This session aims to provide an active forum in which to demonstrate and discuss the integration and appropriate application of emergent techniques and technologies in water-related contexts.
Topics addressed in the session include:
* Predictive and exploratory models based on the methods of statistics, computational intelligence, machine learning and data science: neural networks, fuzzy systems, genetic programming, cellular automata, chaos theory, etc.
* Methods for analysing Big Data and complex datasets (remote sensing, IoT, earth system models, climate models): principal and independent component analysis, time series analysis, clustering, information theory, etc.
* Optimisation methods associated with heuristic search procedures (various types of genetic and evolutionary algorithms, randomised and adaptive search, etc.) and their application to hydrology and water resources systems
* Multi-model approaches and hybrid modelling approaches that blend process-based (mechanistic) and data-driven (machine learning) models
* Data assimilation, model reduction in integrated modelling, and High-Performance Computing (HPC) in water modelling
* Novel methods for analysing and quantifying model uncertainty and sensitivity
* Smart water data models and software architectures for linking different types of models and data sources
* IoT and Smart Water Management solutions
* Digital Twins for hydrology and water resources
Applications could belong to any area of hydrology or water resources, such as rainfall-runoff modelling, hydrometeorological forecasting, sedimentation modelling, analysis of meteorological and hydrologic datasets, linkages between numerical weather prediction and hydrologic models, model calibration, model uncertainty, optimisation of water resources, smart water management.
Data-driven rainfall–runoff models often deliver high predictive skill but provide limited insight into hydrological processes. Classic conceptual models, by contrast, are transparent and process-based but rely on a limited collection of empirically designed structures, so choosing and adapting an appropriate model across diverse catchments remains difficult. To bridge this gap, this study explores a hydrologically constrained genetic-programming (GP) framework that automatically discovers basin-specific conceptual model structures from a shared library of hydrological building blocks. Model structures are assembled from modular storages, flux functions, and routing components adapted from the Modular Assessment of Rainfall–Runoff Models Toolbox (MARRMoT) [1], which assembles and standardizes the storage and flux formulations of 47 established conceptual models. GP, an evolutionary algorithm, is then used to operate on structural flags and parameter values, selecting and combining these components into explicit model equations. Each candidate’s reservoir system is then assembled automatically and advanced with a mass-conserving implicit time-stepping scheme. Calibration uses a multi-objective NSGA-II algorithm, so structural choices and parameters are explored within a single optimization loop.
The framework is evaluated on CAMELS-US basins through three experiments. In a snow-dominated mountain catchment (Buffalo Fork, 13011900), the discovered structure reproduces the snowmelt-driven regime and flow-duration curve in the test period with high efficiency (held-out test period NSE ≈ 0.85). Uncertainty analyses indicate that a snow–soil–single-routing backbone is consistently retained. A transfer experiment to a hydrologically similar basin (Johnson Creek, 13313000) shows that directly reusing the Buffalo Fork structure and parameters already yields useful skill (NSEtest ≈ 0.72), while a short “hot-start” GP run seeded with this transferred solution can reach NSEtest ≈ 0.84, capturing most of the benefit of a much longer optimization (~40× fewer generations, at a small fraction of the computational cost). To evaluate the framework in a broader hydro-climatic context, it is benchmarked against the conceptual and LSTM rainfall–runoff models from the CAMELS benchmark study by Kratzert et al. [2]. We use 18 representative CAMELS-US basins (one medoid per HUC-2 region), asking the system to self-evolve a distinct model structure tuned to the hydro-climate of each basin from the same shared component library. In this multi-basin setting, the GP-derived models achieve a median NSEtest of about 0.70, generally match or exceed the conceptual benchmarks, and remain competitive with the LSTM variants. The results indicate that hydrologically constrained automated model discovery can help narrow the accuracy-interpretability trade-off, yielding transparent, physically consistent rainfall-runoff models and suggesting a potential path toward structure transfer in data-sparse or ungauged basins.
[1] L. Trotter, W. J. M. Knoben, K. J. A. Fowler, M. Saft, and M. C. Peel, “Modular Assessment of Rainfall–Runoff Models Toolbox (MARRMoT) v2.1: an object-oriented implementation of 47 established hydrological models for improved speed and readability,” Geosci. Model Dev., vol. 15, pp. 6359-6369, 2022.
[2] F. Kratzert, D. Klotz, G. Shalev, G. Klambauer, S. Hochreiter, and G. Nearing, “Towards learning universal, regional, and local hydrological behaviors via machine learning applied to large-sample datasets,” Hydrol. Earth Syst. Sci., vol. 23, pp. 5089-5110, 2019.
How to cite: Matin, N., Herath Mudiyanselage, V. V. H., Saha, A., Marshall, L., and Babovic, V.: Hydrologically constrained genetic programming for interpretable rainfall–runoff model discovery, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8757, https://doi.org/10.5194/egusphere-egu26-8757, 2026.
Although numerical representations of river networks are fundamental to hydrological modelling and analysis, their performant and flexible use remains challenging due to inherent spatial dependencies and graph-based structure. Many existing tools are constrained by limited computational efficiency and a lack of support for diverse river network formats. This makes it difficult to compare and analyse data and model outputs from multiple sources.
To address these limitations, we present earthkit-hydro, the hydrological component of ECMWF’s earthkit software for Earth system science workflows. Earthkit-hydro provides a unified interface for operations on river networks, including accumulations, catchment-level statistics, catchment delineation, distance calculations, and computing topological properties. The library supports a wide range of river network formats, including bifurcating river networks, and integrates with major Python array libraries such as NumPy, Xarray, PyTorch, and JAX. In addition, earthkit-hydro is well suited for machine-learning applications, offering GPU support and differentiable operations.
We also present an application to AIFL, ECMWF’s global machine-learning model for streamflow prediction. In this context, earthkit-hydro provides an efficient way of processing ECMWF’s meteorological forecasts by transforming meteorological variables into the catchment-based metrics required as input by AIFL.
How to cite: Morrison, O. M. and Carton de Wiart, C.: Introducing earthkit-hydro: an efficient graph-based library for scalable hydrological modelling and analysis, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-11233, https://doi.org/10.5194/egusphere-egu26-11233, 2026.
The stage-discharge rating curve is crucial for flow estimation in open channels. The power relationship between discharge (Q) and stage (h) is used conventionally to evaluate the discharge from stage measurements. However, this relationship performs poorly under unsteady conditions, varying bed roughness and alteration of cross-sectional geometry. Hysteretic behavior is often found in the rating curves showing different discharges under identical stages because of the unsteadiness. Attempts have been made by the scientific community to develop such Q-h relationship that can be able to capture this hysteresis along with easy computations. Jones' formula comprises steady state discharge and temporal gradient of h is one of the equations that have been used for modeling this hysteresis. Symbolic regression (SR) has also been applied to trained machine learning (ML) models to derive site-specific explicit mathematical Q-h equation of high accuracy. However, the SR-based relationship does not exhibit the realistic hysteretic nature of rating curves. This study aims to find a robust stage-discharge relationship that shall capture the realistic hysteretic nature while having high accuracy. To achieve this, a Physics-Informed Neural Network (PINN) is developed incorporating the Jones formula into its loss function along with the data-driven error term and a term to calibrate the parameters of the Jones formula. Further, SR is implemented using the PySR module to derive a mathematical equation that fits the prediction of the PINN. This equation has no differential term and incorporates the stage on time t, stage on time (t-1) and steady state discharge. For the Q-h data with 15-minute temporal resolution of the River Brays of the USA, the rating curves derived from the Jones formula and this PINN-SR are compared based on their abilities to capture the hysteretic nature of the Q-h relationship. Four metrics of hysteresis capturing performance and an overall score are used for comparison. All data are normalized to avoid mixed units in the overall score. The hysteresis area error to check the magnitude is found to be 1.574 for Jones formula and 0.129 for PINN-SR. For fitting accuracy, the average of Root Mean Square Errors (RMSEs) for rising and falling limbs are 0.425 and 0.360, the hysteresis width errors are 0.007 and 0.138, and the Direction-Aware Dynamic Time Warpings (DTWs) are 18.225 and 5.752. The overall error scores for hysteresis are 5.058 and 1.595 for the Jones formula and PINN-SR-based rating curves, respectively. These results indicate the superior performance of PINN-SR-based rating curve over the Jones formula in capturing the hysteresis under unsteady flow conditions.
How to cite: Dasgupta, R., Das, S., Banerjee, G., and Mazumdar, A.: Modeling hysteresis in stage-discharge: Physics and Artificial Intelligence based Approach, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-981, https://doi.org/10.5194/egusphere-egu26-981, 2026.
Climate-driven alterations of river flow regimes are increasing the occurrence of hydrological droughts and low-flow conditions in many humid watersheds, raising new challenges for basin-scale water management. In the Bécancour River basin (Québec, Canada), these pressures coincide with the expansion of cranberry production, a water-intensive agricultural activity supported by irrigation and reservoir-based water storage. Based on an initial watershed-scale assessment of individual and cumulative agricultural, municipal, and industrial water withdrawals, the first step of this research identifies agricultural water use—and particularly cranberry production—as a water use type that becomes especially influential under low-flow conditions. This highlights the need for a detailed, daily representation of cranberry farm-level water demand and reservoir operations, which cannot be adequately captured by conventional hydrological models alone. Building on this foundation, a Cranberry Farm Water Management Model is developed to explicitly simulate daily water use, storage, and recirculation processes. The model is coupled with HYDROTEL, a distributed hydrological model, allowing direct assessment of the impacts of cranberry farming practices on streamflow dynamics of the hydrographic network of the watershed. The outputs of this coupled framework then will serve as a basis for integration with a socio-economic model, enabling the analysis of farmer behavior, governance regulations, and feedback between water availability and management decisions. Together, this integrated socio-hydrological model provides a structured platform for evaluating future climate change scenarios and exploring mitigation and adaptation strategies for sustainable water governance in the Bécancour River basin. Accordingly, this communication focuses on the development and structure of the Cranberry Farm Water Management Model, as the central building block of this broader socio-hydrological governance framework.
How to cite: Rousseau, A. N., Khoramshokooh, N., and Gumiere, S. J.: Socio-Hydrological Governance for Watershed-Scale Water Management:Evaluating the Influence of Cranberry Production on Water Availability for Various Agricultural, Municipal and Industrial Water Uses, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-22182, https://doi.org/10.5194/egusphere-egu26-22182, 2026.
Accurate streamflow prediction is essential for reliable water resource management and flood forecasting. Recently, deep learning methods, especially Long Short-Term Memory (LSTM), have demonstrated state-of-the-art performance for streamflow prediction when trained in supervised learning (SL) settings. However, robust SL requires large volumes of “labeled” training data, including meteorological inputs paired with corresponding streamflow observations as ground truth. Globally, this poses a problem as only a small fraction of catchments worldwide are monitored with stream gauges. This leaves most regions with abundant “unlabeled” meteorological data but limited 'labels', i.e. discharge observations. This data scarcity limits SL model performance in data-scarce regions, and also limits model generalization and transferability.
To overcome this challenge, we propose a two-stage semi-supervised learning (SSL) method for streamflow prediction based on the Contrastive Predictive Coding (CPC) approach [1]. CPC is a self-supervised learning method that extracts informative feature representations from sequential data (e.g., meteorological time series) without labeled targets (e.g., streamflow observations), by contrasting correct future predictions against incorrect ones. In the first stage, we use CPC to pre-train an encoder (i.e., fully connected layers) and an LSTM network followed by a projection head (i.e., a linear layer without bias), using a large amount of meteorological data (28 years). In the second stage, we add a linear layer to the pre-trained encoder and LSTM, and fine-tune the model for streamflow prediction using a small amount of meteorological data paired with streamflow observations (1 year).
We demonstrate the effectiveness and robustness of our methodology on the CAMELS-DE dataset [2]. We conduct a thorough comparison with a baseline supervised learning model with the same LSTM network. The results suggest that our method improves both in-sample and out-of-sample generalization performances over the SL method, when only a limited amount of discharge data is available. Additionally, the results demonstrate that transfer learning via CPC pre-training provides informative representations for streamflow prediction task, enabling faster convergence and higher model training efficiency, compared to the baseline model trained from scratch.
Our findings highlight a promising direction to leverage self-supervising learning methods for developing hydrological foundation models. Foundation models have revolutionized artificial intelligence applications across diverse domains, and hold large promise for hydrological applications. By scaling our proposed approach with larger and more diverse datasets, we can make significant strides towards multiple downstream prediction tasks, including predicting climate-driven variables (e.g., discharge, groundwater, and soil moisture).
References:
[1] Oord, A. V. D., Li, Y., & Vinyals, O. (2018). Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748.
[2] Loritz, R., Dolich, A., Acuña Espinoza, E., Ebeling, P., Guse, B., Götte, J., ... & Tarasova, L. (2024). CAMELS-DE: hydro-meteorological time series and attributes for 1555 catchments in Germany. Earth System Science Data Discussions, 2024, 1-30.
How to cite: Jia, T., Chen, G., and Ehret, U.: Semi-Supervised Deep Learning for Streamflow Prediction in Data-scarce Regions, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8108, https://doi.org/10.5194/egusphere-egu26-8108, 2026.
Over the past few years, the of use machine learning (ML) models in hydrology has shifted towards deep learning (DL), perhaps because of the icreasing availability of big data and ready-to-be-deployed software tools. It seems that the underlying assumption is that more complex (DL) models are desired, however, there are less efforts to systematically investigate and validate this perception. Deep learning models like LSTM have greatly improved sequential data forecasting, but their success depends on large labeled datasets, limiting their effectiveness in data-scarce domains. This study investigates whether complex ML models offer significant advantages over simpler approaches in predicting seasonal streamflow in Canada. Using a comprehensive case study, we examine multiple subbasin types—mountain, prairie, and non-prairie—along with headwater and downstream locations, exhibiting both natural and human-influenced regulated flow regimes. These variations introduce distinct hydrological behaviors, making them an ideal testbed for assessing model complexity requirements. Our case study includes 135 subbasins from the Canadian Nelson-Churchill River Basin, comprising the vast area starting from the Rocky mountains up to the Hudson Bay, with the monthly temporal resolution and spatial scales of the order of 200 km2 to ~1.0 x106 km2, as reflected by drainage areas of all subbasins.
We implemented a suite of ML techniques, ranging from traditional algorithms to advanced DL architectures. Specifically, we compared models developed based on Artificial Neural Networks (ANNs), Random Forests (RF), Long Short-Term Memory (LSTM) networks, attention-based LSTM networks, and stacked LSTM configurations. Each model was trained and tested using historical flow data across multiple subbasins, with performance evaluated through metrics, such as Nash-Sutcliffe Efficiency and Percent Mean Bias Error. We also experimented with alternative sets of input features, i.e., (i) all potential hydrometeorological inputs, (ii) correlation and partial mutual information-based inputs, and (iii) causality-based inputs.
Our findings reveal that while simpler models like RF and ANNs perform adequately in certain contexts—particularly in headwater subbasins with natural flow regimes—complex architectures, such as LSTM and stacked LSTM configurations demonstrate superior performance for downstream and regulated basins, where flow patterns exhibit higher variability and nonlinearity. In contrast, attention-based LSTM networks do not appear to outperform other options across certain basins. Interestingly, the benefits of DL models are not uniform across all subbasin types; prairie and non-prairie basins show mixed results, suggesting that model complexity should be tailored to basin characteristics rather than universally applied. These results highlight the importance of context-driven model selection to inform operational forecasting. Thus, water managers can leverage simpler models in less complex basins to reduce computational costs and data requirements, while reserving advanced architectures for highly regulated or downstream basins where accuracy gains justify the added complexity. This approach can potentially optimize resource allocation, enhance forecast reliability, and support informed decision-making in water allocations, reservoir operations, and drought/flood preparedness. Additionally, in the absence of data and computational constraints, multi-model outputs can be synthesized further through fusion modelling techniques to enhnace overall prediction accuracy.
How to cite: Elshorbagy, A., Nguyen, D.-H., Khaliq, M. N., Unduche, F., and Akhtar, M. K.: Do We Need Deep Learning Models? Assessing the Complexity of Machine Learning Models for Seasonal Streamflow Forecasting Across Diverse Subbasins, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8281, https://doi.org/10.5194/egusphere-egu26-8281, 2026.
Bangladesh is disaster-prone due to its location, heavy monsoon rainfall, and frequent cyclones, with floods causing major loss of life and property. Accurate and timely flood forecasting and warning systems are essential to reduce flood-related damage and human suffering. Currently, the national flood forecasting system provides reasonably accurate predictions only for short lead times of up to three days (FFWC, 2021). Improving medium to long range flood forecasting with lead times of 5–10 days is therefore critical for enhanced flood preparedness.
This study investigates artificial intelligence-based approaches for medium-range river flow forecasting with a five-day lead time at Hardinge Bridge station in the Ganges basin. Multiple input variables, including precipitation, precipitable water, soil moisture storage, and satellite-derived river water levels, were used. Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting Machine (GBM) algorithms were applied to simulate river water level as an alternative to traditional hydrologic model-based forecasting. Predictions were evaluated against the Bangladesh Water Development Board’s Flood Forecasting and Warning Centre (FFWC).
For each algorithm, 70% of data were used for training, 15% for testing, and 15% for independent validation. Various input combinations, or model scenarios, were examined. The scenario including all variables performed best. Among algorithms, Random Forest showed superior performance, with RMSE of 0.28 m, a coefficient of determination (R²) of 0.99, and a Nash Sutcliffe Efficiency (NSE) of 0.99. Upon evaluating the R² value by comparison in a percentage scale, it was observed that best RF model of scenario-01 demonstrated an improvement of approximately 38% over FFWC's Prediction of water level.
This research establishes that the machine learning algorithms, particularly RF, offers a promising alternative to traditional flood forecasting methods, with significant accuracy in predicting water level at Hardinge bridge station in the Ganges basin. Its capacity to use satellite-derived data improves flood forecasting and leads to more reliable predictions, potentially improving flood preparedness and risk management in Bangladesh.
How to cite: Mahmud, Md. A. and Hossain, Md. B.: Five Days Lead Time Water Level Forecasting in the Ganges–Padma River Using Satellite and Reanalysis Data with Machine Learning, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9860, https://doi.org/10.5194/egusphere-egu26-9860, 2026.
Accurate streamflow prediction remains challenging in monsoon-dominated basins characterized by extreme flow variability. This study evaluated three machine learning approaches for daily streamflow forecasting using 39 years of data (1980-2018) of the Basantpur station, Mahanadi basin, India from the CAMELS-India dataset which are (1) Optuna- optimized Genetic Programming (GP) for interpretable symbolic regression, (2) Optuna- optimized bidirectional LSTM networks, and (3) a novel GP-LSTM meta-learning framework that predicts optimal hyperparameters from time series statistical features.
Analysis of highly skewed flow distributions (97.66% of values <5,000 m³/s) using the False Nearest Neighbor method identified six-day embedding dimensions. For regular flow conditions without extreme outliers, the optimized LSTM achieved superior performance (NSE = 0.92, KGE = 0.93, R² = 0.92) compared to GP (NSE = 0.86, KGE = 0.87, R² = 0.86). However, GP demonstrated lower absolute errors (RMSE = 197.68 vs. 210.46 m³/s) and produced interpretable mathematical expressions that revealed lag-dependent hydrological relationships.
The meta-learning framework showed the best results when tested on complete datasets, including those with extreme events. By extracting thirty-two statistical features that cover central tendency, time-based autocorrelation, complexity measures, and spectral properties, the GP- based meta-model learns to predict the best LSTM configurations for different flow patterns. This flexible approach performed better on test data with outliers, showing improved predictions for rare but important flood events.
The results suggest that standard deep learning is effective in normal conditions. However, meta-learning frameworks, which adjust model structure based on flow characteristics, provide better reliability for operational flood forecasting in complex monsoon-influenced areas. This proposed hybrid meta-learning framework aims to combine the strengths of both methods. Our initial implementation, though, reveals challenges that need more effort.
How to cite: Singh, D. and Jothiprakash, V.: Streamflow Forecasting using Genetic Programming, LSTM, and Hybrid Meta-Learning GP-LSTM model in Monsoon-Dominated Basins, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16415, https://doi.org/10.5194/egusphere-egu26-16415, 2026.
The accuracy and robustness of flood forecasting have long been constrained by model structural uncertainties and runoff generation mechanisms. Single hydrological models or machine learning approaches not only show limited performance improvements but also struggle to achieve balanced simulation of both high-flow and low-flow processes. To address this, this study proposes for the first time a stacking ensemble machine learning framework (TBC-SEML) that integrates multi-model state awareness, terminal bias correction, and interpretability analysis. The framework leverages classical hydrological models (GR4J, HYMOD, SIMHYD) to acquire multi-model state datasets, establishing comprehensive evaluation metrics (NPCEM) as the optimization objective to enhance capture of high-flow processes. Furthermore, this study innovatively proposes a terminal bias correction based on Auto-Regressive with Extra Inputs and Weighted Least Square (ARX-WSL), and the excessive dominance of flood peak on weight estimation is suppressed by the flow attenuation coefficient β. Building on this, eight types of base learners are integrated, including Random Forest (RF), ExtraTrees, XGBoost, LightGBM and CatBoost, Multilayer Perceptron (MLP), Support Vector Regression (SVR), and K-Nearest Neighbors (KNN). Bayesian methods are used to optimize the hyperparameters of the base learners, and a meta-learner is constructed based on linear regression. Meanwhile, the SHAP interpretability analysis method is introduced to quantify the predictive contributions of base learners and state variables, enhancing model transparency. This highly diverse and heterogeneous stacking ensemble framework not only enhances the complementarity among base learners but also achieves good synergy between accuracy, stability, and interpretability, providing a new paradigm for intelligent hydrological forecasting that combines high performance and transparent decision support.
How to cite: Chang, X., Guo, J., Jia, T., Qin, H., and Liu, Y.: A stacking ensemble machine learning framework with terminal bias correction for flood prediction, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8326, https://doi.org/10.5194/egusphere-egu26-8326, 2026.
Upscaling coarse-grid flood maps to achieve fine-grid accuracy using machine learning has emerged as a promising hybrid pathway for operational flood mapping, as fine-grid hydrodynamic models remain computationally prohibitive for real-time and ensemble-based applications. In this context, latent diffusion models (LDMs), a class of generative AI models, have recently demonstrated superior accuracy and generalisability in flood map super-resolution. However, despite their inherently stochastic nature, it remains unclear to what extent LDM-generated ensembles provide meaningful representations of predictive uncertainty in flood depth estimates.
In this study, we develop a conditional latent diffusion framework to generate fine-grid high-resolution flood depth maps using coarse-grid flood simulations and digital elevation models (DEMs) as conditioning inputs. The approach is demonstrated for a coastal floodplain near Tacloban, Philippines, which is subject to complex compound flooding driven by inland rainfall and storm surge. Hydrodynamic simulations are performed using a subgrid-based shallow water solver. The coarse-grid model contains approximately 95 times fewer computational cells than the fine-grid model and executes around 188 times faster, albeit with reduced accuracy (pixel-wise RMSE of 81.2 cm for maximum flood depth map).
Fine-grid model outputs are treated as deterministic ground truth, allowing uncertainty arising solely from the stochastic behaviour of the LDM to be isolated. By repeatedly sampling the trained model for identical inputs (up to 100 stochastic runs), we systematically evaluate accuracy–uncertainty–cost trade-offs using RMSE and pixel-wise 90% confidence interval (CI) coverage of flood depths.
Results show that individual stochastic predictions substantially improve upon the coarse-grid baseline but exhibit notable variability, with RMSE ranging between approximately 19–24 cm (Figure 1). Ensemble averaging rapidly enhances accuracy, with ensemble-mean RMSE converging within 20–40 runs, yielding 26–14 times speed-up compared to fine-grid hydrodynamic simulations. However, despite increasing ensemble size, empirical 90% CI coverage stabilises at around 70%, indicating systematic under-capture of uncertainty. Increasing the number of reverse diffusion steps from 500 to 1000 does not significantly alter this behaviour (Figure 2) suggesting that uncertainty limitations are not driven by insufficient sampling resolution.
Further analysis indicates that uncertainty under-representation arises from overly strong conditional signals learned during training rather than ensemble size. Introducing controlled stochastic perturbations (Figure 2), into the latent representation of coarse-grid flood maps at inference time increases ensemble spread and substantially improves CI coverage, reaching approximately 86% for a noise factor of 0.2, while only marginally increasing RMSE (~0.3 cm).
The study highlights three key insights: (i) stochastic LDM ensembles provide a practical balance between accuracy and computational efficiency for operational flood mapping; (ii) increasing ensemble size alone yields diminishing returns for uncertainty representation under strong conditioning; and (iii) future research should focus on incorporating uncertainty-aware conditioning during training and leveraging advanced diffusion solvers to further reduce inference cost. Together, these findings establish a principled pathway toward fast, uncertainty-aware flood inundation modelling using generative AI.
Figure 1: Accuracy–uncertainty–cost trade-off.
Figure 2: Uncertainty coverage–accuracy trade-off.
How to cite: Herath Mudiyanselage, V. V. H., Saha, A., Rasnayaka, S., and Marshall, L.: Fast and uncertainty-aware super-resolution of compound flooding using latent diffusion models, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10180, https://doi.org/10.5194/egusphere-egu26-10180, 2026.
Recently, differentiable parameter learning was presented as a deep-learning calibration method that estimates transfer relationships between physical characteristics and calibration parameters (Tsai et al., 2021). Such methods are especially important for large-scale hydrological models, as these transfer relationships allow for estimating consistent and seamless parameter fields in regions without observations. Although parameter learning calibration is shown to efficiently improve the simulation performance, the uncertainties related to this approach are poorly understood.
Our study distinguishes and quantifies the various sources of parameter learning calibration uncertainties with a structured set of calibration experiments using a synthetic dataset generated with a physically based global hydrological model. As the “true” parameters are known in each experiment, our study can distinguish and quantify uncertainties related to: the transfer function form, deep-learning surrogate gradient transfer, deep-learning surrogate performance, geographical bias in available observations, and non-uniqueness.
Our results show that the parameter learning calibration approach is robust under a wide range of possible transfer function forms, gradient transfer through a deep-learning surrogate model, and geographical biases in available observations. In addition, parameter learning calibration is somewhat robust non-uniqueness issues. However, parameter learning is most sensitive to errors in the deep-learning surrogate model's predictions or, conversely, the observations. Moreover, estimated parameters improve the simulation performance even when they are erroneous, indicating better results for the wrong reasons.
Our study highlights the significant potential of deep learning to understand and extrapolate relationships from potentially limited observational data. However, when using the parameter learning calibration approach, care should be taken to select the appropriate parameters and introduce some form of regularization to avoid unrealistic parameterizations.
References
Tsai, W. P., Feng, D., Pan, M., Beck, H., Lawson, K., Yang, Y., ... & Shen, C. (2021). From calibration to parameter learning: Harnessing the scaling effects of big data in geoscientific modeling. Nature communications, 12(1), 5988.
How to cite: Droppers, B., Bierkens, M. F. P., and Wanders, N.: Uncertainties in differentiable parameter learning calibration, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10651, https://doi.org/10.5194/egusphere-egu26-10651, 2026.
Numerical flow models are essential tools in hydraulic engineering and form the basis for a wide range of planning and decision‑making processes. Rapid advances in data availability, modelling techniques, and computational power enable increasingly detailed simulations and appealing visualizations, yet this can also lead to overconfidence in the models and obscure errors, while also complicating the assessment of model robustness. Building on an earlier international survey highlighting the lack of standardized procedures in current practice, the DWA (German Association for Water, Wastewater and Waste) working group WW‑1.7 “Qualitätssicherung und -management beim Einsatz mehrdimensionaler Strömungsmodelle” is developing a structured quality management (QM) system for all participants in the process of flow modeling.
This contribution presents the first conceptual version of this QM system. A guiding design principle is the balance between comprehensive and in-depth quality assurance and the clearly expressed need for intuitive and time‑efficient tools. Requirements and expectations of different stakeholder groups are systematically incorporated into the framework to ensure broad acceptance and usability.
A central component of the development concept is a supporting uncertainty analysis designed to identify critical modelling processes that should be given special consideration in the quality management system. The approach aims to systematically assess how variations in data and key modelling parameters influence model outcomes and contribute to overall uncertainty. To this end, selected modelling processes are examined across several representative test cases. The results are used to refine the prioritization and structuring of QM components by indicating which modelling steps require enhanced quality assurance and documentation.
The development concept presented here provides insight into ongoing efforts toward a comprehensive and robust QM framework, with the aim of enhancing transparency, robustness, and reproducibility in hydraulic flow modelling and reducing the dependence of modelling quality on individual or institutional backgrounds.
How to cite: Knepper, F., Oberle, P., and Franca, M. J.: Towards a Quality Management System for Flow Modelling: Integrating Uncertainty Analysis in Conceptual Development, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16777, https://doi.org/10.5194/egusphere-egu26-16777, 2026.
The transition from intermittent water supply (IWS) to continuous water supply (CWS) is a critical goal for global water security, but it continues to be hampered by the difficulty of precisely determining the resources required for 24-hour service. Traditional water audits are often designed for ideal systems with CWS and rely on static, annual, or system-wide water balances that fail to capture temporal variability. This study addresses Network Input Volume (NIV), which represents the daily amount of water delivered to consumers from the service reservoir supplying the city, and hours of supply (SH). By focusing on dynamic system behavior instead of static averages, this approach provides water companies with a more nuanced operational perspective for planning infrastructure transitions in data-scarce environments.
The methodology was applied to the Tillo district of Siirt, Turkey using service reservoir outlet data and recorded hours of supply from January to August 2023. To account for the network's inherent scholastic structure, we used regression analysis across bi-monthly periods, including both 95% confidence intervals and forecast bands. Our findings reveal a strong positive correlation between NIV and SH, but this varies considerably across seasons. For the March-April period, the adapted regression suggests a CWS threshold of approximately 600 m³/day. Specifically, the wider forecast range indicates that continuity could theoretically be ensured at values as low as 390 m³/day, whereas at 650 m³/day, at least 15 hours of supply is guaranteed.
As climatic demand increases towards the summer months, the model captured a significant increase in requirements. Extrapolations for the May-June and July-August periods showed that the CWS thresholds would rise to approximately 870 m³/day and 970 m³/day, respectively. However, the analysis also identified a critical hydraulic phenomenon: "compensatory flow". On days when water returned following periods of supply shortages, the system experienced temporarily elevated NIV values as it compensated for the previous deficit. This cumulative adjustment dynamic demonstrates that the relationship between input and supply is not merely instantaneous but is shaped by the system's memory of previous IWS cycles.
Consequently, this research shows that the seasonal and daily relationship between SH and NIV is another point to consider during the transition to CWS. This daily monitoring approach, which goes beyond annual balances and captures daily and seasonal variability, allows water companies to establish realistic metrics for the transition. Furthermore, these findings highlight the importance of integrated controls that account for both physical losses and human-induced demand shifts, providing a replicable model for improving urban water resilience in similar contexts worldwide.
How to cite: Yamac, E., Collins, R., and Speight, V.: Determining Minimum Network Input Volume for The Transition from Intermittent Water Supply (IWS) To Continuous Water Supply (CWS): A Seasonal Analysis of Supply Hours in Tillo, Turkey, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-13718, https://doi.org/10.5194/egusphere-egu26-13718, 2026.
Accurate precipitation estimation is fundamental for the analysis of hydrological processes, especially in urban areas with limited rain-gauge networks. The objective of this study was to develop two models based on the Random Forest (RF) algorithm for the detection of rainy and non-rainy days and for the estimation of daily precipitation during the wet season in the city of Culiacán, Sinaloa, Mexico. For this purpose, in situ meteorological station data and variables derived from images from the GOES-16 geostationary satellite were used, employing only spectral bands available 24 hours a day, specifically bands 7, 9, 13, 14, and 15. As part of the preprocessing stage, a parallax correction and a temporal adjustment were performed to harmonize the different data sources. Additionally, a prior classification of the days under analysis was implemented to reduce the radiometric heterogeneity of the training dataset. According to the main results, the rainfall detection model showed satisfactory performance, with an accuracy of 88%, a sensitivity of 86%, and a specificity of 89%, indicating an adequate ability to identify the presence and absence of precipitation. In turn, the precipitation estimation model achieved a correlation coefficient (R) of 0.74, a mean absolute error (MAE) of 6.59 mm, and an RMSE of 14.26 mm, demonstrating a good capacity to capture temporal variability, although with a tendency to overestimate intense events. The variable importance analysis showed that infrared bands 13 (10.3 μm) and 14 (11.2 μm) dominate the estimation in most groups, while the band 7 (3.9 μm) band becomes more relevant in events associated with microphysical processes. In conclusion, the integration of GOES-16 data and machine learning models have shown to be a viable alternative for complementing precipitation information in urban areas with scarce rain-gauge instrumentation; however, its application to other regions or periods requires model retraining.
How to cite: Alfaro Valencia, L., Rentería Guevara, S. A., Lobato Sánchez, R., and Monjardín Armenta, S. A.: Urban Precipitation Estimation Using GOES-16 Infrared Observations and Random Forest Models, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15537, https://doi.org/10.5194/egusphere-egu26-15537, 2026.
Spatiotemporal Characteristics of Precipitation and Drought Variability in Taiwan Using Multi-dimensional Complementary Ensemble Empirical Mode Decomposition
Abstract
Most time series observed in natural systems are nonlinear and nonstationary, particularly under the influence of climate change. Taiwan has experienced increasingly frequent drought events in recent decades. Droughts are characterized by their gradual development, cumulative impacts, and lack of clear early warning signals, which makes their detection and analysis challenging.
To address these issues, this study applies Multi-dimensional Complementary Ensemble Empirical Mode Decomposition (MCEEMD) to analyze long-term temperature and precipitation data in Taiwan from 1960 to 2023. MCEEMD is an effective time–frequency analysis method designed for nonlinear and nonstationary time series. It enables the decomposition of multi-dimensional signals into a set of Intrinsic Mode Functions (IMFs), allowing both spatial and temporal characteristics of climate variables to be examined. Through these IMFs, meaningful instantaneous frequencies and long-term trends in the signals can be identified.
This study considers both stochastic and deterministic influences by reconstructing the IMFs into two components based on their autocorrelation coefficients. The relationships between temperature, precipitation variability, and drought-related characteristics are then examined, providing insights into the spatiotemporal behavior of drought events in Taiwan.
Key word: Multi-dimensional Complementary Ensemble Empirical Mode Decomposition(MCEEMD ),Intrinsic Mode Functions (IMFs)
How to cite: Tang, C.-H. and Tsai, C. W.: Spatiotemporal Characteristics of Precipitation and Drought Variability in Taiwan Using Multi-dimensional Complementary Ensemble Empirical Mode Decomposition, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15993, https://doi.org/10.5194/egusphere-egu26-15993, 2026.
Long-term hydrological time series are essential for planning effective water-resource management strategies that balance competing water and energy uses and preserve ecosystem functioning. In particular, long-term large-scale surface water temperature (SWT) time series are crucial for enhancing understanding of climate change impacts and for quantifying uncertainties in the occurrence of critical periods affecting water and energy uses, as well as ecosystem balance. However, these datasets inevitably contain missing observations, and long-term data series with large spatial coverage remain scarce. Modeling approaches provide valuable tools for estimating surface water temperature dynamics when observations are missing. Owing to their low data requirements and fast computation times, statistically based approaches are well suited to large spatial scales, where physically based approaches often become impractical to apply. Among statistically based methods, recurrent neural networks, such as Long Short-Term Memory (LSTM) models, have recently shown considerable potential for time series imputation (Cao et al., 2018 https://doi.org/10.48550/arXiv.1805.10572; Che et al., 2018 https://doi.org/10.1038/s41598-018-24271-9) and for simulating hydrological variables, including SWT (e.g. Saadi et al., 2025 https://doi.org/10.5194/egusphere-2025-3393). The aim of the present work was to develop and assess an approach for reconstructing long-term SWT time series at the scale of a large river basin using an LSTM model. The study was conducted at the scale of the Seine River Basin, including nearly 80 monitoring stations providing daily SWT observations, and relied on continuous meteorological data from 1958 to 2025 derived from the SAFRAN system (Vidal et al., 2010 10.1002/joc.2003). The developed model was designed to simulate a one-year daily SWT sequence, considering both dynamic and static inputs. Dynamic inputs include one-year sequences of meteorological data and the daily SWT time series to be reconstructed, as well as masks used to identify missing values in the SWT input (Quian et al., 2024 arXiv:2405.17508v1). Static inputs include features characterizing the monitoring stations, such as hydrological (mean and low-flow discharges), geographical and meteorological features. The model architecture is composed of two sequential modules: (i) a bidirectional LSTM that encodes basin-scale temporal dynamics from dynamic inputs, and (ii) a multilayer perceptron that combines the LSTM’s final hidden states with a learned embedding representing the target monitoring station to generate the full annual SWT sequence. This approach enables the reconstruction of daily SWT across the basin over multiple decades, handling a wide range of missing-data situations - from sporadic gaps to entirely missing time series - by leveraging covariates and influential drivers, primarily meteorological factors.
How to cite: Metayer, G., Rivière, A., Corral, D., Roy, V., and Thomas, W.: Reconstructing multi-decadal daily river water temperature in the Seine RiverBasin (France) with a bidirectional LSTM and basin-location embeddings, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20702, https://doi.org/10.5194/egusphere-egu26-20702, 2026.
Addressing water availability ecosystem service in human water resources is fundamental for the development of strategies that encompass sustainable pathways in a climate changing era. The total amount of water available for human activities is essential to economic development, influencing food production, power generation, human well-being, and healthy environments. Although key drivers of water availability, such as precipitation and land use, are well established in the literature, other potentially influential factors remain underexplored, including population density, GDP per capita, the human development index (HDI), water governance indicators, and total water demand. In this work, we developed a new concept for water management through the lens of ecosystem services approach. This framework emphasizes understanding the socio-economic and environmental drivers that influence water yield, aiming to enhance human well-being by promoting best practices in water management. This perspective enables a deeper understanding of the mechanisms influencing water availability beyond conventional assessment methods, while prioritizing management and restoration strategies. In this context, hydroinformatics enables advanced spatial analysis for examining water availability and ecosystem services. By integrating data analytics and machine learning (Random Forest) with traditional modeling approaches, it is possible to uncover complex relationships between socio-economic and environmental drivers and their spatial influence on water resources. At the same time, combining with Geographically Weighted Regression (GWR) tool, it is possible to analyze how socio-economic and environmental factors influence water availability ecosystem services across different geographic regions. This is possible because GWR captures spatial variability by estimating local rather than global relationships. Finally, this methodology will be applied globally using Level 5 basins from HydroATLAS, allowing the identification of regional heterogeneities, cross-scale patterns, and dominant local drivers of water availability. This global application provides a robust basis for comparing basins and supporting targeted management and policy interventions. The results are expected to provide a global understanding of how human and environmental factors jointly regulate water availability, supporting the design of more adaptive, equitable, and resilient water management strategies.
How to cite: Silva, G., Silva, P., Benso, M., Morellato, L. P., and Mendiondo, E.: Assessing the Influence of Socio-Economic and Environmental Drivers on Water Availability Ecosystem Services: A Geographically Weighted Regression approach, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-892, https://doi.org/10.5194/egusphere-egu26-892, 2026.
The implementation of managed aquifer recharge (MAR) systems, which have the potential to store surface water underground for future use or environmental advantages, has become increasingly popular due to the world's growing water scarcity. It has been demonstrated that MAR is a successful strategy for mitigating the effects of climate change on the world's water supplies as well as on issues related to spatiotemporal water shortages. However, conflicting goals, such as maximizing recharge efficiency while reducing total operational costs, must be balanced while designing MAR systems. This study aims to create a novel framework for a multi-objective optimization of MAR systems to handle such kind of trade-offs and also support decision-making.
This work introduces the first design steps and the general structure of a framework that integrates the capabilities of the existing web-based groundwater modelling platform INOWAS (www.inowas.com) with a hybrid evolutionary algorithm. The framework effectively explores optimal solutions in complex solution spaces by combining groundwater models implemented on the INOWAS platform with tools from the MODFLOW family (MODFLOW-2005, MT3DMS, SEAWAT) with global search capabilities (e.g., Genetic algorithm) and local refining methods (e.g., Simplex algorithm). The evaluation of the first design steps of the proposed framework was conducted with python, and not through direct implementation on the INOWAS platform.
The proposed framework is applied to the Akrotiri River Basin, a coastal region in the southern part of the Republic of Cyprus, facing complex and competing water management challenges. The region faces a number of key challenges related to water-scarcity, such as seawater intrusion into the coastal aquifer, overexploitation of the groundwater resources, deterioration of hydrochemical water quality, lack of sufficient monitoring infrastructure and low trust from local farmers in current water management strategies. Soil aquifer treatment (SAT) has been implemented at the site since 2016 through the infiltration of tertiary-treated wastewater using seventeen recharge ponds. The proposed framework integrates groundwater flow and transport modeling with a multi-objective optimization algorithm to simultaneously enhance groundwater quantity and quality, mitigate saltwater intrusion, protect drinking wells from any negative impact of injectant flow, hence supporting the reliability of water supply. Several meaningful trade-offs between these competing objectives are explicitly explored. Solutions are expressed as pareto fronts, which represent sets of optimal trade-off solutions that are non-dominated with respect to one another and superior to all other solutions in the search space.
How to cite: Taani, M., Händel, F., Panayiotou, C., Glass, J., Stefan, C., and Scheytt, T.: Multi-Objective Optimization Framework for Managed Aquifer Recharge: A case study in Akrotiri, Cyprus, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-22691, https://doi.org/10.5194/egusphere-egu26-22691, 2026.
A significant share of freshwater discharge from Baltic Sea drainage basin (BSDB) originates from unmonitored or poorly monitored coastal catchments, which increases uncertainty in regional water balance assessments and in estimates of nutrient and pollutant loads to the Baltic Sea. This study presents a data-driven regionalization framework designed to estimate surface discharge in unmonitored Baltic Sea catchments at monthly, seasonal, and annual scales for the period 2001-2020. A large-sample dataset for about 720 monitored basins is compiled using Global Runoff Data Centre discharge records together with hydro-meteorological and land-surface predictors, including precipitation, evapotranspiration and temperature, topographic attributes, and land-cover fractions. Predictors are selected based on the catchment water balance and processed consistently across all basins using zonal statistics. Multiple linear Regression (MLR) and Random Forest (RF) models are trained on specific discharge, and several modelling configurations are evaluated, including temporal grouping, geographical neighbouring strategies, clustering approaches and the inclusion of correlated variables. A hybrid correction method helps identify which parts of each BSDB are monitored and which are not, making sure discharge is predicted only for the unmonitored areas. The most effective configuration was combined temporal grouping with geographical neighbouring, and it achieved satisfactory performance (NSE > 0.5) for roughly 75% of basins and very good performance (NSE > 0.75) for more than half of basins. Median absolute percentage errors were below 30%. Land use characteristics (e.g. crop land, forest, waterbodies) provide important explanatory power alongside climatic and topographical variables. The framework provides consistent discharge estimates for ungauged coastal basins in the Baltic Sea region and can be applied in other areas where data is limited to support regional water balance and pollutant load assessment.
How to cite: Vijayan, A., Airiaud, Y., and Kalantari, Z.: Data-Driven Regionalization of Surface Discharge in Unmonitored Catchments of the Baltic Sea Drainage Basin, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1060, https://doi.org/10.5194/egusphere-egu26-1060, 2026.
The Kis-Balaton Water Protection System (KBWPS) is a complex, semi-constructed wetland habitat consisting of three separate units that play a significant role in protecting water quality of Lake Balaton, the largest shallow freshwater lake in Central Europe (Bhomia et al., 2021). Previously, it was noted that the water balance of the KBWPS can only be determined with high uncertainty; specifically, the seasonal variation in the ratio of evaporation to transpiration. Hovewer, estimating evaporation is one of the most crucial factors in water balance calculations. Until now, a “homogeneous method” developed for Fertő/Lake Neusiedl (Bhomia et al., 2021) has been applied for the KBWPS, which is rather an oversimplification for the highly heterogeneous lake and marsh complex of KBWPS. Therefore, the aim of the current study is to develop a system-specific approach tailored to the KBWPS’ spatial heterogeneity.
To quantify and predict the system’s hydrological behaviour, a Long Short-Term Memory (LSTM) model was developed to estimate daily outflow discharge at a single outlet point. The model was trained using meteorological variables and observed daily discharge time series, allowing the network to capture temporal dependencies and delayed system responses. In parallel, monthly Sentinel-2 imagery and daily in-situ measurements were analysed using trend analysis and seasonal decomposition to investigate the temporal variability of key hydro-meteorological parameters. NDVI-based satellite estimates were applied to characterise evapotranspiration dynamics. A comprehensive statistical analysis of time series, including air humidity, air temperature, wind conditions, and water chemistry data, was carried out to identify correlations between the individual parameters. The applied statistical and machine learning methods effectively captured the temporal dynamics of the system.
In addition, Sentinel-2 satellite data was used to refine the spatial structure of vegetation, which influenced directly the transpiration. The development of a vegetation delineation methodology, based on NDVI classification, contributes to more accurate determination of water balance components by separating water surfaces from vegetation-covered areas.
Another uncertain element of the system is the yield data series from the point-shape civil engineering structure, which connects the Kis-Balaton hydrological system to Lake Balaton. Formerly, the correction of the yield time-series was required human resources. To reduce measurement errors and decrease the need for the manual correction, a deep learning-based model is under development, which determines seasonal correction factors. To address this problem, precipitation and wind speed data are also used as suitable predictors in addition to the daily water flow time series.
The expected outcome of the research is a comprehensive, scientifically sound methodology that will enable more accurate water balance calculations for Kis-Balaton and contribute to more efficient water management support for the system in the long term.
The research was supported by the National Multidisciplinary Laboratory for Climate Change, RRF-2.3.1–21-2022–00014 project.
Bhomia, R. K., Clement, A., Látrányi-Lovász, Z., Kaur, R., Rousseau, D., Louage, F., Wang, Q., Hatvani, I. G. (2021). Case studies of (semi) constructed wetlands treating point and non-point pollutant loads to protect downstream natural ecosystems. In Reference module in earth systems and environmental sciences. Elsevier.
How to cite: Petneházy, A., Szijártó, M., Fórizs, I., Czuppon, G., Kapolcsi, F., Látrányi-Lovász, Z., Clement, A., and Hatvani, I. G.: Integration of time-series analysis, satellite data and machine learning in water balance assessment for the Kis-Balaton Water Protection System, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2375, https://doi.org/10.5194/egusphere-egu26-2375, 2026.
As urban flood risks intensify due to climate change and rapid urbanization, robust assessment of drainage system resilience has become increasingly important. This study proposes a deep learning–based framework to evaluate flood resilience in urban drainage networks (UDNs) using node-level hydraulic predictions. The framework integrates Graph Neural Networks (GNNs) and Transformer models to predict water depth at each network node under multiple storm scenarios. GNNs capture spatial dependencies and network topology within the drainage system, while the Transformer models temporal rainfall–runoff dynamics. Flooding conditions at nodes are identified by applying depth-based thresholds to the predicted water levels, enabling the generation of time-resolved flood maps across the network. Flood resilience is assessed at the node level by adapting the Simple Urban Flood Resilience Index (SUFRI). Three indicators are considered: normalized flood depth at nodes, recovery time required for water levels to return to normal conditions, and flooding frequency. These indicators are combined to derive resilience scores for individual nodes, which are further weighted according to their hydraulic and topological importance within the network, considering factors such as flow capacity, connectivity, and redundancy. System-level resilience is obtained by aggregating the weighted node-level resilience scores. The proposed framework is applied to a real-world urban drainage system to evaluate resilience under diverse storm scenarios. Results reveal critical nodes and vulnerable regions that disproportionately influence overall system performance. Based on the analysis, targeted optimization strategies—such as capacity enhancement, redundancy improvement, and recovery acceleration—are suggested to mitigate future flood risks. The framework provides a scalable and data-efficient decision-support tool for urban flood resilience assessment and infrastructure planning, particularly in data-scarce urban environments.
Acknowledgement
This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Ministry of Science and Technology (RS-2024-00356786) and Korea Environmental Industry & Technology Institute grant funded by the Ministry of Environment (RS-2023-00218973).
How to cite: Li, S. and Park, J.: Assessing flood resilience in urban drainage networks using deep learning–based hydraulic predictions, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-4692, https://doi.org/10.5194/egusphere-egu26-4692, 2026.
River discharge is a key hydrological issue for the river and water resources management. Recently, climate change is likely exacerbating the frequency and intensity of the extreme flood events, which indicates continuous monitoring of water discharge and its variation at different time scales are of prime important, especially for large river basins. The stage–discharge relationship or rating curve in a river is very useful because it allows computing the discharges from measured water levels at a gauge station. A single-valued rating curve can be used for a nearly steady regime. However, a complex relationship between stage and discharge should be established when there is a non-stationary regime due to, for example, the operation of artificial constructions, such as dams and weirs. This study aims to evaluate the stage-discharge relationship considering weir operation. A machine learning architecture, gated recurrent unit (GRU), is developed to determine the complex relationship between water level and discharge at the Yeoju Bridge which is located between Gangcheon and Yeoju weirs. To consider both of the upstream and downstream weir operation, observed upstream and downstream water levels of individual weirs are included as GRU inputs. The root mean squared error (RMSE) is adopted to assess the GRU performance. Our findings show that the GRU model considering the effect of weir operation can estimate the discharge with satisfactory accuracy by establishing the relationship between stage and discharge. The approach introduced in this study enables the estimation of discharge in stream networks with abundant artificial constructions, such as weirs and estuary barrages, where streamflow is highly affected by their operations.
This study was supported by the Korea Environmental Industry andTechnology Institute (KEITI) (Grant number: 2022003460001)
How to cite: Jun, K. S. and Li, L.: Analysis of non-unique stage–discharge relationship affected by downstream weir operation using a deep learning method, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6125, https://doi.org/10.5194/egusphere-egu26-6125, 2026.
The reliability of computationally expensive geophysical and environmental models for simulating land surface processes strongly depends on accurate parameter calibration. However, traditional optimization algorithms often require thousands of model evaluations, making them unsuitable for such complex models. We propose an adaptive surrogate modeling-based optimization algorithm with active learning (ASMOAL), an efficient calibration framework that integrates surrogate modeling with a trust-region active learning strategy. At each iteration, ASMOAL adaptively selects informative parameter samples within a trust region, prioritizing high-potential and physically plausible regions, and updates the surrogate to guide the search toward improved solutions with limited model runs.
We first evaluate ASMOAL on nine benchmark functions to verify convergence behavior and robustness. Then the algorithm is applied to three geophysical models with increasing complexity: the Variable Infiltration Capacity (VIC) model and the Xinanjiang (XAJ) model in two river basins in China, and the flash–Flood Event–based Spatially distributed rainfall–runoff Transformation (FeST) in two river basins in Italy. In addition, we conduct parameter sensitivity analysis to investigate how parameter relevance and interactions shape the search dynamics and accuracy of ASMOAL. The results demonstrate that sensitivity patterns can vary across basins and models, and that accounting for sensitivity information is critical for interpreting calibrated parameters and reducing the risk of equifinality. Moreover, the proposed algorithm exhibits improved convergence, calibration accuracy, and robustness compared to existing surrogate-based methods. The results also reveal that the optimal parameters obtained by ASMOAL tend to cluster within physically meaningful regions, highlighting the importance of focused search. The proposed ASMOAL algorithm offers a promising solution for enhancing parameter calibration in a wide range of computationally expensive geophysical and environmental models.
How to cite: Yang, Q., Sun, R., Mancini, M., and Ravazzani, G.: Active Learning–Based Surrogate Optimization Algorithm for Calibrating Computationally Expensive Geophysical Models, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6713, https://doi.org/10.5194/egusphere-egu26-6713, 2026.
Flooding is the costliest disaster in Canada, yet traditional flood susceptibility modelling is computationally expensive for large-scale applications and often relies on static geospatial features while excluding temporal antecedent conditions. This study uses the Canadian Flood Archive maintained by Natural Resources Canada (NRCan) to develop a large-scale flood susceptibility model for Canada's major drainage areas. The model integrates static terrain derivatives, dynamic climate variables, and semi-static geospatial variables using a hybrid Convolutional Neural Network-Convolutional Long Short-Term Memory (CNN-ConvLSTM) framework. The Canadian Medium Resolution Digital Elevation Model (MRDEM) was used to derive geospatial features, including height above nearest drainage (HAND), Euclidean distance to rivers (EUC), slope, aspect, topographic position index (TPI), and terrain ruggedness index (TRI). Semi-static geospatial variables include land cover (available every 5 years) and the annual normalized difference vegetation index (NDVI), which were temporally matched to each historical flood event. The static and semi-static features were coupled with Daymet meteorological data (precipitation, temperature extremes, snow water equivalent) spanning 1–3-month antecedent windows. The performance of the 2D hybrid CNN-ConvLSTM model will be compared with an Extreme Gradient Boosting (XGBoost) baseline. While XGBoost has performed well in prior research, the hybrid CNN-ConvLSTM is hypothesized to offer superior interpretability of flooding mechanisms. By leveraging the temporal sequence of meteorological drivers, the model captures complex spatiotemporal dependencies that traditional machine learning methods cannot. A preliminary sensitivity analysis of temporal sequence lengths (1-3 months) and resampling ratios (0.1-0.7) showed that the CNN-ConvLSTM architecture achieved the highest predictive accuracy (F1 = 0.89) with a 3-month sequence length and a resampling ratio of 0.5. These initial findings suggest that capturing the full spring snowmelt-to-rainfall cycle is critical for flood susceptibility mapping in Canadian watersheds.
How to cite: Dunbar, K. E., McGrath, H., and Khan, U.: Integration of Temporal Meteorological and Geospatial Data for Flood Susceptibility Modelling in Canada using a Hybrid CNN-ConvLSTM Model, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8478, https://doi.org/10.5194/egusphere-egu26-8478, 2026.
Accurate simulation of hydropower output characteristics is a prerequisite for optimizing long-term reservoir scheduling. However, traditional empirical formulas often fail to capture the complex non-linear relationships between hydraulic head, turbine discharge, and power output, while purely data-driven models lack adherence to physical laws. This paper proposes a Physics-Informed Machine Learning (PIML) method that couples physical prior knowledge with data-driven modeling. By embedding strictly defined physical constraints—specifically dynamic head-dependent capacity limits, hydraulic monotonicity, and tailwater elevation effects—into the loss function of a Deep Neural Network (DNN), the proposed model guarantees physically consistent predictions. The PIML model is further integrated as a high-fidelity surrogate into a long-term scheduling optimization model solved by Particle Swarm Optimization (PSO). Case studies on the Shuibuya Hydropower Station demonstrate that the PIML method achieves high simulation accuracy with an RMSE of 12.25 MW and zero physical violations. Furthermore, under identical hydrological conditions, the PIML-based scheduling strategy increases annual power generation by 4.72% and reduces the water consumption rate by 4.50%, effectively identifying high-efficiency operating zones compared to traditional methods.
How to cite: Liu, Z., Liu, P., and Cheng, L.: A Physics-Informed Machine Learning Method for Long-Term Hydropower Output Simulation and Scheduling, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9281, https://doi.org/10.5194/egusphere-egu26-9281, 2026.
Approximately 30% of the agricultural land in the Czech Republic (around 1 million hectares) has been drained, which significantly affects the water regime and water availability in the landscape. The largest expansion of agricultural drainage systems occurred during the communist and socialist era, particularly between the 1950s and 1980s. Old paper project documentation for these systems has often not been preserved, archival records are highly fragmented, and many existing plans do not correspond to the actual implementation.
Accurate mapping of drainage systems is essential for understanding detailed hydrological processes in the landscape, as well as for designing measures to mitigate their negative impacts. Automatic detection of drainage systems represents an important step toward comprehensive mapping of functional drainage structures. Traditional approaches based on manual interpretation of aerial imagery are time-consuming and practically infeasible for large areas. Therefore, the presented project proposes and tests a method using convolutional neural networks for the segmentation of drainage lines from high-resolution aerial imagery. The aim is to assess the potential of up-to-date machine learning techniques for automated extraction of drainage systems in landscapes with varying vegetation cover.
A critical component of the workflow is the preparation of training data, including manual annotation of drainage lines, creation of mask layers, and data augmentation to enhance model generalization. Preliminary results will be presented, including segmentation examples and discussion of key limitations, such as sensitivity to vegetation cover.
Segmentation is implemented using the U-Net architecture, widely applied for pixel-level classification tasks in geosciences. The encoder is based on ResNet34, enabling hierarchical feature extraction and improving robustness to texture and illumination variability in aerial imagery. The implementation was carried out in PyTorch using the Segmentation Models PyTorch library. Skip connections between corresponding levels ensure preservation of spatial details and accurate localization of linear structures typical of drainage systems.
Results indicate that deep neural networks significantly accelerate and improve the accuracy of drainage feature identification, opening new possibilities for various landscape analyses, incl. hydrology, agricultural management, landscape planning, nature protection etc. Future work will focus on expanding the training dataset, optimizing hyperparameters, and validating the model on a large set of aerial images.
How to cite: Marval, Š., Princ, T., and Poláková, L.: Deep Learning-Based Segmentation of Land Drainage Systems from High-Resolution Aerial Imagery, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10817, https://doi.org/10.5194/egusphere-egu26-10817, 2026.
There is a fast development in technical solutions for water management. Digital Twins are among these fast-developing technologies, offering a platform for scientists, policy makers, and other stakeholders to exchange knowledge. However, the Digital Twins require also efficient hydrological information gain. New measurement techniques are causing a rapid growth of data, often resulting in scattered or incomplete datasets. Machine learning can be used to detect patterns, identify relations between variables, and fill data gaps. Typically, machine learning needs high-quality and long-term data for training. This is not always available, especially for variables that are obtained from short-term field campaigns.
This study explores traditional machine learning algorithms to optimize hydrological information gain from large datasets. Data from four study sites in three intensively studied Finnish rivers are used as a case study. The rivers are in the south, middle, and north of Finland and cover climatic conditions from boreal to sub-arctic. The approach involves the development of a simple application that enables users to gain maximum understanding with minimal user input. The main goals of the application are to detect patterns, recognize different river conditions and seasonality, fill data gaps, identify variable importance under different environmental conditions, and provide insights on variable relationships. The case study shows differences in seasonality, and therefore, differences in variable importance between the different rivers.
How to cite: Olsman, J., Johnson, J., Mäkinen, V., and Lotsari, E.: River system understanding through machine learning in Digital Twins, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-11460, https://doi.org/10.5194/egusphere-egu26-11460, 2026.
A good understanding of uncertainty is of paramount importance in the hydrological sciences, notably in streamflow prediction. Over recent decades, hydroinformatics has played a key role in advancing hydrological prediction through the exploration of physically inspired and conceptual models and data-driven approaches. In particular, machine learning (ML) models demonstrate strong predictive skills. Despite this, limited interpretability and potentially weak extrapolation under extreme conditions remain major disadvantages of ML applications [1,2].
To address these limitations, a hybrid framework that combines conceptual hydrological modelling with machine learning–based probabilistic forecasting is proposed. The so-called Generalized Pareto Uncertainty (GPU) framework can be used to train an ensemble of models (potentially physically based) so that it reliably reproduces the predictive uncertainty of the output [3]. In this case, GPU is employed with the conceptual HYdrological Predictions for the Environment (HYPE) model. By embedding hydrological knowledge into a data-driven uncertainty framework, the proposed approach seeks to improve robustness, generalization, and physical consistency of streamflow forecasts.
GPU relies on finding a multi-objective optimal surface (something akin to a double Pareto surface) that selects model parameters that span the full range of exceedance of simulations—at the extremes, forcing some models to always underpredict and others to always overpredict—while simultaneously searching for optimal error metrics (e.g., Nash-Sutcliffe efficiency, King-Gupta efficiency, mean absolute error, etc.). One promising feature of the framework is that it is not constrained to one type of error metric or even two dimensions (exceedance and error metric), potentially even opening avenues for addressing equifinality challenges.
The methodology is applied to the Nabão and Douro river basin in Portugal, one basin in Sweden, and one in Ireland. The performance of three modelling strategies is compared: (i) a standalone conceptual model (HYPE), (ii) GPU combined with artificial neural networks (missing indirect foreknowledge about hydrological processes), and (iii) a hybrid approach that incorporates HYPE models as ensemble members. Results show that the inclusion of conceptual hydrological information leads to clear improvements in the quality of the predictive uncertainty estimates, including its resolution, reliability, and aggregate metrics (e.g., CRPS).
In this work, we clarify the concept behind GPU, demonstrate its results, address challenges, and discuss potential innovative applications.
[1] Baste, S., Klotz, D., Acuña Espinoza, E., Bardossy, A., & Loritz, R. (2025). Unveiling the limits of deep learning models in hydrological extrapolation tasks. Hydrology and Earth System Sciences, 29(21), 5871–5891. https://doi.org/10.5194/hess-29-5871-2025
[2] Nearing, G. S., Kratzert, F., Sampson, A. K., Pelissier, C. S., Klotz, D., Frame, J. M., Prieto, C., & Gupta, H. v. (2021). What Role Does Hydrological Science Play in the Age of Machine Learning? In Water Resources Research (Vol. 57, Issue 3). Blackwell Publishing Ltd. https://doi.org/10.1029/2020WR028091
[3] Matos, J. P., Hassan, M. A., Lu, X. X., & Franca, M. J. (2018). Probabilistic Prediction and Forecast of Daily Suspended Sediment Concentration on the Upper Yangtze River. Journal of Geophysical Research: Earth Surface, 123(8), 1982–2003. https://doi.org/10.1029/2017JF004240
How to cite: Marinheiro, R. and Matos, J. P.: Non-parametric multidimensional uncertainty estimation employing a hybrid approach, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-12509, https://doi.org/10.5194/egusphere-egu26-12509, 2026.
How to cite: Callau Medrano, S., Nowak, W., Oladyshkin, S., and Seidel, J.: Let the data speak: Catchments as non-autonomous dynamical systems via a modulated storage–discharge function, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-13951, https://doi.org/10.5194/egusphere-egu26-13951, 2026.
Mass balance equation is the fundamental governing equation that links reservoirs’ inflow, storage and discharge; yet, it remains unclear to what extent each component can be inferred from the others. This is particularly the case when considering large samples of reservoirs with wide range of capacities and operational purposes. Here, we use an entropy-based framework to investigate the predictability of reservoirs’ storage and discharge based on the information of content of one another. Using the time series data of inflow, storage, and discharge from 52 reservoirs across the globe, we treat mass balance with two parallel approaches. First, we examine how well discharge can be constrained by antecedent storage and inflow. Second, we assess the predictability of storage based on discharge and inflow. Marginal and conditional entropies are used to measure and quantify information flows from one variable to the other and to evaluate how much uncertainty is reduced when additional information is introduced. We apply this approach to observational data as well as simulated data obtained from reservoir algorithms. Our results reveal considerable variability in entropy measures across reservoirs and between the two approaches. The suggested framework can be considered as flexible and empirical means for assessing reservoir algorithms and for evaluating the role of data assimilation in improving reservoir simulations in hydrology and land-surface models.
How to cite: Nazemi, A. and Mirdarsoltany, A.: On the information content of reservoirs’ storage and discharge for the predictability of one another, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17348, https://doi.org/10.5194/egusphere-egu26-17348, 2026.
TETRA – From Methodology to Operational Tools for Water-Based AI Projects
Maximilian Zenner, Tobias Hellmund, Jürgen Moßgraber, Issa Hansen, Salvador Peña Haro, Divas Karimanzira, Linda Ritzau, Florence Le Ber, Gaëlle Lortal
Fraunhofer IOSB, Karlsruhe, Germany (maximilian.zenner@iosb.fraunhofer.de)
The development of efficient and interoperable tools for monitoring water resources remains essential to ensure the sustainable availability of this vital resource for both society and ecosystems. Recent events such as the fish die-off in the Oder River further emphasize the urgent need for improved river monitoring and protection strategies.
Building on work previously presented, the TETRA project aims to enable and accelerate the practical adoption of artificial intelligence (AI) in water management, while fostering a shared European AI ecosystem through close collaboration between German and French partners. To establish a harmonized approach, the project builds on the PAISE methodology for the development of AI-based products and adapts it to the domain of public water management.
Since the previous contribution, TETRA has progressed toward an operational data pipeline: SEBA contributes an automatic data pipeline for its in-situ measurement stations into the FROST server. The acquired datasets include velocity profiles and bathymetric measurements, which are accessed by the TETRA knowledge base and visualized through an interactive web application.
The application provides a map-based overview of sensor stations and a dedicated analysis view featuring 3D visualizations of velocity and bathymetry profiles (s. Figure 1), including filtering options such as water level and temporal range. Ongoing work focuses on refining the UI/UX to further support data exploration and expert-driven analysis.
Figure 1: 3D visualization of a river’s surface velocity profile over time
Initial experiments in AI-based analysis revealed that the currently available measurement data are not yet sufficient in volume to robustly train data-driven models. To address this limitation, synthetic datasets derived from numerical simulations are used as a first step to evaluate model behavior and feasibility. While not a substitute for real-world measurements, this approach provides initial insights and establishes a foundation for future integration of increasing amounts of real sensor data.
In parallel, significant progress has been achieved within the restoration use case: ICUBE has advanced ontology-driven methods for the automated population of a case-based reasoning knowledge base from unstructured texts using large language models, while THALES has developed a semantic search module enabling concept-based retrieval of multilingual restoration documents beyond keyword-based search.
This research has received funding from the BMBF’s (Bundesministerium für Bildung und Forschung) directive on the funding of Franco-German projects on the topic of artificial intelligence, Federal Gazette of 20th June 2022.
How to cite: Zenner, M., Hellmund, T., Moßgraber, J., Hansen, I., Peña Haro, S., Karimanzira, D., Ritzau, L., Le Ber, F., and Lortal, G.: TETRA – From Methodology to Operational Tools for Water-Based AI Projects , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20836, https://doi.org/10.5194/egusphere-egu26-20836, 2026.
This study presents assessment of different types of water balances that exist within and in the surrounding of the crop using different methods like, 1) frequency analysis of crop specific growth threshold-based water balance; 2) field-scale vadose zone-based water balance. Moreover, a process-based crop water productivity modelling tool was also implemented with two distinct scenarios to simulate the working principle of Subsurface drainage system (SSDS) with CD approach along with adequate assumptions and knowledge of its limitations to simulate SSDS with CD approach for computation of climate and unsaturated soil zone-based fluxes of water balances. The process-based modelling tool also presents the quantification of the benefits to implement SSDS with Controlled Drainage (CD) approach in an agricultural case study field in the Northern Finland in terms of overall crop yield and the Crop Water Use Efficiency (CWUE). The modelling efforts with necessary calibration showed improved overall model performance to predict the crop yield that was measured using R2 and RMSE. The result of crop yield and CWUE was observed as improved on an average from 0.38 & 1.04 tons/ha to 0.92 & 0.40 tons/ha respectively for Scenario1 and from 0.38 & 1.04 tons/ha to 0.92 & 0.40 tons/ha respectively for Scenario2. 22 years average Crop water use efficiency (CWUE) for Scenario1 was observed on an overage of 2.25 kg/m3 and for Scenario2 was on an overage of 2.28 kg/m3. The field-scale vadose zone-based soil water balances were computed through the implementation of the finite difference techniques to govern the soil water fluxes and the equations governing the steady-state groundwater table management. Comprehensive in-situ data collected during 2021 -2022 cropping season and processed using machine learning techniques like multi-linear regression to predict the missing datasets demonstrated the application of hybrid modelling techniques in which process-based modelling blended with machine learning techniques for agricultural water resources management. The volumetric water content (m3.m-3) simulated through combined model approach showed satisfactory results when compared with in-situ datasets with an accuracy of (RMSE) 0.038 and 0.023. This approach also simulated water depth inside agricultural drainage control structure (ADCS) of SSDS with CD approach, and estimation about the total daily controlled discharge from ADCS. The study finally discussed a tri-modular smart, and pro-active decision support system (DSS) that integrates a comprehensive database module required to assess current and future condition of weather, field, and crop development; a data integration and analysis module to collect different datasets, analyse collected dataset using machine learning techniques and process-based numerical techniques; a decision support module to communicate with the user about different operations related to SSDS with CD approach. A DSS which aims to deliver sustainable development goals (SDGs) and relevant initiatives for Nordic agriculture associated with the state of water, agriculture and the environment in multiple ways.
How to cite: Ghag, K. S., Liedes, T., Klöve, B., and Torabi Haghighi, A.: Field-scale assessment of crop yield, crop water use efficiency, and water balances using different techniques to devise an ICT-based decision support solution for sub-surface drainage system with controlled drainage approach in Nordic agriculture, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21611, https://doi.org/10.5194/egusphere-egu26-21611, 2026.
The development of transparent and trustworthy hydrological modes requires careful attention to the choices made throughout the modeling process, including the assumptions made about the model structure. However, the structures of conceptual models are often defined indirectly through equations and the code implementation. This limits reproducibility, comparability and transparency of its structural assumptions. Moreover, the is currently no consistent unifying framework that represents the multitude of conceptual model structures found in hydrological modeling literature, making systematic analysis and comparison difficult.
We introduce a graph-theory based framework for explicitly and coherently representing conceptual model structures, where model compartments are represented as nodes and fluxes as directed edges. This allows mode structures to be defined independently of specific process formulations, while mass balance equations are derived directly from the graphs topology, ensuring consistent balances across all model compartments.
Representing conceptual model structures in an algebraic graph form, also allows to compare, analyze and manipulate in a ways that is difficult to achieve with pure equation-based representations. Graph and matrix encoding allows us to enumerate, compare and modify model structures in a controlled way, enforcing explicit constraints like hydrological plausibility, connectivity and closure. This representation forms a theoretical foundation for flexible and multi-model hydrological frameworks, allowing for the construction, testing and communication of different model hypotheses in a consistent way. Additionally, the graph-based representation support harmonious and unambiguous visual depictions of conceptual model structures, strengthening communication of modelling assumptions alongside their mathematical formulation.
Using examples of watershed models, we illustrate how conceptual models correspond to specific graph and matrix configurations and how structural differences are reflected in the resulting system of ordinary differential equations. In particular, we show that the incidence matrix provides a direct algebraic mapping between hydrologic model structure and the governing system of ordinary differential equations, where state derivatives are obtained as the balance of incoming and outgoing fluxes associated with each node. Moreover, we demonstrate how the graph–matrix representation can be used to systematically sample a space of candidate model structures by permuting adjacency matrices under predefined structural constraints. Invalid or implausible structures are excluded through rule-based filtering, yielding a structured yet unconstrained exploration of the admissible model space.
How to cite: Wischnewski, J., Schütze, N., and Wöhling, T.: A graph-theoretic framework for the systematic representation and generation of conceptual hydrological model structures, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9368, https://doi.org/10.5194/egusphere-egu26-9368, 2026.
Posters virtual: Thu, 7 May, 14:00–18:00 | vPoster spot A
EGU26-8171 | ECS | Posters virtual | VPS10
Hydrological Modelling Framework for Large-Scale Catchments using triangular nonhomogeneous spatial discretizationThu, 07 May, 14:00–14:03 (CEST) vPoster spot A
Abstract:
In this work, we develop a hydrological model designed to simulate the water balance and runoff processes at the catchment-scale. Instead of using a rectangular grid discretization, the model represents the catchment using a non-homogeneous two-dimensional triangular mesh framework (similar to a triangular mesh in the Finite Element method). This discretization fits a more flexible representation of complex topography and land boundaries. The model is implemented in the Fortran programming language. It depends on the Digital Elevation Model (DEM) to extract the flow pathways starting from upstream and reaching downstream. That guarantees a physically consistent and explicit flow-routing structure across the triangular mesh.
Evapotranspiration is calculated using the Penman–Monteith equation, as the parameters are considered to suit coastal climate conditions. The model utilizes temperature, solar radiation, wind speed, and vapor pressure as atmospheric inputs. The SCS Curve Number method is used to estimate the surface runoff, considering slope, land cover, and soil properties. Meteorological data measurements, including precipitation, temperature, humidity, as well as inflow and outflow discharges, are integrated into the simulations.
Due to its efficient numerical structure, the model supports simulations with numerous spatial elements and long time series while maintaining the computational cost at its lowest limits. This makes it well-suited for large-scale watershed applications and provides a strong basis for future high-performance computing developments.
Keywords: hydrological modeling, watershed triangulation, flow routing, numerical simulation
How to cite: Dali, N.: Hydrological Modelling Framework for Large-Scale Catchments using triangular nonhomogeneous spatial discretization, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8171, https://doi.org/10.5194/egusphere-egu26-8171, 2026.
