Spatio-temporal variability of the terrestrial hydrological processes (land heat and water storage anomalies) has important implications in the climate predictability through their effects on surface energy and water fluxes. The changes in seasonal precipitation patterns associated with the Indian Summer Monsoon can alter the hydrological processes; for a given catchment, which in turn can influence the exchange of water and energy at the land surface-atmosphere interface. Hence the reliable prediction of the basin-scale water cycle components in a physically based high-resolution hydrological model equipped with sophisticated Land Surface Models (LSMs) is of prime requirement. The modern LSMs can provide detailed representations of important biophysical, biogeochemical and hydrological processes of varying spatial and temporal scales by incorporating the necessary feedbacks between the land and the atmosphere. When coupled to a physically based fully distributed hydrological model, it can affect the soil moisture patterns means of recycling the surface and sub-surface runoff (lateral terrestrial flow). However, despite the role of lateral terrestrial hydrological processes for the improved simulation of soil moistures, the sensitivity studies involving the land surface and sub-surface feedbacks are less pronounced especially for a tropical humid region with complex physiographic settings (presence of complex topography) under monsoon regimes (strong synoptic forcings). Therefore, in the present study, we examined a process based diagnosis regarding the role of the lateral flow on the terrestrial hydrological processes (Evapotranspiration, surface and sub-surface runoff, stream flow) and surface energy fluxes (latent heat, sensible heat) by using a multi-configured modeling framework of offline WRF-Hydro with Noah-Multi parameterizations (MP) LSM to enable systematic evaluation of the multiple physical parameterizations of hydrologic process representation; the validation has been done with the reanalysis dataset, a remotely sensed product and ground based observations.

How to cite: Sarkar, S. and Lakshmikanthan, P.: Modeling the impact of lateral flow on terrestrial water balance components and surface energy fluxes using WRF-Hydro with multi-configuration ensembles: a study over Krishna River Catchment, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1783, https://doi.org/10.5194/egusphere-egu25-1783, 2025.

In a mountainous watershed, there are many confluences at which two or more streams join. Due to inaccessible terrain and associated costs, river discharge data is collected only at a few confluences. It is, therefore, important to assess which confluence is critical. By critical, we mean the junction which will create maximum fragmentation in a river network. In this study, we analysed river networks with uneven topography in the Alaknanda River basin, which is vulnerable and prone to geo-hydro hazards. We applied Unsupervised Machine Learning (UML) algorithms such as Isolation Forest, Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and Linear Integer Programming (LIP) to identify the critical confluence locations. We compare our results with the well-established graph-based centrality metrics (Degree centrality, Betweenness centrality, Closeness centrality, and Eigen Vector Centrality). Our results suggest that DBSCAN outperformed other approaches in terms of detecting crucial nodes. We obtained better results using LIP than other techniques except DBSCAN. The outcome of this study will help the Central Water Commission, in deciding which confluence to focus on, and in assessing the locations of new gauges.

Keywords: Critical nodes; Alaknanda Basin; Machine Learning; Hazards

How to cite: Rajouria, N., Parajapati, P., and Jha, S. K.: Application of Unsupervised Machine Learning Algorithms for identifying critical river confluence in a mountainous watershed., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19050, https://doi.org/10.5194/egusphere-egu25-19050, 2025.

Predicting streamflow in ungauged basins remains a significant challenge in hydrological studies. In recent years, data-driven models have been shown to outperform traditional physics-based models in streamflow prediction for ungauged catchments. However, few studies have examined the potential of such models for predicting streamflow in ungauged basins within India. This study aims to evaluate the performance of two machine learning models, namely Support Vector Regression (SVR) and Random Forest (RF), alongside two deep learning models, Long Short-Term Memory (LSTM) and Bi-LSTM, in the context of streamflow regionalization within the Krishna River Basin in India. Each prediction model is trained using meteorological variables as input features, with streamflow as the output variable. K-means clustering is employed to group selected catchments (based on data availability) into an optimum number of clusters based on spatial proximity and physical similarity. It is assumed that catchments within a cluster share homogeneous characteristics. Regionalization is achieved by sharing model parameters across catchments within the same cluster. For each cluster, one gauged catchment is designated as the donor catchment, while the others are treated as pseudo-ungauged. Each proposed model is trained and tested using the meteorological inputs and streamflow data available at the gauged donor catchment. The trained model for each cluster is then transferred to the remaining receptor catchments within the cluster, where the meteorological variables corresponding to each ungauged catchment are used as inputs. The performance of the models in ungauged catchments is rigorously evaluated by comparing the simulated streamflow against observed streamflow using metrics such as Nash-Sutcliffe Efficiency (NSE), Root Mean Squared Error (RMSE), Coefficient of Determination (R²), and Percentage Bias (PBIAS). This study highlights the advantages of utilizing data-driven methods for streamflow prediction in both gauged and ungauged basins, particularly due to their ability to capture complex, non-linear relationships between meteorological inputs and streamflow generation. The findings of this study are expected to be instrumental in water resources planning and management, flood assessment, and the design of hydraulic structures in the Krishna River Basin.

How to cite: Kaur, S. and Chavan, S.: Data-driven models for streamflow regionalization in Krishna River Basin, India, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15731, https://doi.org/10.5194/egusphere-egu25-15731, 2025.

The prior literature on hydrologic model performance is dispersed, encompassing a small number of catchments, different methodology, and rarely linking the results to specific catchment characteristics. This study addresses these constraints by systematically attributing model performance to catchment variables in 671 US catchments, providing a formal framework for determining the best models for specific conditions. Daily streamflow estimation was performed using eight process-based (PB) models and three deep learning (DL) models, with performance measured using the Nash-Sutcliffe Efficiency (NSE). The PB models were tested with a variety of optimization techniques, and the most effective approach for each model was chosen based on the number of catchments that exceeded a predetermined performance threshold. Four models were selected as the top performers based on three performance metrics. Further analyses, such as Classification and Regression Tree (CART) and SHAPley, were used to correlate model performance with catchment variables across all models.
The results showed that PB models (GR4J, HBV, and SACSMA) performed well in catchments with low to medium aridity and a high Q/P ratio, indicating quick hydrologic responses. In contrast, the LSTM-based DL model performed well in medium to high aridity regions but had limits in catchments with rapid precipitation responses and low sand percentages. These findings provide a thorough understanding of the links between model performance and catchment descriptors.

Keywords: Process-based models, Deep learning model, CART analysis, SHAPley analysis, catchment characteristics.

How to cite: sourya, D. A., manikanta, V., and rathinasamy, M.: Can the catchment features influence the performance of the conceptual hydrological and deep learning models? A study using large sample hydrologic data , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18458, https://doi.org/10.5194/egusphere-egu25-18458, 2025.

Knowing evapotranspiration (ET) accurately at fine spatial scales is very important. This would improve understanding hydrological processes and contribute to the advancement of water resource management. In this study, we set a framework based on deep learning to downscale Terra Net Evapotranspiration Gap-Filled 8-Day Global 500m dataset, developed and managed by NASA's Earth Observing System. This approach resulted in a scale enhancement of 20 times. The U-Net architecture was used for this purpose. It incorporated MODIS Land Cover Type 1 (LC Type 1) as an auxiliary variable. This was done to account for land cover changes. The study covered a diverse region that encompasses latitudes 28° to 32°N and longitudes 74° to 78°E. A synthetic design of experiments was utilized to systematically generate and evaluate training data, this ensures robust model performance and reliable downscaling outcomes across the heterogeneous terrain of the study area. Model training, validation, and testing were conducted using the 2001–2014 dataset, 2015–2018, and 2019–2023 dataset, respectively. The model showed excellent performance on the testing dataset. The average PSNR was 34.35 dB and the mean SSIM was 0.8517. The U-Net module effectively downscale and enhance the spatial resolution of ET data. The results show ET's spatial and structural features are well preserved. This study shows how deep learning improves climate data spatial resolution. It provides reliable local hydrological and agricultural resources.

How to cite: Jha, S. K. and Gupta, V.: Downscaling MODIS ET using deep learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18599, https://doi.org/10.5194/egusphere-egu25-18599, 2025.

Accurate and timely representation of spatiotemporal precipitation patterns is critical for monitoring and predicting hydrological extremes, particularly in operational hydrology and early warning systems. In regions with limited in-situ precipitation data, satellite precipitation products (SPPs) offer an accessible solution. However, the latency of these datasets—the delay between data collection and availability—remains a key challenge for real-time applications. This study developed a machine learning model based on the Random Forest (RF) algorithm to predict precipitation using low-latency data from GOES-16 Advanced Baseline Imager (ABI) bands. The model was applied to the Jubones River basin (3,391 km²) in southern Ecuador, a region characterized by complex terrain and hosting a key hydropower project. Leveraging hourly data over a five-year period, the RF model addressed the five-hour latency of traditional SPPs by generating near-real-time precipitation maps with a latency of only 10 minutes. The model’s performance was evaluated using quantitative and qualitative metrics across temporal scales, demonstrating progressive accuracy improvements with larger temporal aggregations. Root Mean Square Error (RMSE) values decreased from 0.48 to 0.05 mm/h, while Pearson’s Cross-Correlation (PCC) improved from 0.59 to 0.87 for scales ranging from hourly to monthly. Qualitative metrics, including Probability of Detection (POD), False Alarm Ratio (FAR), and Critical Success Index (CSI), further validated the approach. These findings highlight the potential of integrating advanced hydroinformatics techniques with remote sensing for managing hydrological extremes in diverse basins. The study underscores the importance of leveraging low-latency satellite data and machine learning to enhance real-time forecasting and operational hydrology. Future work will focus on refining the model for improved detection of extreme precipitation events and exploring its integration into stakeholder-driven decision-making frameworks.

How to cite: Muñoz, J., Muñoz, P., Muñoz, D. F., and Célleri, R.: Leveraging machine learning and satellite precipitation data to overcome latency challenges in operational hydrology, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19086, https://doi.org/10.5194/egusphere-egu25-19086, 2025.

Landslides are a widespread geohazard with significant impacts on lives and economies worldwide. While past research has primarily emphasized creating inventories, and analysing spatial and temporal patterns, the objective of this study is to explore the relationship between landslides events taken place in different catchments using only topographical and physical attributes from the disasters’ areas. The aim is to improve the understanding of the occurrence and susceptibility of such events, as well as the possible similarities between the events and the catchments. To this end, multicollinearity and mutual information analysis were performed to identify both linear and nonlinear relationships between the variables, assisting on the identification of the most relevant driving factors to historical landslides in the study area. Furthermore, the events were grouped using 5 different unsupervised clustering techniques, KMeans, Mean Shift, DBSCAN, Hierarchical and Spectral Custering, to analyse the relationship between landslides taken place in different catchments and their underlying driving forces. Clustering evaluation metrics, i.e. Silhouette Score, Davies-Bouldin Index, and Calinski-Harabasz Index, were used assess the performance of these algorithms. The results show that, for a preliminary study and providing insights on the relevance of driving factors and similarities between events, unsupervised learning proves to be an important tool. Nevertheless, to find more applicable and in-depth associations between extreme disasters and its driving factors, more robust machine learning techniques can and should be used.

How to cite: Antunes Meira, M., Xuan, Y., and Wang, H.: Using Unsupervised Learning to Explore Landslides Driving Factors from Topographic and Hydrological Catchment Features, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6708, https://doi.org/10.5194/egusphere-egu25-6708, 2025.

Runoff forecasting remains a critical challenge in many basins worldwide, particularly those featuring a complex topography, where the scarcity of hydrometeorological data is a prevalent challenge. Data fusion offers a promising alternative to conventional single-source data modelling, which often fails to capture the full spatial and temporal variability of precipitation. By integrating multiple sources, data fusion seeks to generate enhanced satellite precipitation datasets, essential for data-driven runoff forecasting models. This study aims to evaluate the effectiveness of the Three-Cornered Hat (TCH) method for fusing satellite precipitation products (SPPs) and its influence on the performance of a Random Forest-based runoff forecasting model.

Three scenarios were evaluated: (i) a TCH-fused dataset combining three SPPs: Integrated Multi-satellitE Retrievals for GPM – Early Run (IMERG-ER), the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks – Cloud Classification System (PERSIANN-CCS) and the Global Satellite Mapping of Precipitation – Near Real Time (GSMaP-NRT); (ii) an individual SPP (IMERG-ER); and (iii) an already fused benchmark product, the Multi-Source Weighted-Ensemble Precipitation (MSWEP). All scenarios performed comparably for lead times of 3, 6, 12, and 24 hours, with MSWEP slightly outperforming across Nash-Sutcliffe Efficiency, Kling-Gupta Efficiency, and Root Mean Square Error metrics. However, TCH demonstrated better bias reduction as reflected by the Percent Bias metric.

A key limitation of the fusion method was identified at hourly scales, where statistical dependence arises during periods with no precipitation over the basin, hindering the effectiveness of TCH. The introduction of a matrix regularization step addressed this issue. This study provides valuable insights for enhancing SPP fusion methods and offers a replicable framework for improving runoff forecasting, particularly in data-scarce regions and other hydrological contexts.

How to cite: Luna-Abril, P., Muñoz, P., Samaniego, E., Muñoz, D. F., Merizalde, M. J., and Célleri, R.: Evaluating the Three-Cornered Hat Method for Satellite Precipitation Data Fusion and its Influence on Runoff Forecasting, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16242, https://doi.org/10.5194/egusphere-egu25-16242, 2025.

Understanding and modeling surface and groundwater resources are critical due to the effects of droughts and climate change, especially in semi-arid, arid, or hyper-arid regions. GeoLinkage, developed by Troncoso (2021), facilitates the creation of linkage files for integrated models. These linkage shapefiles act as a communication interface between a surface hydrological domain (D₁) and an aquifer domain (D₂). The surface domain (D₁) comprises nodes and arcs that represent hydrological elements and their relationships, while the aquifer domain (D₂) contains geometric elements such as grids or Quadtree diagrams. D₁ defines a surface topology (τ₁), D₂ defines a groundwater topology (τ₂), and the linkage file establishes a surface-groundwater topology (τ_1-2). This new topology, τ_1-2 ,imposes constraints that influence the relationship between τ₁ and τ₂. For instance, the superposition of elements in τ_1-2 should be considered a spatial relationship. Depending on the type of superposed elements, this relationship must be reflected in τ₁  or τ₂. To enforce these τ_1-2 specific restrictions, GeoLinkage has been enhanced with a post-processing module called GeoChecker. This module evaluates the quality of the resulting linkage files. GeoChecker currently performs a superposition check to ensure that overlaps between cells in the linkage file—whether between groundwater and catchments or groundwater and demand sites—are accurately represented as connections in the surface model (WEAP). The aquifer is represented by a MODFLOW model fully linked to the WEAP model. GeoLinkage2.0 and GeoChecker were developed using the tutorial WEAP-MODFLOW model, considered a small model, and were tested in large integrated models, such as the Azapa Valley (3,000 km²) and the Limarí River Basin (12,000 km²), Chile.

How to cite: Sanzana, P., Torga, A., Hitschfeld, N., and Lobos, C.: GeoLinkage2.0 and GeoChecker: Hydroinformatics tools for large and complex hydrological-hydrogeological models using WEAP-MODFLOW. Case Study: Severe drought in the Limarí River Basin, Chile, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20678, https://doi.org/10.5194/egusphere-egu25-20678, 2025.

The release of low-temperature water from a reservoir can have negative impacts on downstream fish spawning and crop growth in irrigation areas. Therefore, predicting the discharged water temperature accurately and swiftly is crucial. This study focused on the Pubugou Hydropower Station, a major project situated on the Dadu River in the upper reaches of the Yangtze River, and evaluated the impacts of meteorological factors and reservoir operational parameters on the released water temperature using Spearman correlation coefficients (R). To predict the discharged water temperature of Pubugou Reservoir, five models were optimized by genetic algorithms including random forests, support vector regression, convolutional neural network, long short-term memory network, and the lightweight gradient boosting machine respectively. The results showed that: (1)The dew point temperature exhibited the highest correlation with discharged water temperature (R = 0.89), However, the correlation coefficient between wind speed, cloud cover, solar radiation, dam front water level, and discharge water temperature was not found to be 0.4. (2) All the five models optimized by genetic algorithms performed well on the training set, especially the random forest model (R² = 0.997). The worst performing model is the long short-term memory network model (R² = 0.985). (3) In the prediction of discharge water temperature, all models have good fitting effects, with r² greater than 0.93, average absolute error not greater than 0.662 ℃, and mean square error not greater than 0.852 ℃. Random forest models and lightweight gradient boosting machine models have shown good performance on the most of sample data, with a small residual range, while support vector regression models and convolutional neural network models have smaller maximum residuals. This research indicated that machine learning methods can effectively predict water temperature released from reservoirs, providing more reliable decision support for formulating relevant measures to alleviate the impact of reservoir discharge water temperature.

How to cite: Junguang, C.: Application of Machine Learning in Predicting the Water Temperature Released from Reservoirs, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18342, https://doi.org/10.5194/egusphere-egu25-18342, 2025.

Drought is one of the most severe climate-induced phenomena; with significant impacts on agriculture, water resources, and ecosystems. Drought monitoring under climate change scenarios becomes crucial, particularly in regions vulnerable to water scarcity, such as semi-arid areas in Iran. Although Global Climate Models (GCMs) contain coarse spatial resolutions, they provide valuable insights in better assessing the variability of drought characteristics—such as duration, severity, and intensity in the future. To achieve this aim, downscaling of climate variables as triggers of droughts is required to monitor drought in local scale. Latyan region in Iran, as an important area to supply water, is a critical place based on its climate, drought event occurrences, and water demand and supply stress. This study tried to accurately downscale and bias-correct the climate variables utilizing the latest CMIP6 models (ACCESS-CM2, BCC-ESM1, CanESM5, HadGEM3-GC31-LL, and MIROC6) and AI techniques in the case study. This research employs a predictor selection technique in conjunction with a stack generalization model to improve the accuracy of the downscaling process. After careful examination of predictors, surface temperature, precipitation, and surface air pressure have been used along with annual cycles for training four machine learning models including Multilayer Perceptron (MLP), Support Vector Regression (SVR), Random Forest and Stack Generalization (SG) models for the sake of downscaling. Results showed that MIROC6 model is the best model according to all downscaling methods. In addition, among MLs, stacked generalization model improved the statistical metrics considerably with a Nash-Sutcliffe Efficiency (NSE) of 0.64, Mean Squared Error (MSE) of 1051.3, and Kling-Gupta Efficiency (KGE) of 0.68 for MIROC6 model. Selection of the proper GCM and downscaling method can help decision-makers take proper measures against drought to reduce drought impacts.

How to cite: Mirdarsoltany, A., Rahimi, L., Anderson, C., and Graf, T.: Fusion of Stacked Generalization and Predictor Selection Technique for Downscaling in Drought Monitoring: A Case Study in a Semi-Arid Area, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7150, https://doi.org/10.5194/egusphere-egu25-7150, 2025.

The São Francisco River Basin is crucial for Brazil’s agriculture, hydropower, and water security. However, climate change has intensified challenges like reduced water flow and frequent extreme events, threatening its socio-economic sustainability. This study aims to forecast flow in the São Francisco River Basin, enabling proactive decision-making to mitigate risks associated with both droughts and floods. To address these challenges, this study propose a novel methodology based on Artificial Intelligence (AI), combining Recurrent Neural Networks (RNN) and complex network techniques. The method creates new features and assigns importance weights to enhance the algorithm’s ability to generate probabilistic flow forecast. The results are promising, demonstrating the method’s ability to deliver accurate probabilistic forecasts. This research can support risk mitigation strategies and improve water resource management in the São Francisco Basin. Additionally, the proposed framework is scalable, offering potential applications to other critical watersheds facing similar challenges

How to cite: Caseri, A., Aparecido Rodrigues, F., and Victal Cerqueira, M.: Ensemble Approach for Hydrological Forecasting Based on Recurrent Neural Networks and Complex Networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13991, https://doi.org/10.5194/egusphere-egu25-13991, 2025.

The increasing frequency and intensity of climate hazards, as emphasized by the IPCC’s Sixth Assessment Report, underscore the urgent need for effective disaster risk reduction strategies. Using the devastating floods of September 2024 in Nepal’s Kathmandu Valley, and the April 2024 floods in Kenya’s Nairobi, this study examines the persisting gaps in flood resilience despite early warnings using disaster forensics techniques. The Kathmandu floods, which were triggered by an extreme rainfall event resulting from the convergence of a low-pressure system from the Bay of Bengal and a cyclonic circulation from the Arabian Sea, caused extensive loss of life, property damage, and economic disruption in the Nakhu Khola watershed, despite timely early warnings issued by the government. In Kenya, a notable gap exists in the warning systems, whether in their issuance, dissemination, or uptake, despite the presence of advanced operational forecasting systems. Encroachment on floodplains, unplanned urbanization, and land-use changes have exacerbated vulnerability, while weak governance and poor enforcement of disaster risk management legislation has left populations and assets exposed. Additionally, risk assessment efforts are scarcely integrated into government plans or those of other stakeholders, highlighting a critical area for improvement in disaster preparedness and management.

Using the UNDRR’s forensic disaster analysis framework, this research investigates the underlying causes, risk drivers, and lessons from these events. The populations most affected are identified, including those living in floodplains, including marginalized communities, and critical infrastructure. Local investments in disaster preparedness are also critically examined for efficacy. Results highlight that while early warnings were disseminated through various channels, gaps in risk communication and community-level preparedness persisted. The findings emphasize the need for education and awareness and integrated approaches to disaster risk management that address root causes such as unplanned urban growth and environmental degradation. Empowering youth and fostering leadership in disaster risk reduction are critical to ensure climate resilient societies of tomorrow. This research contributes actionable insights to reduce vulnerability, enhance preparedness, and minimize losses in future climate hazard events in the Kathmandu Valley and Kenya, as well as similar rapidly urbanizing regions.

How to cite: Kettner, A., Gupta, A., Singh Shrestha, M., Trigg, M., Cohen, S., Hawker, L., Prades, L., Rudari, R., Salamon, P., Tellman, B., Sperna Weiland, F., and Wu, H.: Devastating Flooding Despite Early Warning: Lessons Learned from the Nepal and Kenya Floods, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21199, https://doi.org/10.5194/egusphere-egu25-21199, 2025.

Flooding has posed a significant challenge to urban infrastructure, necessitating effective and real-time risk management strategies [1]. One of the most devastating impacts is on urban transportation, where disruption can lead to significant economic losses or even human casualties [2-3]. This study has focused on the key financial and commercial areas in downtown London, where an innovative system has been developed to integrate real-time flood risk forecasting with traffic data visualisation and dynamic decision support for emergency response and resource allocation. First, with access to the Google Maps API, real-time and forecast traffic data have been collected for local streets. Then, these datasets can facilitate a 15-minute resolution forecast for the next 8 hours, enabling an in-depth understanding of traffic flow patterns during flood events. Furthermore, by employing flood forecasting measures on these real-time datasets, streets at risk of inundation can be identified faster, with their traffic conditions assessed accordingly.

A key aspect of this study is to consider different factors dynamically for weighting and prioritising streets. On one hand, pre-existing factors such as road hierarchy, connectivity, access to critical facilities, land use, infrastructure vulnerability, and proximity to evacuation zones are converted into dynamic factors by attaching a temporal variable to these pre-existing factors. On the other hand, real-time dynamic ones include flood depth, traffic congestion, accessibility for emergency services, and community needs reported. The integration of all these factors leads to the development of a transportation-based decision support system (TBDSS) tailored to urban flood management. The TBDSS has facilitated the allocation of emergency resources, prioritisation of street reopening, and planning for evacuation or relief operations. For instance, streets connecting to hospitals or shelters have been given higher priority, while those serving industrial or low-density areas have been weighted lower. As such, our proposed system can dynamically adjust priorities based on evolving flood and traffic conditions, ensuring optimal response strategies.

The findings have demonstrated the feasibility of leveraging real-time data and advanced modeling to enhance urban flood resilience. By combining flood risk maps, traffic forecasts, and a comprehensive prioritisation framework, this approach has provided a promising tool for urban planners and emergency responders.

[1] Piadeh, F., Behzadian, K., Alani, A. (2022). A critical review of real-time modelling of flood forecasting in urban drainage systems. Journal of Hydrology, 607, p.127476.

[2] Gao, G., Ye, X., Li, S., Huang, X., Ning, H., Retchless, D., Li, Z. (2024). Exploring flood mitigation governance by estimating first-floor elevation via deep learning and google street view in coastal Texas. Environment and Planning B: Urban Analytics and City Science, 51(2), 296-313.

[3] Naghedi, S. N., Piadeh, F., Behzadian, K., and Hemmati, M.: Unveiling the Interplay: Flood Impacts on Transportation, Vulnerable Communities, and Early Warning Systems, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-13189, https://doi.org/10.5194/egusphere-egu24-13189, 2024.

How to cite: Naghedi, R., Piadeh, F., Huang, X., and Wu, M.: Real-time Transportation-Based Flood Warning System: A Case Study in Downtown London, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3543, https://doi.org/10.5194/egusphere-egu25-3543, 2025.

Digital twins, virtual representations of physical systems, integrate sensor data and predictive models to enable real-time simulation and analysis. They are instrumental in monitoring weather, infrastructure health, and water levels, particularly in flood management. By modeling mitigation techniques, forecasting risks, and enhancing emergency responses, digital twins improve decision-making, reduce economic losses, and enhance public safety in flood-prone areas [1][2]. This study developed a digital twin system to monitor and forecast flood disasters in western Iran. The system combined multidimensional sensor data on temperature, flood flow, vegetation cover, and water levels using an offline databank. Time-series analysis tracked trends, while a linear regression-based predictive model estimated future flood conditions. Threshold values for flood warnings and high-risk alerts were defined using hydrological principles and environmental data [3]. Game theory concepts were employed to optimize flood management strategies by modeling interactions among stakeholders, including authorities, responders, and communities. A non-cooperative game theory approach simulated conflicting objectives, such as minimizing economic losses and optimizing resource allocation. Stable solutions were identified through the Nash equilibrium, ensuring no stakeholder could unilaterally improve outcomes. Visualization dashboards presented time-series data, risk levels, and stakeholder strategies, facilitating informed decision-making. Simulation results demonstrated the system's effectiveness in flood risk assessment. Water levels remained below the 2.5-meter warning threshold but rose significantly during simulated abnormal conditions. In later stages, some areas approached the 3.0-meter high-risk threshold, indicating zones vulnerable to flooding. Flood flow rates frequently exceeded the 40 m³/s threshold, with peaks above 60 m³/s, highlighting the need for continuous flow monitoring. Temperature fluctuations were minimal, consistently below the 25°C threshold, suggesting limited influence on flood risks during the study. However, vegetation cover often fell below the 30% threshold, correlating with increased flood risks and reinforcing its importance in mitigation. The system effectively categorized risk levels, with most instances classified as "Normal" or "Warning." High-risk alerts were concentrated during elevated water levels and flows. This research highlights the potential of digital twins for real-time flood monitoring and collaborative decision-making, providing a robust framework to enhance disaster resilience.

Keywords: Digital Twin; Flood Risk Assessment; Game Theory; Predictive Modeling; Multidimensional Data Analysis.

References

[1] Ghaith, M., Yosri, A., & El-Dakhakhni, W. (2021, May). Digital twin: a city-scale flood imitation framework. Canadian Society of Civil Engineering Annual Conference (pp. 577–588). Singapore: Springer Nature Singapore.

[2] Gheibi, M., & Moezzi, R. (2023). A Social-Based Decision Support System for Flood Damage Risk Reduction in European Smart Cities. Quanta Research, 1(2), 27–33.

[3] Kreps, D. M. (1989). Nash equilibrium. In Game Theory (pp. 167–177). London: Palgrave Macmillan UK.

How to cite: MohammadZadeh, F., Eghbalian, H., Gheibi, M. G., Yeganeh-Khaksar, R., Ghazikhani, A., and Behzadian, K.: A Digital Twin Framework for Real-Time Flood Monitoring and Multidimensional Prediction: A case study in Iran, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20713, https://doi.org/10.5194/egusphere-egu25-20713, 2025.

Asset-based Dynamic Flood Risk Assessment: Case Study of London Downtown

Flooding poses significant risks to urban centres, with particular challenges faced by business hubs where disruptions can have devastating consequences on national and global economies [1]. Business hubs are the lifeblood of national and global economies. During flood events, businesspeople encounter disruptions that not only obstruct daily operations but also ripple through supply chains and financial systems [2-3]. This study emphasises the importance of protecting critical assets in Downtown London, a vital business hub, to mitigate economic and social impacts during floods. Through a watershed-based approach, Downtown London, a vibrant business hub with numerous critical assets, has been selected as the case study area. The district contains key commercial buildings and infrastructure that are vital to economic and social continuity. Using Digimap and Verisk, essential commercial buildings and critical assets are pinpointed based on their usage and significance. These tools facilitate generating an accurate map of assets requiring priority attention during flood events.

The proposed decision support system (DSS) is developed to aid risk management authorities, including policy-makers, decision-makers, and technical staff. The system operates on two key bases. Real-time population density data for critical assets is obtained using Google API. This data helps evaluate the human vulnerability component during flood scenarios. A flood forecasting system is integrated to predict water levels at 15-minute intervals for the coming hours. This system provides granular and actionable insights into evolving flood conditions. For each critical asset, two risk values are computed: one based on population density and another on forecasted water levels. These values are combined to derive a dynamic risk level for each time step, enabling authorities to respond effectively. The integration of real-time data and predictive modeling in the DSS offers a comprehensive framework for flood risk assessment. By prioritising critical assets based on dynamic risk levels, authorities can implement targeted preparedness and response measures such as early warnings and evacuation plans. This approach ensures both human safety and economic resilience. The findings have demonstrated the feasibility of applying real-time data and cutting-edge modeling to enhance urban flood resilience. By combining flood risk maps, real-time population density, and a comprehensive prioritisation framework, this approach provides a promising tool for urban planners and emergency responders to protect critical business assets and ensure economic continuity during flood events.

References

[1] Bakhtiari, V., Piadeh, F., Behzadian, K. and Kapelan, Z. (2023). A critical review for the application of cutting-edge digital visualisation technologies for effective urban flood risk management. Sustainable Cities and Society, p.104958.

[2] Bakhtiari, V., Piadeh, F., Chen, A.S. and Behzadian, K. (2024). Stakeholder analysis in the application of cutting-edge digital visualisation technologies for urban flood risk management: A critical review. Expert Systems with Applications, 236, p.121426.

[3] Piadeh, F., Behzadian, K. and Alani, A.M. (2022). A critical review of real-time modelling of flood forecasting in urban drainage systems. Journal of Hydrology, 607, p.127476.

How to cite: Bakhtiari, V. and Piadeh, F.: Asset-based Dynamic Flood Risk Assessment: Case Study of London Downtown, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3504, https://doi.org/10.5194/egusphere-egu25-3504, 2025.

Social media applications have emerged as reliable communication channels, especially when traditional methods falter [1]. Their integration into emergency management presents significant advantages, including enhanced situational awareness during unfolding events, rapid dissemination of news and alerts to broader audiences, and improved coordination among decision-makers and stakeholders [2]. Both remote sensing and social media data offer distinct advantages in large-scale flood monitoring and near-real-time flood monitoring [3]. To better understand these advantages and challenges, a comprehensive review and analysis of the literature on the application of social media in this field was conducted.  Social media facilitates participatory and collaborative structures, enabling collective knowledge-building in public information and warning systems. To realise this vision, the authors examined, 73 studies conducted from 2014 to 2024 to systematically evaluate the current literature surrounding communication on social media and the latest research in social media informatics related to disasters. This review identified key challenges within existing studies. The articles included 23 related to pluvial floods, 12 related to fluvial floods, 17 related to storm floods and 21 paper that were unspecified The majority of the studies were conducted in China, followed by the United States. Various software platforms, including Twitter, YouTube, and other social media networks, were analysed. Data extraction from these platforms was performed using Python programming. The study periods ranged from 1 to 3,650 days. These findings serve as guidance for researchers examining the relationship between social media and disaster management. They aim to develop the use of social networks during disasters, analyse patterns, and create programming to identify best practices for utilising social media in times of crisis. In the future, a mapping framework and tool can be developed to automatically extract information from social media through text and image analysis. By integrating this data with other available information sources, it will be possible to generate more accurate inundation maps in real-time. It is essential to recognise that information about floods obtained from social media may be incomplete during communication interruptions. To address this issue, future research should prioritise integrating big data from urban Internet of Things networks and improving communication infrastructure repairs. By adopting this strategy, we can collect more comprehensive disaster information to enhance flood emergency response effectiveness.

References

[1] Piadeh, F., Behzadian, K. and Alani, A.M. (2022). A critical review of real-time modelling of flood forecasting in urban drainage systems. Journal of Hydrology, 607, p.127476.

[2] Piadeh, F., Ahmadi, M., Behzadian, K. (2020). A Novel Planning Policy Framework for the Recognition of Responsible Stakeholders in the of Industrial Wastewater Reuse Projects. Journal of Water Policy, 24 (9), pp. 1541–1558.

[3] Bakhtiari, V., Piadeh, F., Chen, A., Behzadian, K. (2024). Stakeholder Analysis in the Application of Cutting-Edge Digital Visualisation Technologies for Urban Flood Risk Management: A Critical Review. Expert Systems with Applications, p.121426.

How to cite: Panahi, A., Karkhaneh, N., and Piadeh, F.: Community-based flood early warning system: Current practice and Future directions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3526, https://doi.org/10.5194/egusphere-egu25-3526, 2025.

About 90% of extreme precipitation in the midlatitudes can be assoicated with front boundaries. Therefore, it is important to identifiy frontal locations for short term weather forecasting or long-term prediction of precipitation in climatology. Deep learning (DL) refers to machine learning alogrithms that use multiple layers of neural networks to derive features from input data. It is generaly useful to process 2-dimensional image data. One of the potential advantages of employing DL to detect surface weather fronts is that the developed DL based functions can be applied to automatic detection of surface weather fronts for climate models. However, justifications of its applicability on climate models are needed.

In this study, we developed deep learning based methodology to detect surface weather fronts. Specifically, a U-shape convolutional network (U-net) based deep learning model is developed to predict surface weather fronts over Japan and surrounding sea in summer (June, July, and August). We justify the applicability of the deep learning model in predicting surface fronts in summer on outputs from large-scale Global Climate Models (i.e. GCMs) from two aspects. First, the coarse resolution of GCMs (e.g., 1.25 degrees) can capture the general morphological features of surface fronts. Second, models trained in a colder climate are applied to predict fronts in a warmer climate with some decrease in predicted peak frequency of fronts, but the general features of the spatial distribution of fronts can be represented by the deep-learning model predictions. We also see that the locations of peak frequency tend to move slightly more southwesterly in a slant zone within the belt region between 25N to 40N as climate warms in the future.

How to cite: Mao, Y. and Yamada, T.: Applicability of deep learning based detection of surface weather fronts on large scale climate models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19530, https://doi.org/10.5194/egusphere-egu25-19530, 2025.

Flash floods are one of the most devastating natural disasters, posing significant risks to both human life and infrastructure. The classification of their underlying drivers—such as high precipitation, dam breaches, landslides, and melting snow—remains a critical yet challenging task, especially in regions like China, where diverse geographical and climatic factors exacerbate disaster complexity. In this study, we propose a Transformer-based Graph Network (TGN) designed to tackle these challenges by leveraging a dataset of over 53,000 flash flood events, each characterized by non-uniform geographical attributes and varying levels of data completeness. Unlike traditional graph neural networks (GNNs) that depend on predefined graph structures, TGN dynamically learns and refines edge weights during training, enabling it to uncover asymmetric dependencies. This adaptability is particularly valuable when explicit relationships between nodes are unavailable or incomplete.

Integrating multi-head self-attention mechanisms from Transformer architectures, TGN captures complex interdependencies across watershed features while maintaining interpretability through sparsity and diversity constraints. A distinguishing feature of this framework is its ability to identify meaningful graph structures without prior knowledge, offering insights into critical connections and interactions within disaster-prone regions. For instance, our experiments demonstrate how TGN emphasizes high-risk upstream-downstream relationships, providing actionable knowledge for localized flood management. The model significantly outperforms traditional GNNs and machine learning methods in accuracy and robustness, achieving superior classification performance across all four disaster categories. Furthermore, the TGN framework is supported by rigorous evaluation metrics, including Precision, Recall, F1-score, and Overall Accuracy, ensuring its reliability in real-world applications.

By combining innovative graph-based modeling with interpretable mechanisms, this study bridges the gap between theoretical advancements and practical disaster management. The proposed approach not only enhances prediction capabilities but also provides an analytical lens for understanding the intricate relationships among flash flood drivers, paving the way for more effective mitigation strategies and informed decision-making. This work underscores the transformative potential of adaptive graph neural networks in addressing complex environmental challenges and advancing the state of flood risk assessment.

How to cite: Wang, H., Xuan, Y., Zhang, Z., Antunes Meira, M., Li, Q., and Liu, C.: A Transformer-based Graph Network for Flash Flood Disaster Classification, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2646, https://doi.org/10.5194/egusphere-egu25-2646, 2025.

South Korea experiences significant regional variation in precipitation due to its unique topographical features. Over the years, the intensification of summer rainfall concentration has led to recurring damage from floods and torrential downpours. To mitigate such impacts, the Korea Meteorological Administration monitors precipitation using observed data from weather stations and estimated values for non-observed locations. The Generalized Extreme Value (GEV) distribution is commonly employed to model annual maximum precipitation, enabling the calculation of return levels that serve as foundational data for flood prevention. This study aims to estimate the spatially generalized additive GEV distribution of daily maximum precipitation using data from 54 Automatic Synoptic Observation System (ASOS) stations between 1972 and 2024. Spatial elements (latitude, longitude, altitude) and temporal elements (year) were incorporated into the model. The location, scale, and shape parameters of the GEV distribution were estimated using the maximum likelihood method, with smoothing functions accounting for spatial and temporal factors. The results indicate that the location and scale parameters, influenced by latitude and longitude, are relatively lower in central regions, while the shape parameter, influenced by altitude, shows similar trends. Furthermore, return levels for 50-year and 100-year return periods are notably higher in mountainous regions. Goodness-of-fit tests, such as the Anderson-Darling test, were performed on the GEV distributions of 53 ASOS stations, excluding one. However, 12 stations located in island regions, high-altitude areas, or regions affected by typhoons exhibited distributions that were difficult to explain spatially. These findings are expected to aid in the development of efficient water resource management strategies and regional flood prevention measures based on the distribution characteristics of precipitation.

How to cite: Lee, B., Hwang, Y., and Yoon, S.: Spatial and Temporal Extreme Modeling of Daily Maximum Precipitation Based on a Generalized Additive Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2663, https://doi.org/10.5194/egusphere-egu25-2663, 2025.

We investigate the potential of using Long Short-Term Memory (LSTM) neural networks for estimating streamflow in (sub)tropical catchments under contrasting hydroclimatic regimes (semi-arid and humid). We have used 176 Brazilian catchments with at least 30 years of streamflow data and LSTM models with 16 static catchment attributes as input features. We tested different LSTM model configurations to assess their sensitivity to varying input sequence lengths (lookbacks). The primary objective was to explore the hydrological insights offered by LSTM-based streamflow models and compare their performance with the traditional GR4J hydrological model. With this design, we aim to address two research questions: (i) Does the performance of LSTM models depend on catchments' hydroclimatic characteristics? (ii) How effective are LSTM-based models for streamflow simulation in tropical and subtropical catchments under semi-arid and humid conditions? We adopt two modeling approaches: (1) regionalized models trained on catchments within the same hydroclimatic regime and (2) a composite model trained on a heterogeneous sample combining both arid and humid catchments. The findings reveal distinct sensitivities of LSTM models to hydroclimatic conditions. LSTM models exhibit higher sensitivity to the length of input sequences (lookbacks) in humid catchments, with longer sequences yielding better performance. This is attributed to the dominant hydrological processes in humid regions, which are influenced by long-term memory effects such as soil moisture and groundwater storage. Conversely, this sensitivity is not observed in semi-arid catchments, where streamflow dynamics are primarily driven by short-term precipitation events and exhibit less dependence on long-term hydrological processes. Furthermore, the composite model, which combines semi-arid and humid catchments, demonstrates a decrease in performance for semi-arid catchments. This suggests that adding catchments with contrasting hydroclimatic characteristics introduces heterogeneity in the dataset, potentially reducing the model's ability to capture the specific dynamics of semi-arid catchments. Overall, the regionalized LSTM models outperformed the GR4J model in both semi-arid and humid regimes, particularly in humid catchments. Approximately 87% of humid catchments and 50% of semi-arid catchments achieved Kling-Gupta Efficiency (KGE) values above 0.60 during the testing phase of the regionalized LSTM models. These results highlight the potential of LSTM networks for streamflow regionalization, especially in humid regions where long-term hydrological memory plays a critical role. The study underscores the strengths and limitations of LSTM models in tropical and subtropical catchments with contrasting hydroclimatic regimes. The findings suggest that LSTM models could serve as valuable tools for regional hydrological applications, aiding local and regional decision-making processes. Additionally, the results emphasize the importance of tailoring LSTM model configurations to the specific hydrological characteristics of catchments, particularly the choice of input sequence length, to maximize model performance. 

How to cite: Maria de Andrade, J., Nóbrega, R., Ribeiro Neto, A., Rico-Ramirez, M., Coxon, G., and Montenegro, S.: Streamflow simulations using regionalized Long Short-Term Memory (LSTM) neural network models in contrasting climatic conditions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-580, https://doi.org/10.5194/egusphere-egu25-580, 2025.

Soil moisture dynamics play a crucial role in hydrological processes, influencing runoff generation, drought stress, and water management. To better understand the complex drivers of soil moisture dynamics, we present a novel hybrid architecture integrating Vision Transformers (ViT), spatial attention mechanisms, and Long Short-Term Memory (LSTM) networks. This architecture enables investigation of controlling factors across diverse landscapes in the Continental United States (CONUS) by incorporating spatial awareness at two levels: through ViT's ability to capture spatial patterns and through explicit spatial attention between neighboring stations. We leverage a comprehensive set of environmental data sources, including in-situ measurements from the International Soil Moisture Network (ISMN), ERA5 climate reanalysis, USGS elevation products, MODIS land cover, and SoilGrids soil characteristics. Initial results from a one-year training period and three-month testing period (R² = 0.73, 0.72, 0.73 for 24h, 48h, and 72h predictions) reveal important insights about the hierarchical importance of different drivers across prediction windows. Our preliminary analysis shows that static physical properties (particularly slope and soil structure) and hydraulic characteristics maintain high importance across temporal scales, while the influence of dynamic weather features varies with prediction horizon. The model's dual spatial attention mechanisms and temporal components enable discovery of both local and regional controls on soil moisture dynamics. The identified feature importance hierarchies provide initial insights into the spatiotemporal controls on soil moisture dynamics across CONUS. Ongoing work extends the training to the full temporal extent of available data to develop a more comprehensive understanding of these driving factors. This approach advances our fundamental understanding of soil moisture processes at continental scales, with implications for a future tool for land characterization and ecological site classification.

How to cite: Rahman, M., Meles, M., Bradford, S., and Nearing, G.: Drivers of Soil Moisture Dynamics over Continental United States, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7192, https://doi.org/10.5194/egusphere-egu25-7192, 2025.

This study proposes a novel approach for analyzing block minima data within the Generalized Extreme Value Distribution (GEVD) framework by incorporating the Negative Power Transformation (NPT). The NPT method, which includes a hyper-parameter to adjust data bounds (effectively reducing to the Reciprocal Transformation (RT) when the hyper-parameter is 1), aims to improve the accuracy and robustness of long-term return level (RL) estimations. Traditional transformation methods often exhibit limitations in accurately predicting RLs for extended return periods. Through extensive Monte Carlo simulations, we demonstrate that the NPT-GEVD method outperforms conventional approaches in terms of bias, standard error (SE), and root mean square error (RMSE) for return periods of 25, 50, and 100 years. Notably, the NPT-GEVD consistently provides reliable RL estimates across various parameterizations and sample sizes, particularly when using L-moments for estimation with smaller datasets. The efficacy of the NPT-GEVD method is further validated through its application to inter-arrival time (IAT) rainfall data from South Korea. The analysis revealed that RLs for detecting the time to exceed hourly cumulative rainfall thresholds of 60 mm, 90 mm, and 110 mm varied significantly across locations, ranging from 30 minutes to over 4 hours. This research underscores the significance of advanced transformation techniques in enhancing the accuracy and reliability of environmental risk assessments. The NPT-GEVD method offers valuable insights for improving flood prediction and mitigation strategies in the face of climate change.

How to cite: Yoon, S. and Prahadchai, T.:  Transformed Technique for Applying the Generalized Extreme Value Distribution to Block Minima, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2660, https://doi.org/10.5194/egusphere-egu25-2660, 2025.

Climate change and urbanization intensify urban pluvial flooding, posing significant threats to human lives and infrastructure. This situation underscores the critical need for efficient and accurate predictive systems for disaster prevention and mitigation. Traditional flood simulation models, while precise, are often limited by their data-intensive requirements and substantial computational complexity. In contrast, deep learning (DL) models show their advantages by high efficiency and powerful capability in processing large-scale non-linear data, making them highly appropriate for modeling complex flood dynamics. Consequently, integrating DL with conventional urban flood models has emerged as a promising strategy to enhance the accuracy and efficiency of flood prediction systems. However, existing research predominantly focuses on inland flooding, with limited attention to the role of tidal levels in coastal cities, which can significantly impact the accuracy of urban flood simulations.
To bridge the GAP, this study proposes an innovative hybrid DL approach that explores spatial and temporal data to improve the accuracy and efficiency of urban flood simulations, particularly in coastal areas. Simulation results from physics-based urban flood models are utilized to construct a comprehensive database for the DL model. Afterwards, patch-size and random sampling methods are employed to construct the sample dataset for training DL models. The convolutional neural network (CNN)-based data-driven urban pluvial flood model can simulate floods using topographic, rainfall, and tidal data, enabling the simulation of large urban areas within seconds. Incorporating diverse input data and advanced network architectures enhances model robustness and generalization across various scales and rainfall events. Fusion models that combine the strengths of DL and traditional hydrological models demonstrate improved prediction accuracy and computational efficiency by integrating tidal data and other environmental factors. Consequently, these hybrid models hold significant potential for integration into early warning systems and supporting decision-making processes in urban flood risk management.

How to cite: Zeng, B., Huang, G., and Yang, G.:  A Diversity Driven Deep Convolutional Network for Enhanced Coastal Urban Flood Risk Assessment, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2662, https://doi.org/10.5194/egusphere-egu25-2662, 2025.

Surface water contamination by pesticides is widespread across the European Union (European Environment Agency, 2024). A primary pathway for pesticide transfer from agricultural fields to surface waters is surface runoff (Wauchope et al., 1995 [https://doi.org/10.1162/neco.1997.9.8.1735]; Louchart et al., 2001 [https://doi.org/10.2134/jeq2001.303982x]; Reichenberger et al., 2007 [https://doi.org/10.1016/j.scitotenv.2007.04.046]). This process is influenced by various spatial and temporal factors, including compound properties, topography, application date and methods, climate, soil properties, and agricultural practices (Shipitalo and Owens, 2003 [https://doi.org/doi: 10.1021/es020870b]). Richards-based models are valuable for predicting the temporal variability of pesticide runoff (Métayer et al., 2024 [https://doi.org/10.1016/j.scitotenv.2023.167357]), especially in regions with high rainfall intensity variability, such as the Mediterranean. However, their operational application is constrained by substantial computational demands and extensive data requirements. Meta-modeling approaches provide a means to reduce the computational time of an initial physically-based model. Among these, the Long Short-Term Memory (LSTM; Hochreiter and Schmidhuber, 1997 [https://doi.org/10.1162/neco.1997.9.8.1735]) model has demonstrated high efficiency in replicating hydrological (Kratzert et al., 2018 [https://doi.org/10.5194/hess-22-6005-2018]) and hydrochemical time series (Pyo et al., 2023 [https://doi.org/10.1016/j.wroa.2023.100207]), making them a promising meta-modeling strategy for pesticide runoff models. This study aimed to develop and evaluate a meta-modeling approach using LSTM models for a Richards-based model to simulate hourly variations in water and pesticide runoff over an entire year while minimizing computation times. The proposed approach was applied to a field-scale pesticide runoff model implemented in the fully spatially distributed hydrological model MHYDAS-Pesticide 1.0 (Crevoisier et al., 2021 [https://hal.inrae.fr/hal-04090048v1]) that integrates Richards and convection-dispersion equations, the uniform mixing cell concept, and an overland flow routine. This represents a challenge for at least the following three reasons: i) the time series contains mainly zero values of runoff discharge, ii) the prediction of pesticide runoff requires an efficient prediction of water runoff and ii) the order of magnitude of the targeted non-zero values of runoff concentration varies by several orders of magnitude. The LSTM meta-model was trained and validated using 560,560 annual time series simulations generated by the initial physically-based model. The training dataset comprised 70% of the simulations, with the remaining 30% reserved for validation. The resulting meta-model accounted for meteorological conditions, compound properties, and pesticide application date and rate. It demonstrated high accuracy in simulating hourly runoff and pesticide concentrations, achieving significant reductions in computation time. However, challenges remain, such as improving the precision of runoff occurrence simulation and enhancing the meta-model's generalizability by incorporating additional static parameters as inputs.

The poster will focus on the methodology for the meta-model’s development and the results of its evaluation. The meta-model has been implemented within a fully spatially distributed physically-based hydrological model, MHYDAS Pesticide, to form an hybrid version.

How to cite: Métayer, G., Dagès, C., Voltz, M., and Bailly, J.-S.: Meta-modeling of a physically-based pesticide runoff model with a Long-Short term Memory approach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19160, https://doi.org/10.5194/egusphere-egu25-19160, 2025.