, https://doi.org/10.5194/egusphere-egu26-8138, 2026.
Impact-based early warning (IbEW) aims at integrating knowledge about risks and impacts with timely, understandable and actionable warnings, thus enabling targeted early actions that can help reduce risks in the face of impending hazards. However, applications are still scarce, and an established risk-informed framework to guide assessment and inform early actions has yet to emerge. Drawing on the outcomes of a research project in Eastern Africa, with pilot studies in Kenya and Ethiopia, we have developed an IbEW application for drought and flood risks, spanning from conceptualization to co-development and implementation, informed by a novel IbEW framework. Drought risks are of particular significance in the region, with recent events exacting disruptive tolls on the lives and livelihoods of millions of people. To capture their characteristics and warn for these impacts, we have developed a drought IbEW methodology for rainfed agriculture (informed by co-developed conceptual risk models) that combines spatial hazard information (using the combined drought indicator - CDI), dynamic exposure of cropland (by accounting for crop-specific calendar variability and phenological stages), and contextual warning information on multiple dimensions of vulnerability of rainfed farming households and specific vulnerable groups (women and girls, persons with disabilities, and people in camps setting). Focusing on three staple crops (maize, wheat, sorghum), our application produces automated assessments of multiple combinations of drought hazard, crop types, phenological stages, and possible impacts on crop production quantity at both dekadal and monthly accumulation periods, packaging contextualized warning messages in an intuitive narrative format. Our system was co-developed with and validated by national and subnational experts and stakeholders through multiple stages, and aims to deliver actionable information to people at risk and to organizations and institutions responsible for disaster response and risk management.
How to cite: Cotti, D., Pfeiffer, S., Dewi, M., Kiptum, A., Musa, J., Okoth, V., Lelaono, M., Limo, E., Ouma, J., Nyaga, J., Mwangi, P., Rono, F., Alfieri, L., Trasforini, E., Amdihun, A., Massabo, M., Werners, S., and Hagenlocher, M.: From risk knowledge to effective early actions: a novel framework and application for impact-based early warning with a pilot study in Eastern Africa, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15034, https://doi.org/10.5194/egusphere-egu26-15034, 2026.
Continuous Risk Monitoring and Assessment (CRMA) is widely used in financial auditing and cyber-risk management to update risks in real-time and escalate them as conditions evolve. Hydrometeorological early warning systems typically operate in a cycle of repeated hazard and threshold monitoring, usually daily for floods and monthly or seasonally for droughts. The current study introduces a method tailored for operational Impact-Based Forecasting (IBF) for flood and drought hazards in East Africa, developed under the Complex Risk Analytics Fund(CRAF'd) project. The method formalizes existing monitoring practices into a continuous, conditional, evidence-driven hydrometeorological risk assessment process, in which evolving observations, forecasts, and expert knowledge are systematically integrated, documented, and auditable across time.
The method combines forecast and observation indicators using probabilistic Bayesian networks to aggregate risks and provide decision support. For drought, it uses multi month Combined Drought Indicators (CDI) as observed antecedent conditions, along with ECMWF SEAS5 standard precipaiton index (SPI) ensemble forecasts across agricultural seasons. For floods, antecedent rainfall and soil saturation indicators from satellite observations are fused with short-range ensemble precipitation forecasts from NOAA GEFS. In both hazard contexts, Bayesian Networks encode expert knowledge through Conditional Probability Tables(CPT) to represent compound risk mechanisms, temporal persistence, spatial coverage, and data confidence, enabling transparent, uncertainty quantification and reproducible inference of evolving risk states.
The output produces admin-2–level traffic-light risk communcation categories linked to anticipatory action decision pathways. Validation results from pilot study demonstrate that Bayesian Networks implemented using the Python pgmpy library enable cost-effective and repeatable continuous risk monitoring when combined with analysis-ready, cloud-optimized datasets. The results show that parsimonious hazard modelling, using Prefect automation tool for operational impact-based forecasting, a calendar-based web app, and structured CPT management support transparent risk assessment, traceable record-keeping, and auditable decision histories. Integration with storymaps complements this method by enabling event-based climate storylines that link risk knowledge with operational decision communication.
How to cite: Kalladath, N., Tucci, R., Koros, H., Zablone, O., Mahzabeen, A., Gudoshava, M., and Amdihun, A.: Continuous Risk Monitoring and Assessment (CRMA) for Operational Impact-Based Forecasting: A Bayesian Network method for Flood and Drought Hazards in East Africa , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18323, https://doi.org/10.5194/egusphere-egu26-18323, 2026.
Despite major advances in the accuracy and lead time of hydrometeorological forecasting, significant gaps persist in the early warning-early action chain, limiting the ability of warnings to trigger timely and effective protective actions at the municipal level. Impact-based early warning systems have emerged as a promising pathway to address these gaps; however, their operational implementation, particularly the systematic availability, integration and usability of impact data for warning validation and improvement, remains a key challenge.
In this contribution, we present an impact database environment designed to collect, structure and support the analysis of observed impacts from hydrometeorological events. The database links reported impacts to forecasts, warning levels and predefined response actions, and is dynamically populated through a mobile application that enables users to submit geolocated impact reports, including text descriptions, images and links to official information sources. A central component of the database is its connection to in situ sensors, forecasts and warning thresholds, enabling comparisons of observed impacts with forecasted conditions and triggered warning levels to support warning validation and refinement. In parallel, artificial intelligence techniques are being integrated to support the organisation and filtering of incoming impact reports, and to explore the extraction of event-based impact information, with the aim of informing future impact-based warning threshold assessment.
This impact database ecosystem is embedded within the Site-Specific Early Warning System (SS-EWS) architecture, an operational framework for designing and implementing impact-based warnings at vulnerable locations to trigger self-protection actions. The SS-EWS, including the database prototype, is currently being implemented, improved and evaluated in close collaboration with civil protection and emergency authorities across vulnerable municipalities in Europe within the Horizon Europe GOBEYOND project, and in Catalonia (Spain) through the SAAI project, providing a broad co-design and real-world evaluation environment.
How to cite: Meléndez-Landaverde, E., Sempere-Torres, D., González, V., and Sanz-García, R.: From impact data to impact-based warnings: developing an impact-centred database to support local warning validation, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20343, https://doi.org/10.5194/egusphere-egu26-20343, 2026.
This work presents the methodological framework of the HIDROANDES project, involving the participatory installation of rainfall and streamflow monitoring stations in indigenous and rural communities of the Mulato River sub-basin (Mocoa, Colombia). Precipitation and water level measurements constitute the foundation for the development of an integrated early warning system aimed at reducing vulnerability to rapid-onset flooding events.
The proposed methodology consists of three interconnected components. First, real-time community-based monitoring, in which local actors operate hydrometeorological stations, generating geo-referenced datasets while integrating traditional knowledge and ensuring inclusive participation. Second, AI-assisted hydrological modelling, based on neural networks trained with locally generated and synthetic data to capture the specific hydrological response dynamics of the basin. Third, a generation of tailored alerts, designed according to the socio-territorial characteristics of each community and supported by fast-response predictive models capable of issuing warnings within seconds.
The central hypothesis of this research states that AI-driven, locally tailored hydrological models trained with community-generated data will provide faster and more accurate flood predictions than conventional hydrological models, especially in steep, fast-responding Andean basins such as the Mulato River.
This methodological approach is expected to strengthen local capacities for risk management, improve anticipatory response to extreme events, and provide a replicable framework for early warning systems in vulnerable Andean–Amazonian watersheds.
Keywords: community-based monitoring, early warning systems, artificial intelligence, participatory hydrology, rapid-response basins, flood risk management.
How to cite: Román-Chaverra, D., Romero-Hernández, C.-P., and Rodrigo-Ilarri, J.: Integrated Early Warning System Based on Community Monitoring and Artificial Intelligence: Methodological Framework for the Mulato River Sub-basin (Mocoa, Colombia), EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-668, https://doi.org/10.5194/egusphere-egu26-668, 2026.
Flood forecasting is evolving from predicting hydrological variables to estimating potential impacts, bridging the gap between hazard anticipation and decision-making. Flood impact forecasts are obtained by combining hazard forecasts with the relatively high-resolution exposure datasets; e.g., population density, health and education facilities, transport networks, and energy infrastructure, to support decision-making before and during the event. However, the evaluation of the impact forecast remains challenging. Beyond hydrometeorological forecast skill, a meaningful evaluation of impact forecasts must incorporate ground truth evidence of real-world impacts and feedback from operational end-users.
A Pan-European real-time flood impact forecasting system has been designed within the European project INLINE. The system uses precipitation forecasts generated by seamless blending of probabilistic radar-based nowcasts and the precipitation simulations of the ECMWF EPS (maximum lead time: 120 hours). These are the inputs to estimate the flash flood hazard probabilities throughout Europe, which are integrated with high-resolution open-source exposure datasets to estimate flash flood impacts. INLINE is conducting a 15-month large-scale demonstration with an extensive Community of Interest (COI) including hydrological institutions, civil protection agencies, and emergency managers.
This study presents a multi-criteria evaluation framework applied to assess the performance of the system during the demonstration period. The evaluation integrates four components: (i) Hydrometerological skill, comparing the blended forecast product against radar and gauge observations to evaluate accuracy, reliability and timeliness; (ii) Impact-based verification, evaluating the forecasted impact levels against a newly created real-world impact database, which collects impact information using an LLM-based algorithm through news and social media; (iii) User-centric operational value, quantifying the system’s usefulness, clarity and operational relevance through structured surveys within the COI; and (iv) added value, comparing the complementary of the project developments with the current operational tools used by stakeholders to quantify the improvement for emergency management.
Several representative flood events are analysed in detail to showcase the applicability of the evaluation framework applied to the different developments of the project, and particularly impact-based forecasts. The results underline the importance of combining technical performance metrics with real-world impacts and stakeholder perspectives to guide future operational implementation.
How to cite: li, X., Berenguer, M., Park, S., and Sempere-Torres, D.: Evaluating a Pan-European Flood Impact Forecasting System: A Multi-Criteria Framework Integrating Hydrological Skill and End-User Perspectives, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-848, https://doi.org/10.5194/egusphere-egu26-848, 2026.
Impact-based flood forecasting remains a major challenge for early warning systems, particularly in regions subject to rapid hydrological transitions and high societal vulnerability. Conventional approaches relying on pre-computed inundation maps and fixed impact thresholds often fail to capture event-specific dynamics, anticipate cascading impacts, and support timely emergency response. This study presents a real-time impact-based forecasting system that integrates physics-enhanced LSTM streamflow prediction, two-dimensional hydrodynamic simulation, and automated GIS-based impact assessment within a unified Python framework.
The workflow begins with a physics-enhanced LSTM model trained to provide short-range streamflow forecasts at key upstream stations. These forecasts drive an automatically executed HEC-RAS 2D model, producing time-evolving floodplain conditions beyond the static assumptions of threshold-based systems. By adopting dynamic simulations rather than pre-calculated inundation products, the system captures spatially and temporally explicit flood characteristics—advancing the representation of timing, extent, and hydraulic intensity during extreme or atypical events.
Hydrodynamic outputs are post-processed through a Python module that derives key impact metrics, including (i) direct economic losses via depth–damage functions, (ii) exposed and affected population, (iii) disruption of transportation links, (iv) impacts on critical facilities (e.g., hospitals, schools, emergency services), and (v) flood arrival times at operationally relevant locations. The arrival-time analysis provides essential lead-time information for emergency mobilisation, substantially enhancing situational awareness.
The system is demonstrated in the 8,850 km² upstream drainage area of the Piracicaba Basin (São Paulo, Brazil), a region characterised by hydrological sensitivity, rapid urbanisation, and recurrent flood emergencies. Results show that integrating machine learning, hydrodynamic modelling, and automated geospatial impact quantification improves the timeliness, accuracy, and operational relevance of flood warnings. The framework advances beyond hazard-centric forecasts by delivering transparent, event-specific impact information essential for effective early action.
All components of the framework rely on free and open-source tools, and all scripts developed in this study are openly available on GitHub to support transparency, reproducibility, and operational scalability.
How to cite: Perdigão Gomes Bezerra, R., Brentan, B., Solha, P., Eleutério, J., and Rodrigues, A.: Real-Time Impact-Based Flood Forecasting in the Piracicaba Basin, Brazil, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-867, https://doi.org/10.5194/egusphere-egu26-867, 2026.
Accurate tropical cyclone (TC) intensity forecasting remains challenged due to the lack of high-spatiotemporal-resolution observations of inner-core dynamics. This study introduces a novel structural indicator, Area-Mean Vorticity (VORm), derived from minute-scale FY-4B atmospheric motion vectors using the SPA-FABI framework. We identify a distinct "U-shaped" lifecycle in vorticity variability, identifying anomalous high-frequency fluctuations as robust precursors for TC rapid intensity change. Integrating VORm into linear (MLR, R2=0.97, RMSE=5.239 kt) and non-linear (XGBoost, RMSE=5.778 kt) models significantly enhances 6-hour forecast skill, with VORm ranking as a top-tier indicator alongside other well-known dynamical and thermodynamic environmental drivers. In physical terms, a critical synergy is established: environmental factors such as sea surface temperature (SST) define the theoretical ceiling of potential intensity, while VORm quantifies the efficiency of the TC inner-core engine in realizing this potential. Furthermore, SHAP (Shapley Additive Explanation) analysis also reveals that VORm serves as a low-variance "anchor" signal, stabilizing predictions against environmental uncertainty. Operationally, VORm fills the critical gap for real-time, high-fidelity structural predictors, offering a novel and effective pathway to reduce short-term TC intensity forecast errors.
How to cite: Xia, P.: A Vorticity-Based Indicator for Typhoon Intensity Forecasting, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5009, https://doi.org/10.5194/egusphere-egu26-5009, 2026.
Recent extreme precipitation events across Europe, including those in autumn 2024, underscore the need to strengthen proactive disaster risk reduction through improved impact-based early warning. In Austria, precipitation-related hazards such as landslides, flash floods and hailstorms repeatedly result in considerable impacts on people, infrastructure and economic assets. These challenges are expected to intensify under ongoing climate and environmental change. In response, European national meteorological and hydrological services are increasingly pursuing a paradigm shift in their warning strategies, from traditional weather warnings towards impact-based warnings (IbW). IbW focus on the consequences of weather events (“what the weather will do”) rather than solely on meteorological conditions (“what the weather will be”). However, data-driven and applicable approaches to predict precipitation-induced impacts at the national scale remain limited.
The PRE4IMPACT-AT project is part of the Austrian Climate Research Programme (ACRP) and addresses this gap by developing explainable and user-oriented impact-based predictive models for precipitation-related hazards in Austria. This contribution presents the overall project concept and the methodological framework, exemplified through a recent transferable and generalizable approach (Steger et al., 2025; https://doi.org/10.5194s/egusphere-2025-4940). PRE4IMPACT-AT focuses on processes whose impacts that typically occur in temporal and spatial proximity to precipitation events, namely landslides, flash floods and hailstorms.
Adopting a risk-oriented perspective, PRE4IMPACT-AT first conceptualizes impacts as the outcome of interacting atmospheric drivers, biophysical and geomorphological preconditions, and socioeconomic exposure and vulnerability. These relationships are formalized using an impact-chain framework, which supports the systematic identification and prioritization of key impact drivers for each hazard type. In subsequent steps, the selected drivers are parameterized and harmonized using a wide range of national datasets, including meteorological and geo-environmental information, as well as socioeconomic data. Model training relies on available national and international damage databases (landslides, flash floods) and agricultural insurance loss data (hail). Based on these datasets, explainable machine learning is applied to derive spatiotemporal predictive rules linking static and dynamic drivers to observed impacts. The resulting models aim to characterize typical impact conditions, with a strong emphasis on interpretability to enhance transparency and allow plausibility checks. The models are evaluated in hindcast and nowcast settings to assess their suitability for short-term impact-based warning applications. In addition, long-term analyses, synthesizing large numbers of hindcasts, are used to identify trends in critical conditions and emerging patterns. Finally, individual hazard-specific models are combined to provide a multi-hazard impact perspective. A core element of PRE4IMPACT-AT is continuous user engagement through iterative evaluation workshops with stakeholders who hold warning mandates. Overall, the project contributes to advancing impact-based forecasting, early warning and climate impact assessment by providing Austria with a transparent and operationally relevant foundation, while offering transferable insights for national services facing similar challenges across Europe.
This project is funded by the Climate and Energy Fund in the course of the Austrian Climate Research Programme (ACRP) and the FFG (www.ffg.at). The FFG is the central national funding agency and strengthens Austria’s innovative capacity.
How to cite: Imgrüth, D., Spiekermann, R., Schlögl, M., Lehner, S., Enigl, K., Schwarz, L., Ortner, G., Meyer, V., Hadzimustafic, J., Parajka, J., Valent, P., Komma, J., Gebhart, V., Bresch, D. N., Maraun, D., and Steger, S.: Methodical framework of the PRE4IMPACT-AT project: Exploiting explainable machine learning for impact-based early warning and trend analysis in Austria, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7318, https://doi.org/10.5194/egusphere-egu26-7318, 2026.
Drought forecasting is a key component of agricultural risk management, yet important gaps still remain in linking drought hazard indicators to measurable impacts on crop yields. To translate hydro-climatic drought information into actionable insights for agricultural decision-making, a systematic investigation of relationships between hazard variables and impact indicators is needed to support process understanding and predictive modelling.
In this study, we focus on selected crop yield anomalies in Sweden as key agricultural impact indicators, and characterise the timing, magnitude, and persistence of drought-related yield reductions. Then, we identify their links to drought hazard indicators, e.g. a set of meteorological, soil moisture, and hydrological drought indicators across relevant spatial and temporal scales, and explore their explanatory and predictive power. Building on the Framework for Index-based Drought Analysis (FRIDA), we leverage Machine Learning algorithms to elucidate the non-linear relationships between drought hazard indicators and crop yield impacts. Our results contribute to advancing impact-based drought early warning in Sweden and supports the development of more actionable drought information for agricultural stakeholders.
How to cite: Canedo Rosso, C., Mohammadi, B., Merlo, M., Giuliani, M., Pechlivanidis, I., and Du, Y.: Drought impact-based forecasting of crop yield in Sweden through a machine-learning framework , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7652, https://doi.org/10.5194/egusphere-egu26-7652, 2026.
Flooding is a natural hazard arising from complex and non-linear interactions between hydrometeorological forcing and landscape characteristics, and therefore cannot be reliably represented using simple empirical relationships. The objectives of this study are (1) to identify which hydrological and physiographic variables, or which combinations of them, are most strongly associated with flood-related consequences, and (2) to develop a national-scale flood susceptibility framework for Sweden that can be integrated with forecast information to support operational warning decisions.
The novelty of this work lies in the use of a large, nationwide impact dataset consisting of road closure records from 2000–2023, provided by the Swedish Traffic Agency, as the target for training and validation of a data-driven impact model. Each road closure location is characterized using a comprehensive set of predictors derived from the SHYPE hydrological model — including precipitation, runoff, soil moisture, groundwater storage, and short-term intensity metrics (e.g. 3-hour maxima) — together with topographic and environmental descriptors such as slope, elevation range, upstream contributing area, distance to water bodies and culverts, and land-use classes.
An Extreme Gradient Boosting (XGBoost) classifier was used to learn the relationship between these predictors and observed impacts. The model achieves strong predictive skill (accuracy = 0.977), with a balanced confusion matrix indicating strong ability to distinguish impacted and non-impacted areas. Feature importance analysis reveals that short-term hydrological response dominates model behavior. Surface runoff is the most influential predictor, followed by local runoff and groundwater storage, highlighting the critical role of near-surface hydrological dynamics in translating meteorological forcing into damaging outcomes. Topographic and land-use variables, such as slope and industrial land cover, further modulate susceptibility, emphasizing the influence of local terrain and exposure.
The resulting framework enables the generation of a dynamic flood susceptibility map for Sweden. When driven by real-time or forecast hydrometeorological inputs, the model can function as a “copilot” for forecasters, indicating where events are most likely to produce consequences. This would support more targeted warnings, reduces false alarms, and strengthens proactive risk communication in vulnerable areas.
How to cite: Karimi, S., Brendel, C., Lindqvist, K., Hjerdt, N., and Du, Y.: Integrating Machine Learning for Flood Impact Prediction in Swedish Operational Forecasting and Warning Services, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9728, https://doi.org/10.5194/egusphere-egu26-9728, 2026.
Typhoon Hagibis (2019), one of the most powerful storms to strike Japan in recent years, caused widespread flooding and severe damage. Impact-based forecasting play a critical role in planning effective mitigation measures and enhancing disaster preparedness and responses. In this study, we employ the Integrated Land Simulator (ILS) coupled with the Nonhydrostatic Icosahedral Atmospheric Model (NICAM) to evaluate the effects of typhoon intensity modification on flood damage mitigation associated with Typhoon Hagibis.
To systematically assess uncertainties in typhoon forecasts, we conducted ensemble simulations consisting of a control run and ten ensemble members. The results show that the spatial distribution of heavy rainfall and flooding is closely linked to the typhoon track. When the typhoon track shifted westward, heavy rainfall and flooding expanded over southwestern Japan. In contrast, eastward shifts in the typhoon track led to increased heavy rainfall and flooding in central Japan, with particularly strong impacts over the densely populated Kanto region.
To further investigate the effects of typhoon intensity modification on flood damage mitigation, the central pressure of the typhoon was artificially increased by 1 to 15 hPa at 1-hPa intervals on 10 and 11 October. These intensity modification experiments demonstrate that human intervention generally led to reductions in heavy rainfall and flood damage across Japan. Moreover, modifications applied on October 10 resulted in greater reductions in both heavy rainfall and flood damage than those applied on October 11.
These findings highlight the critical importance of both the intensity and timing of human intervention in influencing flood risk. By simulating different modification intensities and timings and explicitly evaluating the role of weather modification, this study advances our understanding of flood hazards and provides valuable insights for improving disaster preparedness and flood mitigation strategies.
How to cite: Li, X., Yoshimura, K., Nasuno, T., and Yamada, Y.: Impact-Based Ensemble Flood Forecasting in Japan: Effects of Typhoon Intensity Modification on Flood Damage Mitigation during Typhoon Hagibis (2019), EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16290, https://doi.org/10.5194/egusphere-egu26-16290, 2026.
In recent years, with improvements in weather forecasting technology, the development of long-term flood forecasting has advanced globally. This technology is expected to mitigate damage from large-scale floods, which occur infrequently but cause immense damage. However, it is also anticipated to be highly effective against frequent high-water events that occur routinely. This is particularly relevant for urban rivers, where frequent high-water limits the utilization of waterfront areas. Therefore, this study aims to expand the scope of long-term flood forecasting to address these high-frequency events, using the Oto River in Okazaki City, Aichi Prefecture, Japan, as a case study. We analyzed the impact of long-term flood forecast information on the decision-making of waterfront stakeholders through group interviews and workshops with 28 participants, including riverside business owners, municipal river managers, and academic experts.
The findings revealed that the level of demand for long-term flood forecasts varies significantly depending on the type of riverside use. For use that contains many physical installations or hardware, such as urban furniture and temporary structures, the evacuation process requires significant physical effort and time. Therefore, a high accuracy forecast with a lead time of 24 hours or more is essential, as it ensures a safe evacuation timeframe while avoiding unnecessary evacuations due to false alarms. Conversely, for “soft operations” like event hosting or rental businesses, a shorter lead time of 12 to 18 hours was shown to be an ample amount of time to determine event feasibility the day before and notify customers, allowing continued operations while controlling business risk.
A notable finding was that, regardless of whether the usage style was physical installations or soft operation based, when prediction accuracy exceeded 40-60%, users became more willing to accept risk, and the number of waterfront usage ideas increased dramatically. Furthermore, private businesses demonstrated a flexible stance, accepting false alarms in forecasts as an “insurance” cost for business continuity. With this approach, the construction of physical installations, which previously have been impossible due to high risks and strict standards, can broaden the types of businesses that can operate on the riverside, realizing a future urban landscape where permanent installations are standard. Based on these findings, it can be concluded that implementing long-term flood forecasting has the potential to significantly enhance the value of river spaces in daily life, extending beyond providing disaster prevention information for evacuation actions. By presenting appropriate lead times and accuracy levels, it suggests the potential to foster a new urban culture that coexists with waterfronts while accounting for flood risks, ultimately creating more diverse and resilient riverside urban spaces.
How to cite: Horie, K., Nakamura, S., and Morita, H.: The Impact of Long-Term Flood Forecasting on Waterfront Utilization and Stakeholder Decision-Making - A Case Study of the Oto River in Okazaki City, Japan, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16815, https://doi.org/10.5194/egusphere-egu26-16815, 2026.
Ensuring road safety is a critical responsibility for public organizations such as road network operators, emergency services, and national meteorological services (NMS). Traffic accidents arise from a complex interplay of environmental and human factors, making proactive risk management essential for road network operations. In practice, emergency services and road operators predominantly collect high-precision records of accident locations, resulting in presence-only datasets that lack explicit non-accident observations.
Unlike traditional accident modeling approaches that rely on labeled non-accident data or synthetically constructed negative classes, this study investigates one-class learning as a natural and operationally realistic framework for traffic accident analysis. Researchers at the Royal Netherlands Meteorological Institute (KNMI) explore the use of AI/ML methods to model high-resolution presence-only accident data using five years of traffic accident locations (2018–2022) provided by the Dutch road authority. Each accident is characterized by a set of weather and traffic intensity features describing the conditions under which it occurred.
Traffic accidents are modeled using neural one-class classification to obtain a high-dimensional embedding of accident conditions, which is subsequently analyzed using dimensionality reduction techniques to identify clusters of accidents with similar environmental signatures. By learning directly from observed accident occurrences, the approach enables the identification and comparison of recurring accident patterns associated with specific weather and traffic conditions, providing a structured basis for further analysis of weather-related traffic risk.
How to cite: Garcia-Marti, I., Whan, K., van Dijk, T., Stepek, A., Schönthaler, A., van den Besselaar, E., Zaanen, K., Derks, R., Ubels, S., and den Dulk, T.: Exploring weather and traffic conditions in traffic accidents using one-class learning , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19313, https://doi.org/10.5194/egusphere-egu26-19313, 2026.
Recent advances in ensemble meteorological forecasting, hydrological modelling, and flood inundation mapping have substantially improved flood hazard prediction. However, major gaps persist in translating hazard information into understandable, trusted, and actionable warnings based on the impacts of flood events, limiting the effectiveness of early warning–early action systems. A key challenge lies in linking hydrological hazard forecasts with exposure and vulnerability information to support impact-based decision-making.
Constructing and evaluating flood disaster risk forecasts remains a complex and uncertain process, particularly due to the multi-dimensional and spatially heterogeneous nature of vulnerability and exposure data. Impact-based Forecasting (IbF) of flooding seeks to address these challenges by explicitly connecting flood hazard forecasts to potential societal impacts in space and time.
FHIM-India – Flood Hazard Impact Model for India – is an impact-based flood forecasting framework that integrates ensemble numerical weather predictions, distributed hydrological modelling (Grid-to-Grid), and hydrodynamic flood simulations (SynxFlow) with exposure and vulnerability datasets. Here, FHIM-India is evaluated for fluvial flood impacts in the state of Kerala, south-western India using over 30 years of recorded impacts.
The FHIM-India framework is repurposed to generate daily flood impact hindcasts for multiple districts in Kerala over the period 1991–2022 using observed rainfall data as input. Modelled impact indicators related to affected population and property are evaluated against reported historical impact data. The performance of the impact-based hindcasts is assessed relative to warnings derived using fixed rainfall threshold-based approaches.
Results indicate that FHIM-India improves the identification and spatial discrimination of mid- to high-severity flood events compared with warnings based on fixed rainfall thresholds. The framework demonstrates strong potential for use in operational impact-based flood forecasting to support early warning systems and risk-informed decision-making.
How to cite: Mashhadi, A., Cole, S. J., Wells, S. C., Xia, X., and Moore, R. J.: Impact-based flood forecasting in India: evaluation of FHIM-India for fluvial flood impacts in Kerala, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20831, https://doi.org/10.5194/egusphere-egu26-20831, 2026.
Millions of people worldwide are affected by river floods each year. To facilitate early action, humanitarian organisations have adopted anticipatory action frameworks that link pre-agreed activities and their funding with forecasted peak river flow thresholds. However, the scale of humanitarian needs is primarily determined by flood impacts rather than hazard magnitude alone, limiting the effectiveness of streamflow-based triggers.
In the Humanitarian Action Challenges project we work closely with our humanitarian partners, UN OCHA and the Netherlands Red Cross, to engage with national Red Cross societies and key stakeholders in Ethiopia, Nigeria, and Uganda. The goal of the project is to move beyond streamflow thresholds alone to additionally provide impact forecasts, such as estimates of affected populations, in order to improve anticipatory action of humanitarian organisations.
As a first step, and to assess the feasibility of this approach, we analyse past river flood events in Ethiopia, Nigeria, and Uganda. We combine flood extents derived from Global Flood Awareness System (GloFAS) discharge forecasts and JRC hazard maps with geospatial data on population exposure and vulnerability using the open-source risk assessment platform CLIMADA. Modelled affected populations are compared with reported impacts using an event severity ranking. No systematic bias is observed, with both over- and underestimation across events. Rankings are highly sensitive to the inclusion of flood protection standards from the FLOPROS dataset. Comparisons with remotely sensed flood extents and analyses of model drivers highlight key limitations and sources of uncertainty for trigger calibration. These preliminary insights support the development of impact forecasts and the design of impact-based triggers for anticipatory action by humanitarian partners.
How to cite: Kobler, E., McCaughey, J., Severino, L., Riedel, L., van den Homberg, M., Teklesadik, A., Milano, L., and Bresch, D.: River Flood Impact Forecasting to Support Humanitarian Anticipatory Action, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-22105, https://doi.org/10.5194/egusphere-egu26-22105, 2026.
Small Island Developing States (SIDS) in the Eastern Caribbean face escalating hurricane risk under climate change, with impacts driven by compound hazards including extreme wind, rainfall, and storm surge. Anticipatory Action (AA) mechanisms—where predefined actions are activated based on forecast thresholds—offer a means to translate advances in climate and weather prediction into timely, risk-reducing interventions. However, designing robust, decision-relevant trigger models that balance forecast skill, uncertainty, and operational feasibility remains a key challenge, particularly in multi-country contexts.
We studied the feasibility of a sub-regional Early Action Protocol (EAP) covering Saint Kitts and Nevis, Dominica, and Antigua and Barbuda, and focused specifically on how to design a sub-regional trigger model. Using stakeholder consultations, analysis of national disaster management systems, analysis of historical and synthetic events by modelling wind, surge, and rainfall, and review of existing forecasting products, we assessed trigger options across temporal scales, compound hazard components, and impact relevance.
Results show that the wind and track forecasts from the US National Hurricane Centre demonstrated substantial improvements in accuracy over recent decades. The NHC’s 48-hour track error now averages about 90 km, meaning that areas at risk can be identified with an acceptable uncertainty in terms of storm size and asymmetry. Early actions possible within this lead time can include mobilizing communities, cash distributions, and prepositioning stock. Also, the NHC forecast is the official source, widely adopted by the respective national agencies in the three countries. In the future, the trigger model could be improved by, for example, ECMWF’s AIFS, Google DeepMinds GraphCast, or Microsoft Research’s Aurora, as these have demonstrated the ability to deliver medium-range forecasts with skill comparable to or surpassing traditional numerical models. While these AI models are not yet operational tools at national centres, they are available for experimental use and could be incorporated through the Caribbean Institute for Meteorology and Hydrology (CIMH) as complementary resources for rapid local updates and scenario planning within a newly developed anticipatory framework. In that case, a layered trigger architecture could be designed, containing: (i) probabilistic tropical cyclone track and intensity forecasts; (ii) impact-oriented thresholds linked to rainfall accumulation, wind exposure, and storm surge; and (iii) contextual readiness criteria reflecting response capacities.
Our study highlights key design principles for anticipatory trigger models in SIDS now and in the future: transparency, simplicity, tolerance to forecast uncertainty, and alignment with decision timelines for early action. By articulating how forecast information can be operationalised across borders, this contribution advances the integration of climate services and anticipatory humanitarian action in highly exposed island regions. A sub-regional trigger model can leverage shared meteorological information and pooled technical expertise, while allowing country-specific activation thresholds to account for differing exposure and coping capacities. A future initiative will focus on scaling up to Barbados and Belize.
How to cite: van den Homberg, M., Teklesadik, A., Markodimitraki, C., Viton, M.-T., Mouton, J. ., Duchemin, M., de Valk, L., and Kumbikano, M.: Designing Sub-Regional Anticipatory Action for Hurricanes in the Eastern Caribbean, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-23190, https://doi.org/10.5194/egusphere-egu26-23190, 2026.
Posters virtual: Fri, 8 May, 14:00–18:00 | vPoster spot A
EGU26-1658 | ECS | Posters virtual | VPS11
Integrating Intensity–Duration and Antecedent Rainfall Thresholds for Shallow Landslide Prediction in the Eastern Himalaya, IndiaFri, 08 May, 14:03–14:06 (CEST) vPoster spot A
- Rainfall-induced shallow landslides pose a persistent hazard in the Eastern Himalaya, particularly along the strategically important Balipara–Charduar–Tawang (BCT) corridor in western Arunachal Pradesh, India. This study develops a region-specific rainfall threshold framework by integrating long-term rainfall trend analysis with empirical landslide-triggering thresholds to enhance early warning capabilities in this data-scarce, high-relief terrain. Daily gridded rainfall data from the India Meteorological Department (2000–2020) and an inventory of 236 landslide events recorded between 2008 and 2015 were analyzed. Trend analysis reveals a statistically significant decline in annual rainfall (–81.05 mm yr⁻¹), accompanied by pronounced inter-annual variability and persistent monsoonal dominance. Empirical analysis indicates that short-term antecedent rainfall plays a critical role in slope failure initiation, with 3-day and 5-day cumulative rainfall showing the strongest correlation with landslide occurrence (R² = 0.508 and 0.480, respectively). Corresponding 80th percentile thresholds of ≥89.24 mm (3-day) and ≥118.80 mm (5-day) are proposed as practical triggering criteria. In addition, an intensity–duration (I–D) threshold derived from 95 rainfall-induced landslides follows a negative power-law relationship (I = 17.26·D⁻⁰·¹⁰), capturing the influence of short-duration, high-intensity rainfall events. The combined use of antecedent rainfall and I–D thresholds effectively represents both progressive soil saturation and rapid-onset rainfall triggers. This integrated threshold framework provides a robust and scalable basis for landslide early warning system development along the BCT corridor and offers broader applicability to similar monsoon-dominated Himalayan regions.
How to cite: kar, S., Chaturvedi, P., and Negi, H. S.: Integrating Intensity–Duration and Antecedent Rainfall Thresholds for Shallow Landslide Prediction in the Eastern Himalaya, India, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1658, https://doi.org/10.5194/egusphere-egu26-1658, 2026.
