The wetlands and flood plains are among the freshwater bodies that play important role for the land-atmosphere interactions and hydrological processes. They provide important ecosystem services, among which biodiversity preservation, flood prevention and mitigation, biomass production, water purification, reduction of eutrophication, and carbon sequestration. Recently, the attention of researchers was attracted by the Danube River Basin which covers 801.463 km² with specific wetlands, flood plains, coastal wetlands and salt marshes in 19 countries.

The DaWetRest (Danube Wetlands and flood plains Restoration through systemic, community-engaged, and sustainable innovative actions) project performs research and innovation actions to improve the linkage between the Danube and its tributaries with the neighbouring wetlands. The activities include methodologies for freshwater monitoring to improve and control the freshwater ecosystems in line with the EU Mission “Restore our Ocean and Waters by 2030” and specifically the Lighthouse “Danube and the Black Sea”.

Centuries of anthropogenic activities with low attention to nature preservation caused the loss of more than 70% of wetlands, flood plains, river meandering, and coastal wetlands, such as salt marshes in the Danube River Basin. The restoration and protection of some the above listed types of areas is considered as one type of mitigation measures for control of greenhouse gases emissions among a number of other topics.

DaWetRest applies a holistic approach for restoration of deteriorated connections of Danube River and its tributaries with still existing wetlands. Active and passive approaches are used in order to  ensure the improvement of ecosystems using nature-based solutions providing sustainable social and economic activities for the local population.

How to cite: Batchvarova, E. and Koulov, B.: The role of wetlands in the land-atmosphere interactions and hydrological processes, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-642, https://doi.org/10.5194/ems2025-642, 2025.

Hydrological forecasts provide essential information to prevent flood and drought disasters, as well as to manage water, agriculture, and hydropower generation. Global forecast approaches are crucial in regions lacking hydrological in situ registers to develop and evaluate forecast systems. In this direction, in our work “Global Streamflow Seasonal Forecast by a Novel two-way AOGCM/Land/River Coupling” presented at the 2024 EMS annual meeting, we demonstrated the ability of the operational Météo-France seasonal prediction system (SYS8) to deliver skilful global streamflow forecasts. We also highlighted the impact of soil moisture initialisation on atmospheric forecasts, which is associated with land-atmosphere coupling.

Given the strong dependence of land-atmosphere water and energy exchanges on vegetated land cover, we explore the impact of Leaf-Area-Index (LAI) assimilation in the initialisation run on streamflow predictions. The control forecast of SYS8 derives land initial conditions from a historical run where the atmosphere/ocean is nudged to the ERA5/GLORYS re-analysis and climatological LAI (control initialisation). This study improves the initialisation run by taking the soil moisture and temperature from an offline land simulation where the LAI, derived from the THEIA satellite-based product, is assimilated (LDAS initialisation). Daily streamflow ensemble hindcasts of 25 members are generated, with a lead time of up to 4 months initialised on 1st of February/May/August/November between 1993 and 2017. Forecast performance is assessed against observed streamflow in flow-gauged basins worldwide and compared with the hindcasts from the control land initialisation. The performance metrics are the Kling-Gupta Efficiency (KGE) and its components (correlation, simulated-to-observed discharge mean and deviation ratios).

The LDAS initialisation run shows reduced discharge mean bias and increased correlation compared to the control SYS8 initialisation run, especially in Europe and South America. The LDAS initialisation generally leads to more accurate discharge forecasts for all seasons (spring-MAM, summer-JJA, autumn-SON and winter-DJF) in South American river basins. Similar conclusions apply to Europe, except for the spring season, where reduced performance
is observed in the Iberian Peninsula and basins in the Scandinavian Peninsula. North America only reports lower KGE for the winter forecast in basins on the western side, where the streamflow production is mainly driven by snow or drier conditions. Current results encourage ongoing efforts to enhance land initialisation through land data assimilation of other variables, particularly snow water equivalent and/or soil moisture.

How to cite: Narváez, G. and Ardilouze, C.: Global Hydrological Seasonal Prediction with an ESM: Impact of land data assimilation on streamflow forecast skill, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-58, https://doi.org/10.5194/ems2025-58, 2025.

Machine learning (ML) and deep learning (DL) models can play an important role when it comes to modelling complicated processes. Such capability is necessary for hydrological and climate-related applications. Generally, ML models utilize precipitation and temperature time series of a basin as input to develop a lumped rainfall-runoff model to simulate streamflow at the basin outlet. However, when it is divided into several sub-basins, Graph Neural Networks (GNN) can consider each sub-basin as a node and link them together using a connectivity matrix to account for spatial variations of hydroclimatic variables. In this study, GNN and various ML models with different types of architecture, ranging from neural networks, tree-based structure, and gradient boosting, were exploited for daily streamflow simulation over different case studies. For each case study, the basin was divided into a few sub-basins for which daily precipitation and temperature data were aggregated and used as input. For training GNN, the connection matrix of sub-basins was also used as input. Basically, 75% of historical records were utilized to train GNN and different ML models, e.g., artificial neural networks, support vector machine, decision tree, random forest, eXtreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LightGBM), and Category Boosting (CatBoost), while the rest was used for testing. Streamflow simulation was conducted with/without considering seasonality impact and lag times. The obtained results clearly demonstrate that considering seasonality and time lags can enhance accuracy of streamflow predictions based on Kling–Gupta efficiency (KGE). Furthermore, GNN with seasonality impact and time lags achieved promising results across different case studies with KGE>0.85 for training and KGE>0.59 for testing data, respectively. Among ML models, boosting models, e.g., LightGBM and XGBoost, performed slightly better than other ML models. for Finally, this comparative analysis provides valuable insights for ML/DL applications in climate change impact assessments.

Acknowledgements: This research work was carried out as part of the TRANSCEND project with funding received from the European Union Horizon Europe Research and Innovation Programme under Grant Agreement No. 10108411.

How to cite: Niazkar, M., Mozzi, G., Pal, J., and Mysiak, J.: Hydrological modelling using machine and deep learning models across multiple case studies, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-562, https://doi.org/10.5194/ems2025-562, 2025.

This study aims to improve anomaly detection in urban flooding sensor data by utilizing the Long Short-Term Memory (LSTM)-Autoencoder technique. Flooding problems caused by climate change and rapid urbanization are becoming increasingly frequent in densely populated urban areas, making it crucial to build robust real-time monitoring and rapid response systems. Traditional flood monitoring systems often rely on static thresholds or short-term pattern analysis, which limits their ability to detect complex and subtle anomalies that can develop in sensor data over extended periods of time. To address these limitations, we developed an LSTM-Autoencoder-based anomaly detection model capable of identifying abnormal signals from inundation sensors in real time. The LSTM component excels at processing time-series data with temporal dependencies, while the Autoencoder is characterized by its ability to extract and reconstruct meaningful features of the input data. In this study, we used measurement data collected under both ideal experimental conditions and artificially generated abnormal conditions as training inputs. Data with reconstruction errors exceeding a predefined threshold were classified as anomalies or outliers. The experimental results demonstrated that the proposed model significantly improved inundation prediction accuracy compared to conventional methods, and it consistently maintained high sensitivity under a wide range of environmental changes and unpredictable anomaly scenarios. The LSTM-Autoencoder model developed in this research effectively captures temporal dynamics and variations in sensor data, thereby enhancing the reliability of urban flood monitoring systems. Ultimately, this approach is expected to contribute to more accurate urban flood forecasting and play a key role in advancing smart urban flood management systems.

How to cite: An, S. W., Lee, B. H., and Kim, B. S.: A Study on Abnormal Detection in Urban Inundation Sensor Using LSTM-Autoencoder Techniques, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-106, https://doi.org/10.5194/ems2025-106, 2025.

Strategic ecosystem restoration is central to achieving both climate resilience and biodiversity goals. The EU Biodiversity Strategy for 2030 calls for converting at least 10% of agricultural land into high-diversity landscape features. However, the hydrological implications of such vegetation-based transformations remain underexplored. At the same time, effective water policy requires actionable insights into how land-use interventions interact with a changing climate. Vegetation plays a pivotal role in hydro-meteorological extremes, influencing evapotranspiration, soil moisture dynamics, infiltration, and runoff generation. Here, we use a kilometre-scale hydrological model integrated with a machine learning optimisation algorithm to evaluate the impacts of afforestation on European water systems under both current and +2°C warming conditions. Three afforestation scenarios — ranging from a hypothetical full land conversion to smart, spatially targeted, biodiversity-aligned strategies — were tested for their effects on evapotranspiration, river discharge, and groundwater levels. Results reveal that smart afforestation reduces seasonal flood peaks by up to 15%, with the strongest effects observed in Central Europe during winter. It also improves groundwater resilience by tripling minimum storage during dry periods, despite a modest 5% decline in average levels. While climate warming results in a 16% reduction in water availability, afforestation leads to a smaller, consistent 6% decrease that remains stable across climate scenarios. Zones with 40–50% forest cover offer the most effective balance, maximising flood mitigation and buffering hydrological drought risk. These findings emphasise the importance of nature-based solutions, such as spatially targeted afforestation, as viable strategies to moderate hydro-meteorological extremes and enhance climate-adaptive water management.

How to cite: El Garroussi, S., Wetterhall, F., Barnard, C., Di Giuseppe, F., and Mazzetti, C.: Balancing flood mitigation and water availability through smart afforestation, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-157, https://doi.org/10.5194/ems2025-157, 2025.

In recent years, deep learning models have gained traction in hydrology, particularly in streamflow modeling, which is a prerequisite for accurate riverine flood forecasts. However, current state-of-the-art streamflow models are generally lumped setups that ingest meteorological information only as spatial averages over the upstream area. Unfortunately, this withholds information that is relevant to precisely predict floods: for example, rainfall far upstream will take much longer to arrive at the outlet than rainfall further downstream.

Classical hydrologic models use distributed or semi-distributed setups to solve this problem: they divide the basin into pixels or subpolygons and route streamflow along the river graph. There are first attempts to translate this semi-distributed modeling paradigm to end-to-end deep learning models, but so far they are typically trained only on individual river networks (e.g., Kratzert et al., 2021), lag behind the performance of lumped models (e.g., Kirschstein et al., 2021), or cannot generalize to unseen river networks (e.g., Vischer et al., 2025).

With the learnings and experience from operating lumped LSTM models at a global scale (Nearing et al., 2024), we revisit semi-distributed modeling with deep learning, however, at a much larger scale than previous efforts. While the core idea remains the same—i.e., processing time series with Long Short-Term Memory networks (LSTMs) and performing routing through Graph Neural Networks—many details have changed since 2021.

In this submission, we therefore present our version of a global end-to-end semi-distributed hydrologic model. We detail the model setup, its training procedure, and compare this model to the lumped setup. Our evaluation shows that the semi-distributed model has strong performance especially for large, ungauged rivers. Finally, we highlight how this modeling approach is a step towards a broader multi-output system that provides more information than just streamflow.

References:

• Kirschstein, Nikolas, et al. "The Merit of River Network Topology for Neural Flood Forecasting." Forty-first International Conference on Machine Learning. 2024.
• Kratzert, Frederik, et al. "Large-scale river network modeling using graph neural networks." EGU General Assembly Conference Abstracts. 2021.
• Nearing, Grey, et al. "Global prediction of extreme floods in ungauged watersheds." Nature 627.8004 (2024): 559-563.
• Vischer, Marc Aurel, et al. "Spatially Resolved Rainfall Streamflow Modeling in Central Europe." EGUsphere 2025 (2025): 1-26.

How to cite: Gauch, M., Kratzert, F., Metzger, A., Shenzis, S., Klotz, D., Cohen, D., and Gilon, O.: Semi-Distributed Hydrological Modeling Based on Deep Learning at Scale, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-31, https://doi.org/10.5194/ems2025-31, 2025.

In the German administrative setup, meteorology and hydrology are separate. The task to ensure and improve the communication and exchange between the two instances is embedded in the "Co-Design Project" run between the German meteorological service (DWD) and the German flood forecasting authorities. One of the activities in the project focusses on the verification aspect in relation to the use of the flood forecasting centers. 
Here, we present our ongoing work that follows a parallel approach to 
(a)    allow the user, i. e. the flood forecasters to review in retrospect the precipitation forecasts for a specific event, and 
(b)    derive general recommendations for the users concerning the precipitation in different weather models for their application in hydrological models.  
The retrospective analysis is embedded in a web application that can be accessed by the flood forecasters. The event is defined by the user by the choice a catchment area, date and time and a duration. The result is displayed in a predictability plot showing forecasted areal precipitation from different models (deterministic and ensemble) and lead times in combination with the observed precipitation and meteorological return periods from extreme value analysis. 
Routine verification methods applied to the specifics of the hydrological interests will be the base for the recommendations. Here, a focus is on weather model characteristics such as lead time and resolution in combination with weather conditions and the sizes of the catchments of interest, so that especially spatial uncertainty will be addressed. The considered weather models concentrate on the ICON model chain (global, EU, D2, RUC). Also, the added value of ensemble prediction systems is considered. 

How to cite: Blumenstein-Weingartz, I., Bondy, J., Fundel, F., Fundel, V., and Keller, J.: Evaluation of precipitation forecast data as input for hydrological models, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-680, https://doi.org/10.5194/ems2025-680, 2025.

Despite the effectiveness of flood early warning systems as mitigation tools, significant losses persist even in developed countries. Most operational systems provide discharge or water level forecasts at gauge locations, but their predictive accuracy remains limited. Recent advancements in earth observation and data collection technologies offer a wealth of high-resolution hydrological datasets, creating an unprecedented opportunity to apply Artificial Intelligence (AI) algorithms for improved monitoring and prediction of hydrological processes. While Long Short-Term Memory (LSTM) networks are widely used for discharge prediction due to their ability to model temporal dependencies in hydrological time series, they often overlook spatial dependencies within river networks.

This study explores whether Graph Neural Networks (GNNs), which represent river networks as graphs to capture spatial relationships, can improve discharge predictions when integrated with LSTM models. Leveraging the LamaH-CE dataset for the Danube Basin (1981–2017), we incorporate dynamic variables such as daily precipitation, temperature, and soil moisture, alongside 59 static catchment attributes, including digital elevation models, river density, basin area, and soil properties, and etc. We evaluate two approaches: (1) local discharge prediction at sub-basin scales using LSTM models, and (2) a hybrid LSTM-GNN framework where GNNs model water routing across the river network, accounting for upstream-downstream connectivity.

Our findings reveal that the hybrid LSTM-GNN approach significantly outperforms the standalone LSTM model, particularly in capturing spatial propagation of flows during high-discharge events. By explicitly modelling routing dynamics, GNNs enhance prediction accuracy in complex river systems, addressing a critical gap in current forecasting methods. These results underscore the value of integrating spatial context into hydrological modelling and highlight the transformative potential of graph-based deep learning for flood prediction. This framework offers a pathway to strengthen flood early warning systems, supporting more effective mitigation strategies.

How to cite: Mosaffa, H., Prudhomme, C., Chantry, M., Rüdiger, C., Pappenberger, F., and Cloke, H.: Graph Neural Networks in Hydrology: Improving River Discharge Forecasts for Flood Early Warning Systems, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-583, https://doi.org/10.5194/ems2025-583, 2025.