HS3.3 | Deep Learning in Hydrology
Deep Learning in Hydrology
Co-organized by ESSI1/NP4
Convener: Frederik Kratzert | Co-conveners: Basil Kraft, Daniel Klotz, Martin Gauch, Shijie Jiang
Orals
| Mon, 24 Apr, 16:15–18:00 (CEST)
 
Room 3.29/30, Tue, 25 Apr, 10:45–12:30 (CEST)
 
Room 3.29/30
Posters on site
| Attendance Tue, 25 Apr, 08:30–10:15 (CEST)
 
Hall A
Posters virtual
| Attendance Tue, 25 Apr, 08:30–10:15 (CEST)
 
vHall HS
Orals |
Mon, 16:15
Tue, 08:30
Tue, 08:30
Deep Learning has seen accelerated adoption across Hydrology and the broader Earth Sciences. This session highlights the continued integration of deep learning and its many variants into traditional and emerging hydrology-related workflows. Abstracts are solicited related to novel theory development, new methodologies, or practical applications of deep learning in hydrological modeling and process understanding. This might include, but is not limited to, the following:

(1) Development of novel deep learning models or modeling workflows.
(2) Integrating deep learning with process-based models and/or physical understanding.
(3) Improving understanding of the (internal) states/representations of deep learning models.
(4) Understanding the reliability of deep learning, e.g., under non-stationarity.
(5) Deriving scaling relationships or process-related insights with deep learning.
(6) Modeling human behavior and impacts on the hydrological cycle.
(7) Extreme event analysis, detection, and mitigation.
(8) Natural Language Processing in support of models and/or modeling workflows.

Orals: Mon, 24 Apr | Room 3.29/30

Chairpersons: Frederik Kratzert, Daniel Klotz, Martin Gauch
16:15–16:20
16:20–16:40
|
EGU23-4179
|
solicited
|
On-site presentation
David Lambl, Mostafa Elkurdy, Phil Butcher, Laura K Read, and Alden Keefe Sampson

Producing accurate hourly streamflow forecasts in large basins is difficult without a distributed model to represent both streamflow routing through the river network and the spatial heterogeneity of land and weather conditions. HydroForecast is a theory-guided deep learning flow forecasting product that consists of short-term (hourly predictions out to 10 days), seasonal (10 day predictions out to a year), and daily reanalysis models. This work focuses primarily on the short-term model which has award winning accuracy across a wide range of basins.

In this work, we discuss the implementation of a novel distributed flow forecasting capability of HydroForecast, which splits basins into smaller sub-basins and routes flows from each subbasin to the downstream forecast points of interest. The entire model is implemented as a deep neural network allowing end-to-end training of both sub-basin runoff prediction and flow routing. The model's routing component predicts a unit hydrograph of flow travel time at each river reach and timestep allowing us to inspect and interpret the learned river routing and to seamlessly incorporate any upstream gauge data. 

We compare the accuracy of this distributed model to our original flow forecasting model at selected sites and discuss future improvements that will be made to this model.

How to cite: Lambl, D., Elkurdy, M., Butcher, P., Read, L. K., and Sampson, A. K.: Improving Data-Driven Flow Forecasting in Large Basins using Machine Learning to Route Flows, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4179, https://doi.org/10.5194/egusphere-egu23-4179, 2023.

16:40–16:50
|
EGU23-3125
|
On-site presentation
Alexander Ley, Helge Bormann, and Markus Casper

Machine Learning and Deep Learning have been proving their potential for streamflow modelling in various studies. In particular, long short-term memory (LSTM) models showed exceptionally good results. However, machine learning models often are considered “black boxes” with limited interpretability. Explainable artificial intelligence (XAI) comprise methods that analyze the internal processes of the machine learning network and allow to have a glance in the “black box”. Most proposed XAI techniques are designed for the analysis of images, and there is currently only limited work on time series data available.

In our study, we applied various XAI algorithms including gradient-based methods (Saliency, InputXGradient, Integrated Gradient, GradientSHAP) but also perturbation-based methods (Feature Ablation, Feature Permutation) to compare their applicability for reasonable interpretation in the hydrological context. To our knowledge, only Integrated Gradient has been applied to a LSTM in hydrology so far. Gradient-based methods analyze the gradient of the output with respect to the input feature. Whereas perturbation-based methods gain information by altering or masking specific input features. The different methods were applied to a LSTM trained for the low-land Ems catchment in Germany, which has a major baseflow share of total streamflow.

We analyzed the results regarding their “timestep of influence”, which describes the amount of past days having importance for the prediction of streamflow at a particular day. All of the algorithms applied result in a comparable annual pattern, characterized by relatively small timesteps of influence in spring (wet season) and increasing timesteps of influence in summer and autumn (dry season). However, the range of the absolute days of attribution varies between the methods. In conclusion, all methods produces reasonable results and appear to be suitable for interpretation purposes.

Furthermore, we compare the results to ERA-5 reanalysis data and gained evidence that the LSTM recognizes soil water storage as the main driver for streamflow generation in the catchment: we found an inverse seasonality of soil moisture and timestep of influence.

How to cite: Ley, A., Bormann, H., and Casper, M.: Exploring different explainable artificial intelligence algorithms applied to a LSTM for streamflow modelling, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-3125, https://doi.org/10.5194/egusphere-egu23-3125, 2023.

16:50–17:00
|
EGU23-16658
|
ECS
|
Virtual presentation
Tadd Bindas, Wen-Ping Tsai, Jiangtao Liu, Farshid Rahmani, Dapeng Feng, Yuchen Bian, Kathryn Lawson, and Chaopeng Shen

Differentiable modeling has been introduced recently as a method to learn relationships from a combination of data and structural priors. This method uses end-to-end gradient tracking inside a process-based model to tune internal states and parameters along with neural networks, allowing us to learn underlying processes and spatial patterns. Hydrologic routing modules are typically needed to simulate flows in stem rivers downstream of large, heterogeneous basins, but obtaining suitable parameterization for them has previously been difficult. In this work, we apply differentiable modeling in the scope of streamflow prediction by coupling a physically-based routing model (which computes flow velocity and discharge in the river network given upstream inflow conditions) to neural networks which provide parameterizations for Manning’s river roughness parameter (n). This method consists of an embedded Neural Network (NN), which uses (imperfect) DL-simulated runoffs and reach-scale attributes as forcings and inputs, respectively, entered into the Muskingum-Cunge method and trained solely on downstream discharge. Our initial results show that while we cannot identify channel geometries, we can learn a parameterization scheme for roughness that follows observed n trends. Training on a short sample of observed data showed that we could obtain highly accurate routing results for the training and inner, untrained gages. This general framework can be applied to small and large scales to learn channel roughness and predict streamflow with heightened interpretability. 

 

How to cite: Bindas, T., Tsai, W.-P., Liu, J., Rahmani, F., Feng, D., Bian, Y., Lawson, K., and Shen, C.: Improving large-basin streamflow simulation using a modular, differentiable, learnable graph model for routing, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16658, https://doi.org/10.5194/egusphere-egu23-16658, 2023.

17:00–17:10
|
EGU23-5736
|
ECS
|
Virtual presentation
Rendani Mbuvha, Peniel Julien Yise Adounkpe, Mandela Coovi Mahuwetin Houngnibo, and Nathaniel Newlands

Streamflow predictions are a vital tool for detecting flood and drought events. Such predictions are even more critical to Sub-Saraharan African regions that are vulnerable to the increasing frequency and intensity of such events. These regions are sparsely gauged, with few available gauging stations that are often plagued with missing data due to various causes, such as harsh environmental conditions and constrained operational resources. 

This work presents a novel workflow for predicting streamflow in the presence of missing gauge observations. We leverage bias correction of the GEOGloWS ECMWF streamflow service (GESS) forecasts for missing data imputation and predict future streamflow using the state-of-the-art Temporal Fusion transformers at ten river gauging stations in the Benin Republic.

We show by simulating missingness in a testing period that GESS forecasts have a significant bias that results in poor imputation performance over the ten Beninese stations. Our findings suggest that overall bias correction by Elastic Net and Gaussian Process regression achieves superior performance relative to traditional imputation by established methods such as Random Forest, k-Nearest Neighbour, and GESS lookup. We also show that the Temporal Fusion Transformer yields high predictive skill and further provides explanations for predictions through the weights of its attention mechanism. The findings of this work provide a basis for integrating Global streamflow prediction model data and state-of-the-art machine learning models into operational early-warning decision-making systems (e.g., flood/ drought alerts) in resource-constrained countries vulnerable to drought and flooding due to extreme weather events.

How to cite: Mbuvha, R., Adounkpe, P. J. Y., Houngnibo, M. C. M., and Newlands, N.: A Novel Workflow for Streamflow Prediction in the Presence of Missing Gauge Observations, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5736, https://doi.org/10.5194/egusphere-egu23-5736, 2023.

17:10–17:20
|
EGU23-6466
|
ECS
|
On-site presentation
Marvin Höge, Andreas Scheidegger, Marco Baity-Jesi, Carlo Albert, and Fabrizio Fenicia

Neural Ordinary Differential Equation (ODE) models have demonstrated high potential in providing accurate hydrologic predictions and process understanding for single catchments (Höge et al., 2022). Neural ODEs fuse a neural network model core with a mechanistic equation framework. This hybrid structure offers both traceability of model states and processes, like in conceptual hydrologic models, and the high flexibility of machine learning to learn and refine model interrelations. Aside of the functional dependence of internal processes on driving forces, like of evapotranspiration on temperature, Neural ODEs are also able to learn the effect of catchment-specific attributes, e.g. land cover types, on processes when being trained over multiple basins simultaneously.

 

We demonstrate the performance of a generic Neural ODE architecture in a hydrologic large-sample setup with respect to both predictive accuracy and process interpretability. Using several hundred catchments, we show the capability of Neural ODEs to learn the general interplay of catchment-specific attributes and hydrologic drivers in order to predict discharge in out-of-sample basins. Further, we show how functional relations learned (encoded) by the neural network can be translated (decoded) into an interpretable form, and how this can be used to foster understanding of processes and the hydrologic system.

 

Höge, M., Scheidegger, A., Baity-Jesi, M., Albert, C., & Fenicia, F.: Improving hydrologic models for predictions and process understanding using Neural ODEs. Hydrol. Earth Syst. Sci., 26, 5085-5102, https://hess.copernicus.org/articles/26/5085/2022/

How to cite: Höge, M., Scheidegger, A., Baity-Jesi, M., Albert, C., and Fenicia, F.: Neural ODE Models in Large-Sample Hydrology, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-6466, https://doi.org/10.5194/egusphere-egu23-6466, 2023.

17:20–17:30
|
EGU23-16947
|
ECS
|
Virtual presentation
Dapeng Feng and Chaopeng Shen

Although deep learning (DL) models have shown extraordinary performance in hydrologic modeling, they are still hard to interpret and not able to predict untrained hydrologic variables due to lacking physical meanings and constraints. This study established hybrid differentiable models (namely the delta models) with regionalized parameterization and learnable structures based on a DL-based differentiable parameter learning (dPL) framework. The simulation experiments on both US and global basins demonstrate that the delta models can approach the performance of the state-of-the-art long short-term memory (LSTM) network on discharge prediction. Different from the pure data-driven LSTM model, the delta models can output a full set of hydrologic variables not used as training targets. The evaluation with independent data sources showed that the delta models, only trained on discharge observations, can also give decent predictions for ET and baseflow. The spatial extrapolation experiments showed that the delta models can surpass the performance of the LSTM model for predictions in large ungauged regions in terms of the daily hydrographic metrics and multi-year trend prediction. The spatial patterns of the parameters learned by the delta models remain remarkably stable from the in-sample to spatial out-of-sample predictions, which explains the robustness of the delta models for spatial extrapolation. More importantly, the proposed modeling framework enables directly learning new relations between intermediate variables from large observations. This study shows that the model performance and physical meanings can be balanced with the differentiable modeling approach which is promising to large-scale hydrologic prediction and knowledge discovery.

How to cite: Feng, D. and Shen, C.: A differentiable modeling approach to systematically integrating deep learning and physical models for large-scale hydrologic prediction and knowledge discovery, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16947, https://doi.org/10.5194/egusphere-egu23-16947, 2023.

17:30–17:40
|
EGU23-14399
|
ECS
|
On-site presentation
Christina Lott, Leonardo Martins, Jonas Weiss, Thomas Brunschwiler, and Peter Molnar

Simulation of the catchment rainfall-runoff transformation with physically based watershed models is a traditional way to predict streamflow and other hydrological variables at catchment scales. However, the calibration of such models requires large data inputs and computational power and contains many parameters which are often impossible to constrain or validate. An alternative approach is to use data-driven machine learning for streamflow prediction.

In the past few years, LSTM (long short-term memory) models and its variants have been explored in rainfall-runoff modelling. Typical applications use daily climate variables as inputs and model the rainfall-runoff transformation processes with different timescales of memory. This is especially useful as delays in runoff production by snow accumulation and melt, soil water storage, evapotranspiration, etc., can be included. In contrast to feed-forward ANNs (artificial neural networks), LSTMs are capable of maintaining the sequential temporal order of inputs, and compared to RNNs (recurrent neural networks), of learning the long-term dependencies. [1]

However, current work on LSTMs mostly focuses on the USA, the UK and Brazil, where CAMELS datasets are available [1, 2, 3]. Catchments at higher altitudes with snow-driven dynamics and sometimes glaciers are present in small number in these datasets (if at all). Systematic applications of LSTMs for streamflow prediction in climates where a significant part of the catchments are snow and ice dominated are missing. In this work, an FS-LSTM (fast slow-LSTM) previously applied in Brazil is adapted for Swiss catchments to fill this gap [3]. The FS-LSTM explored builds on the work of Hoedt et al. (2021) that imposed mass constraints on an LSTM, called MC-LSTM [4]. FS-LSTM adds a fast and slow part for streamflow, containing rainfall and soil moisture respectively. We will discuss benchmark results against an existing semi-distributed conceptual model widely used in Switzerland for streamflow simulation [5].

 

References:

[1]: Kratzert et al., Rainfall-runoff modelling using Long Short-Term Memory (LSTM) networks, 2018.

[2]: Lees et al., Hydrological concept formation inside long short-term memory (LSTM) networks, 2022.

[3]: Quinones et al., Fast-Slow Streamflow Model Using Mass-Conserving LSTM, 2021.

[4]: Hoedt et al., MC-LSTM: Mass-Conserving LSTM, 2021.

[5]: Viviroli et al., An introduction to the hydrological modelling system PREVAH and its pre- and post-processing-tools, 2009.

How to cite: Lott, C., Martins, L., Weiss, J., Brunschwiler, T., and Molnar, P.: LSTMs for Hydrological Modelling in Swiss Catchments, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-14399, https://doi.org/10.5194/egusphere-egu23-14399, 2023.

17:40–17:50
|
EGU23-5044
|
Virtual presentation
Dagang Wang

The objective function plays an important role in the training process for deep learning models, since it largely determines the trained values of the model parameters and influences the model performance. In this study, we establish two application-orientated objective functions, namely high flow balance error (HFBE) and transformed mean absolute percentage error (MAPE*), for the forecasts of high flows and low flows, respectively, in the LSTM model. We examine the strength and weakness of these streamflow forecast models trained on HFBE, MAPE* and mean square error (MSE) based on multiple performance metrics. Furthermore, we propose the objective function-based ensemble model (OEM) framework that integrates the models trained on different objective functions, so as to take advantages of the trained models focusing on different aspects of streamflow and thus achieve a better overall performance. Our results in 273 catchments over USA show that the models trained on HFBE can alleviate underestimation in high flows existing in the models trained on MSE, and perform remarkably better for high flows. It is also found that the models trained on MAPE* outperform the other two models in low flow forecast, no matter what algorithm is used for the model establishment. By incorporating the three models trained on HFBE, MAPE* and MSE, respectively, our proposed OEM performs well in the forecasts of both high flows and low flows, and realistically capture the mean and the variability of the observational streamflow under different scenarios under a variety of hydrometeorological conditions. This study highlights the necessity of applying application-orientated objective functions for given projects and the great potential of the ensemble learning methods for multi-optimization in hydrological modeling.

How to cite: Wang, D.: The role of ensemble learning in multi-optimization for streamflow prediction, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5044, https://doi.org/10.5194/egusphere-egu23-5044, 2023.

17:50–18:00
|
EGU23-16974
|
On-site presentation
Grey Nearing, Martin Gauch, Daniel Klotz, Frederik Kratzert, Asher Metzger, Guy Shalev, Shlomo Shenzis, Tadele Tekalign, Dana Weitzner, and Oren Gilon

Deep learning has become the de facto standard for streamflow simulation. While there are examples of deep learning based streamflow forecast models (e.g., 1-5), the majority of the development and research has been done with hindcast models. The primary challenge in using deep learning models for forecasting (e.g., flood forecasting) is that the meteorological input data are drawn from different distributions in hindcast vs. forecast. The (relatively small) amount of research that has been done on deep learning streamflow forecasting has largely used an encoder-decoder approach to account for forecast distribution shifts. This is, for example, what Google’s operational flood forecasting model uses [4]. 

In this work we show that the encoder-decoder approach results in artifacts in forecast trajectories that are not detectable with standard hydrological metrics, but which can cause forecasts to have incorrect trends (e.g., rising when they should be falling and vice-versa).  We solve this problem using regularized embeddings, which remove forecast artifacts without harming overall accuracy. 

Perhaps more importantly, input embeddings allow for training models on spatially and/or temporally incomplete meteorological inputs, meaning that a single model can be trained using input data that does not exist everywhere or does not exist during the entire training or forecast period. This allows models to learn from a significantly larger training data set, which is important for high-accuracy predictions. It also allows large (e.g., global) models to learn from local weather data. We demonstrate how and why this is critical for state-of-the-art global-scale streamflow forecasting. 

 

  • Franken, Tim, et al. An operational framework for data driven low flow forecasts in Flanders. No. EGU22-6191. Copernicus Meetings, 2022.
  • Kao, I-Feng, et al. "Exploring a Long Short-Term Memory based Encoder-Decoder framework for multi-step-ahead flood forecasting." Journal of Hydrology 583 (2020): 124631.
  • Liu, Darong, et al. "Streamflow prediction using deep learning neural network: case study of Yangtze River." IEEE access 8 (2020): 90069-90086.
  • Nevo, Sella, et al. "Flood forecasting with machine learning models in an operational framework." Hydrology and Earth System Sciences 26.15 (2022): 4013-4032.
  • Girihagama, Lakshika, et al. "Streamflow modelling and forecasting for Canadian watersheds using LSTM networks with attention mechanism." Neural Computing and Applications 34.22 (2022): 19995-20015.

 

How to cite: Nearing, G., Gauch, M., Klotz, D., Kratzert, F., Metzger, A., Shalev, G., Shenzis, S., Tekalign, T., Weitzner, D., and Gilon, O.: From Hindcast to Forecast with Deep Learning Streamflow Models, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16974, https://doi.org/10.5194/egusphere-egu23-16974, 2023.

Orals: Tue, 25 Apr | Room 3.29/30

Chairpersons: Basil Kraft, Shijie Jiang, Martin Gauch
10:45–10:50
10:50–11:00
|
EGU23-1526
|
ECS
|
On-site presentation
Ralf Loritz and Hoshin Gupta

Neural networks belong to the best available methods for numerous hydrological model challenges. However, although they have shown to outperform classical hydrological models in several applications there is still some doubt whether neural networks are, despite their excellent interpolation skills, capable to make predictions beyond the support of the training data. This study addresses this issue and proposes an approach to infer the ability of neural network to predict unusual, extreme system states. We show how we can use the concept of data surprise and model surprise in a complementary manner to assess which unusual events a neural network can predict, which it can predict but only with additionally data and which it cannot predict at all hinting toward the wrong model choice or towards an incomplete description of the data.

How to cite: Loritz, R. and Gupta, H.: Extrapo… what? Predictions beyond the support of the training data, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1526, https://doi.org/10.5194/egusphere-egu23-1526, 2023.

11:00–11:10
|
EGU23-14631
|
ECS
|
On-site presentation
Xinqi Hu, Ye Tuo, and Markus Disse

Improving the understanding of processes is vital to hydrological modeling. One key challenge is how to extract interpretable information that can describe the complex hydrological system from the growing number of observation data to advance our understanding of processes and modeling. To address the problem, we propose a data-driven framework to discover coordinate transformation, which transfers original observations to a reduced-dimension system. The framework combines deep learning method with sparse regression to approximate the specific hydrological process: deep learning methods have a rich representation to promote generalization, and sparse regression can sparsely identify parsimonious models to promote interpretability. By doing so, we can identify the essential latent variables under a physically meaning-wise coordinate system where the hydrological processes are linearly and sparsity represented to capture the behavior of the system from observations. To demonstrate the framework, we focus on the evaporation process. The relationships between potential evaporation and climate variables including long/short wave radiation, air temperature, air pressure, relative humidity, and wind speed are quantified. The connection between the climate variables and coordinates components extracted are evaluated to capture the pattern of climate variables in the component space. The robustness and statistical stability of the framework is examined based on distributed observations from FluxNet towers over North America. The resulting modeling framework shows the potential of deep learning methods for improving our knowledge of the hydrological system.

How to cite: Hu, X., Tuo, Y., and Disse, M.: Deep learning based coordinates transformations for improving process understanding in hydrological modeling system, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-14631, https://doi.org/10.5194/egusphere-egu23-14631, 2023.

11:10–11:20
|
EGU23-4842
|
ECS
|
On-site presentation
|
Bram Droppers, Myrthe Leijnse, Marc F.P. Bierkens, and Niko Wanders

Process-based global hydrological models are an important tool for sustainable development and policy making in today’s water-scarce world. These models are able to inform national to regional scale water management with basin-scale accounting of water availability and demand and project the impacts of climate change and adaptation on water resources. However, the increasing need for better and higher resolution hydrological information is proving difficult for these state-of-the-art process-based models as the associated computational requirements are significant.

Recently, the deep-learning community has shown that neural networks (in particular the LSTM network) can provide hydrological information with an accuracy that rivals, if not exceeds, that of process-based hydrological models. Although the training of these neural networks takes time, prediction is fast compared to process-based simulations. Nevertheless, training is mostly done on historical observations and thus projections under climate change and adaptation are uncertain.

Inspired by the complementary strengths and weaknesses of the process-based and deep-learning approaches, we present DL-GLOBWB: a deep-learning surrogate of the state-of-the-art PCR-GLOBWB global hydrological model. DL-GLOBWB predicts all water-balance components from the process-based model, including human water demand and abstraction, with a nRSME of 0.05 (range between 0.0001 and 0.32). The DL-GLOBWB surrogate is orders of magnitudes faster than its process-based counterpart, especially as surrogates trained at low resolutions (e.g. 30 arc-minute) can effectively be downscaled to higher resolutions (e.g. 5 arc-minute).

In addition to introducing DL-GLOBWB, our presentation will explore future applications of this deep-learning surrogate, such as (1) improving model calibration and performance by comparing DL-GLOBWB outputs with ins-situ data and satellite observations; (2) training DL-GLOBWB on  future model projections to include global change; and (3) the implementation of DL-GLOBWB to dynamically, and at high resolution, visualize the impact of climate change and adaptation to stakeholders.

How to cite: Droppers, B., Leijnse, M., Bierkens, M. F. P., and Wanders, N.: Introducing DL-GLOBWB: a deep-learning surrogate of a process-based global hydrological model, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4842, https://doi.org/10.5194/egusphere-egu23-4842, 2023.

11:20–11:30
|
EGU23-12952
|
ECS
|
On-site presentation
Roberto Bentivoglio, Elvin Isufi, Sebastian Nicolaas Jonkman, and Riccardo Taormina

The high computational cost of detailed numerical models for flood simulation hinders their use in real-time and limits uncertainty quantification. Deep-learning surrogates have thus emerged as an alternative to speed up simulations. However, most surrogate models currently work only for a single topography, meaning that they need to be retrained for different case studies, ultimately defeating their purpose. In this work, we propose a graph neural network (GNN) inspired by the shallow water equations used in flood modeling, that can generalize the spatio-temporal prediction of floods over unseen topographies. The proposed model works similarly to finite volume methods by propagating the flooding in space and time, given initial and boundary conditions. Following the Courant-Friedrichs-Lewy condition, we link the time step between consecutive predictions to the number of GNN layers employed in the model. We analyze the model's performance on a dataset of numerical simulations of river dike breach floods, with varying topographies and breach locations. The results suggest that the GNN-based surrogate can produce high-fidelity spatio-temporal predictions, for unseen topographies, unseen breach locations, and larger domain areas with respect to the training ones, while reducing computational times.

How to cite: Bentivoglio, R., Isufi, E., Jonkman, S. N., and Taormina, R.: On the generalization of hydraulic-inspired graph neural networks for spatio-temporal flood simulations, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12952, https://doi.org/10.5194/egusphere-egu23-12952, 2023.

11:30–11:40
|
EGU23-1278
|
ECS
|
Virtual presentation
Nikunj K. Mangukiya and Ashutosh Sharma

Accurate flood frequency analysis is essential for developing effective flood management strategies and designing flood protection infrastructure, but it is challenging due to the complex, nonlinear hydrological system. In regional flood frequency analysis (RFFA), the flood quantiles at ungauged sites can be estimated by establishing a relationship between interdependent physio-meteorological variables and observed flood quantiles at gauge sites in the region. However, this regional approach implies a loss of information due to the prior aggregation of hydrological data at gauged locations and can be difficult for data-sparse regions due to limited data. In this study, we evaluated an alternate approach or path for RFFA in two case studies: a data-sparse region in India and a data-dense region in the USA. In this approach, daily streamflow is predicted first using a deep learning-based hydrological model, and then flood quantiles are estimated from the predicted daily streamflow using statistical methods. We compared the results obtained using this alternate approach to those from the traditional RFFA technique, which used the Random Forest (RF) and eXtreme Gradient Boosting (XGB) algorithms to model the nonlinear relationship between flood quantiles and relevant physio-meteorological predictor variables such as meteorological forcings, topography, land use, and soil properties. The results showed that the alternate approach produces more reliable results with the least mean absolute error and higher coefficient of determination in the data-sparse region. In the data-dense region, both traditional and alternate approaches produced comparable results. However, the alternate approach has the advantage of being flexible and providing the complete time series of daily flow at the ungauged location, which can be used to estimate other flow characteristics, develop flow duration curves, or estimate flood quantiles of any return period without creating a separate traditional RFFA model. This study shows that the alternate approach can provide accurate flood frequency estimates in data-sparse regions, offering a promising solution for flood management in these areas.

How to cite: Mangukiya, N. K. and Sharma, A.: Evaluating Machine Learning Approach for Regional Flood Frequency Analysis in Data-sparse Regions, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1278, https://doi.org/10.5194/egusphere-egu23-1278, 2023.

11:40–11:50
|
EGU23-4137
|
ECS
|
On-site presentation
Robin Thibaut, Ty Ferré, Eric Laloy, and Thomas Hermans
The groundwater-surface water (GW-SW) exchange fluxes are driven by a complex interplay of subsurface processes and their interactions with surface hydrology, which have a significant impact on the water and contaminant exchanges. Due to the complexity of these systems, the accurate estimation of GW-SW fluxes is important for quantitative hydrological studies and should be based on relevant data and careful experimental design. Therefore, the effective design of monitoring networks that can identify relevant subsurface information are essential for the optimal protection of our water resources. In this study, we present novel deep learning (DL)-driven approaches for sequential and static Bayesian optimal experimental design (BOED) in the subsurface, with the goal of estimating the GW-SW exchange fluxes from a set of temperature measurements. We apply probabilistic Bayesian neural networks (PBNN) to conditional density estimation (CDE) within a BOED framework, and the predictive performance of the PBNN-based CDE model is evaluated by a custom objective function based on the Kullback-Leibler divergence to determine optimal temperature sensor locations utilizing the information gain provided by the measurements. This evaluation is used to determine the optimal sequential sampling strategy for estimating GW-SW exchange fluxes in the 1D case, and the results are compared to the static optimal sampling strategy for a 3D conceptual riverbed-aquifer model based on a real case study. Our results indicate that probabilistic DL is an effective method for estimating GW-SW fluxes from temperature data and designing efficient monitoring networks. Our proposed framework can be applied to other cases involving surface or subsurface monitoring and experimental design.

How to cite: Thibaut, R., Ferré, T., Laloy, E., and Hermans, T.: Sequential optimization of temperature measurements to estimate groundwater-surface water interactions, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4137, https://doi.org/10.5194/egusphere-egu23-4137, 2023.

11:50–12:00
|
EGU23-7347
|
ECS
|
On-site presentation
Mathilde de FLEURY, Laurent Kergoat, Martin Brandt, Rasmus Fensholt, Ankit Kariryaa, Gyula Mate Kovács, Stéphanie Horion, and Manuela Grippa

Inland surface water, especially lakes and small water bodies, are essential resources and have impacts on biodiversity, greenhouse gases and health. This is particularly true in the semi-arid Sahelian region, where these resources remain largely unassessed, and little is known about their number, size and quality. Remote sensing monitoring methods remain a promising tool to address these issues at the large scale, especially in areas where field data are scarce. Thanks to technological advances, current remote sensing systems provide data for regular monitoring over time and offer a high spatial resolution, up to 10 metres.  

Several water detection methods have been developed, many of them using spectral information to differentiate water surfaces from soil, through thresholding on water indices (MNDWI for example), or classifications by clustering. These methods are sensitive to optical reflectance variability and are not straight forwardly applicable to regions, such as the Sahel, where the lakes and their environment are very diverse. Particularly, the presence of aquatic vegetation is an important challenge and source of error for many of the existing algorithms and available databases.  

Deep learning, a subset of machine learning methods for training deep neural networks, has emerged as the state-of-the-art approach for a large number of remote sensing tasks. In this study, we apply a deep learning model based on the U-Net architecture to detect water bodies in the Sahel using Sentinel-2 MSI data, and 86 manually defined lake polygons as training data. This framework was originally developed for tree mapping (Brandt et al., 2020, https://doi.org/10.1038/s41586-020-2824-5).   

Our preliminary analysis indicate that our models achieve a good accuracy (98 %). The problems of aquatic vegetation do not appear anymore, and each lake is thus well delimited irrespective of water type and characteristics. Using the water delineations obtained, we then classify different optical water types and thereby highlight different type of waterbodies, that appear to be mostly turbid and eutrophic waters, allowing to better understand the eco-hydrological processes in this region.  

This method demonstrates the effectiveness of deep learning in detecting water surfaces in the study region. Deriving water masks that account for all kind of waterbodies offer a great opportunity to further characterize different water types. This method is easily reproducible due to the availability of the satellite data/algorithm and can be further applied to detect dams and other human-made features in relation to lake environments. 

How to cite: de FLEURY, M., Kergoat, L., Brandt, M., Fensholt, R., Kariryaa, A., Kovács, G. M., Horion, S., and Grippa, M.: Deep learning for mapping water bodies in the Sahel, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-7347, https://doi.org/10.5194/egusphere-egu23-7347, 2023.

12:00–12:10
|
EGU23-8218
|
ECS
|
Virtual presentation
Pragay Shourya Moudgil and G Srinivasa Rao

Terrestrial water storage (TWS) anomalies from Gravity Recovery and Climate Experiment (GRACE) and its follow on GRACE-FO satellite missions provide a unique opportunity to measure the impact of different climate extremes and human intervention on water use at regional and continental scales. However, temporal gaps within GRACE and GRACE-FO mission (GRACE: 20 months, between GRACE and GRACE-FO: 11 months and GRACE-FO: 2 months) pose difficulties in analyzing spatiotemporal variations in TWS. In this study, Convolutional Long Short-Term Memory Neural Networks (CNN-LSTM) model was developed to fill these gaps and reconstruct the TWS for the Indian subcontinent (April 2002-July 2022). Various meteorological and climatic variables, such as precipitation, temperature, run-off, evapotranspiration, and vegetation, have been integrated to predict GRACE TWS. The performance of the models was evaluated with the help of Pearson’s correlation coefficient (PR), Nash-Sutcliffe efficiency (NSE), and Normalised Root Mean Square Error (NRMSE). Results indicate that the CNN-LSTM model yielded a mean PR of 0.94 and 0.89, NSE of 0.87 and 0.8, and NRMSE of 0.075 and 0.101 on training and testing, respectively. Overall, the CNN-LSTM achieved good performance except in the northwestern region of India, which showed a relatively poor performance might be due to high anthropogenic activity and arid climatic conditions. Further reconstructed time series were used to study the Spatiotemporal variations of TWS over the Indian Subcontinent.

Keywords: GRACE; Deep Learning; TWSA; Indian subcontinent

How to cite: Moudgil, P. S. and Rao, G. S.: Filling Temporal Gaps within and between GRACE and GRACE-FO Terrestrial Water Storage Changes over Indian Sub-Continent using Deep Learning., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8218, https://doi.org/10.5194/egusphere-egu23-8218, 2023.

12:10–12:20
|
EGU23-15575
|
Virtual presentation
Hamidreza Mosaffa, Paolo Filippucci, Luca Ciabatta, Christian Massari, and Luca Brocca

Reliable and accurate precipitation estimations are a crucial hydrological parameter for various applications, including managing water resources, drought monitoring and natural hazard prediction. The two main approaches for estimating precipitation from satellite data are the top-down and bottom-up. The top-down approach uses data from Geostationary and Low Earth Orbiting satellites to infer precipitation from atmosphere and cloud information, while the bottom-up approach estimates precipitation using soil moisture observations, e.g.  the SM2RAIN algorithm. The main difference between these approaches is that the top-down approach is a more direct method of measuring precipitation that estimates it instantaneously, which may lead to underestimation, while the bottom-up approach measures accumulated rainfall with more reliable precipitation estimation between two consecutive SM measurements. In this study, we develop the deep convolutional neural networks (CNN) algorithm to combine the top-down and bottom-up approaches for estimating precipitation using the satellite level 1 products including the satellite backscatter information from the Advanced SCATterometer (ASCAT), infrared (IR) and water vapor (WV) channels from geostationary satellites. This algorithm is assessed at 0.1° spatial and daily temporal resolution over Italy for the period of 2019-2021. The results show that the developed model improves the accuracy of precipitation estimation. Additionally, it indicates that there is a significant potential for global precipitation estimation using this model.

How to cite: Mosaffa, H., Filippucci, P., Ciabatta, L., Massari, C., and Brocca, L.: Application of deep convolutional neural networks for precipitation estimation through both top-down and bottom-up approaches, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-15575, https://doi.org/10.5194/egusphere-egu23-15575, 2023.

12:20–12:30
|
EGU23-4887
|
On-site presentation
Ho Tin Hung and Li-Pen Wang

IMERG is a global satellite-based precipitation dataset, produced by NASA. It has provided valuable rainfall information to facilitate the design or the operation of the disaster and risk management worldwide. In operation, NASA offers three types of IMERG Level 3 (L3) products, with different levels of trade-offs in terms of time latency and accuracy. These are Early run (4-hour latency), Late run (14-hour latency) and Final run(3.5-month latency). The final-run product integrates multi-sensor retrievals and provides the highest-quality precipitation estimates among three IMERG products. It however suffers from a long processing latency, which hinders its applicability to near real-time applications. In the past 10 years, deep learning techniques have made significant breakthroughs in various scientific fields, including short-term rainfall forecasting. Deep learning models have shown to have the potential to learn the complex variations in weather systems and to outperform the Numerical Weather Prediction (NWP) in terms of short lead-time predictability and the required computational resources for operation.

 

In this research, we would like to explore the potential of deep learning (DL) in generating high-quality satellite-based precipitation product with low latency. More specifically, we investigate if DL models can learn the difference between Final- and Early-run products, and thus predict a Final-run-like product using Early-run product as input. Low-latency yet high-quality IMERG precipitation product can be therefore obtained. Various DL techniques are being tested in this work, including Auto-Encoder(AE), ConvLSTM and Deep Generative model. IMERG data between 2018 and 2020 over a rectangular area centred in the UK is used for model training and testing, and ground rain gauge records will be used to evaluate the performance of the original and predicted products. This pilot includes both ocean and land regions, which enables the comparison of the model performance between two different surface conditions. Preliminary analysis suggests that given patterns do exist in the differences between Early- and Final-run products, and the capacity of the selected DL models to learn the differences will be further investigated. The proposed work is of great potential to improve the applicability of IMERG products in an operational context.

How to cite: Hung, H. T. and Wang, L.-P.: IMERG Run Deep: Can we produce a low-latency IMERG Final run product with a deep learning based prediction model?, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4887, https://doi.org/10.5194/egusphere-egu23-4887, 2023.

Posters on site: Tue, 25 Apr, 08:30–10:15 | Hall A

Chairpersons: Daniel Klotz, Basil Kraft, Shijie Jiang
A.53
|
EGU23-245
Dongkyun Kim and Yongoh Lee

An LSTM-based distributed hydrologic model for an urban watershed of Korea was developed. The input of the model is the time series of the 10-minute radar-gauge composite rainfall data and 10-minute temperature data at the 239 model grid cells, and the output of the model is the 10-minute flow discharge at the watershed outlet. The Nash-Sutcliffe Efficiency (NSE) coefficients of the calibration period (2013-2016) and validation period (2017-2019) were 0.99 and 0.67, respectively. Normal events were better predicted than the extreme ones. Further in-depth analyses revealed that: (1) the model composes the watershed outlet flow discharge by linearly superimposing multiple time series created by each of the LSTM units. Unlike conventional hydrologic models, most of these time series greatly fluctuated in both positive and negative domain; (2) the runoff to rainfall ratio of each of the model grid cells does not reflect its counterpart parameters of the conceptual hydrologic models  revealing that the model simulates the watershed responses in a unique manner; (3) the model successfully reproduced the soil-moisture dependent runoff processes, which is an essential prerequisite of continuous hydrologic models; (4) Each of the LSTM units have different temporal sensitivity to a unit rainfall stimulus, and the LSTM units that is sensitive to rainfall input have greater output weight factors nearby the watershed outlet, and vice versa. This means that the model learned a mechanism to separately consider the hydrologic components with distinct response time such as direct runoff and the low frequency baseflow. 

Acknowledgement

This research was supported by the Basic Science Research Program (Grant Number: 2021R1A2C2003471) and the Basic Research Laboratory Program (Grant Number: 2022R1A4A3032838) through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT.

How to cite: Kim, D. and Lee, Y.: Machines simulate hydrologic processes using a simple structure but in a unique manner – a case study of predicting fine scale watershed response on a distributed framework, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-245, https://doi.org/10.5194/egusphere-egu23-245, 2023.

A.54
|
EGU23-339
|
ECS
Jeonghun Lee and Dongkyun Kim

This study developed a distributed hydrologic model based on Long Short-Term Memory (LSTM) to predict flow discharge of Joongrang stream located in a highly urbanized area in Seoul, Korea. The model inputs are the time series of 10-minute radar-gauge composite precipitation data at 239 grid cells (1km2) in the watershed and the Normalized Difference Vegetation Index (NDVI) data derived from Landsat 8 images and the model output is the 10-minute flow discharge at the watershed outlet as output. The model was trained for the calibration period of 2013-2016 and was validated for the period of 2017-2019. The NSE value over the validation period corresponding to the optimal model architecture (256 LSTM hidden layers) with and without NDVI input data was 0.68 and 0.52, respectively, which suggests that the machine can learn dynamic processes of soil infiltration and plant interception from the remotely sensed information provided by satellite.

This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. NRF-2022R1A4A3032838). 

How to cite: Lee, J. and Kim, D.: Effectiveness of Satellite-based Vegetation Index for Simulating Watershed Response Using an LSTM-based model in a Distributed Framework, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-339, https://doi.org/10.5194/egusphere-egu23-339, 2023.

A.55
|
EGU23-1218
Ina Vertommen, Xin Tian, Tessa Pronk, Siddharth Seshan, Sotirios Paraskevopoulos, and Bas Wols

Natural Language Processing (NLP), empowered by the most recent developments in Deep Learning, demonstrates its potential effectiveness for handling texts. Urban water research  benefits from both subfields of NLP, namely, Natural Language Understanding (NLU) and Natural Language Generation (NLG). In this work, we present three recent studies that use NLP for: (1) automated processing and responding to registered customer complaint within Dutch water utilities, (2) automated collection of up-to-date water-related information from the Internet, (3) extraction of key information about chemical compounds and pathogen characteristics from scientific publications. These applications, using the latest NLP models and tools (e.g., Rasa, Spacy), take into account studies on both water quality and quantity for the water sector. According to our findings, NLU and rule-based text mining are effective in extracting information from unstructured texts. In addition, NLU and NLG can be integrated to build a human-computer interface, such as a value-based Chabot to understand and address the demands made by customers of water utilities.

How to cite: Vertommen, I., Tian, X., Pronk, T., Seshan, S., Paraskevopoulos, S., and Wols, B.: Exploring the Value of Natural Language Processing for Urban Water Research, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1218, https://doi.org/10.5194/egusphere-egu23-1218, 2023.

A.56
|
EGU23-4801
Bhanu Magotra, Manabendra Saharia, and Chandrika Thulaseedharan Dhanya

Streamflow modelling plays a critical role in water resource management activities. The “physically based" models require high computation resources and large amounts of input meteorological data which results in high operating costs and longer running times. On the other hand, with advancements in deep-learning techniques, data-driven models such as long short-term memory (LSTM) networks have been shown to successfully model non-linear rainfall-runoff relationships through historically observed data at a fraction of computation cost. Moreover, using physics-informed machine learning techniques, the physical consistency of data-driven models can be further improved. In this study, one such method is applied where we trained a physics-informed LSTM network model over 278 Indian catchments to simulate streamflow at a daily timestep using historically observed precipitation and streamflow data. The ancillary data included meteorological forcings, static catchment attributes, and Noah-MP simulated land surface states and fluxes such as soil moisture, latent heat, and total evapotranspiration. The LSTM model's performance was evaluated using error metrics such as Nash-Sutcliffe Efficiency (NSE), Kling-Gupta Efficiency (KGE) and its components, along with skill scores based on 2x2 contingency matrix for hydrological extremes. The trained LSTM model shows improved performance in simulating streamflow over the catchments compared to the physically based model. This will be the first study over India to generate reliable streamflow simulations using a hybrid state-of-the-art approach, which will be beneficial to policy makers for effective water resource management in India. 

How to cite: Magotra, B., Saharia, M., and Dhanya, C. T.: Improving Streamflow Predictions over Indian Catchments using Long Short Term Memory Networks, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4801, https://doi.org/10.5194/egusphere-egu23-4801, 2023.

A.57
|
EGU23-4970
|
ECS
|
Imad Janbain, Julien Deloffre, Abderrahim Jardani, Minh Tan Vu, and Nicolas Massei

Missing data is the first major problem that appears in many database fields for a set of reasons. It has always been necessary to fill them, which becomes unavoidable and more complicated when the missing periods are longer. Several machine-learning-based approaches have been introduced to deal with this problem. 

The purpose of this paper is to discuss the effectiveness of a new methodology added prior to the LSTM deep learning algorithm to fill in the missing data in the hourly surface water level time series of some stations installed along the Seine River in Normandy-France. In our study, due to a lack of data, a challenging situation was faced where only the water level data in the same station, which contain many missing parts, were used as input and output variables to fill the station itself in a self-learning approach. This contrasts with the common work on imputing missing data, where several features are available to take advantage of in a multivariate and spatiotemporally way, e.g.: using the same variable from other stations or exploiting other physical variables and metrological data, etc. The reconstruction accuracy of the proposed method depends on both the size of the available/missing data and the parameters of the networks. Therefore, we performed sensitivity analyses on both the properties of the networks and the structuring of the input and output data to better determine the appropriate strategy. During this analysis process, a data preprocessing method was developed and added prior to the LSTM model. This data processing method was discovered by presenting many scenarios, each of which was an updated version of the last one. Along with these scenarios, limitations were also addressed and overcome. Finally, the last model version was able to impute missing values that may reach one year of hourly data with high accuracy (One-year RMSE = 0.14 m) regardless of neither the location of the missing part in the series nor its size.  

How to cite: Janbain, I., Deloffre, J., Jardani, A., Vu, M. T., and Massei, N.: Use of Long-Short Term Memory network (LSTM) in the reconstruction of missing water level data in the Seine River., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4970, https://doi.org/10.5194/egusphere-egu23-4970, 2023.

A.58
|
EGU23-5445
|
ECS
James Donnelly, Alireza Daneshkhah, and Soroush Abolfathi

The application of numerical models for flood and inundation modelling has become widespread in the past decades as a result of significant improvements in computational capabilities. Computational approaches to flood forecasting have significant benefits compared to empirical approaches which estimate statistical patterns of hydrological variables from observed data. However, there is still a significant computational cost associated with numerical flood modelling at high spatio-temporal resolutions. This limitation of numerical modelling has led to the development of statistical emulator models, machine learning (ML) models designed to learn the underlying generating process of the numerical model. The data-driven approach to ML involves relying entirely upon a set of training data to inform decisions about model selection and parameterisations. Deep learning models have leveraged data-driven learning methods with improvements in hardware and an increasing abundance of data to obtain breakthroughs in various fields such as computer vision, natural language processing and autonomous driving. In many scientific and engineering problems however, the cost of obtaining data is high and so there is a need for ML models that are able to generalise in the ‘small-data’ regime common to many complex problems. In this study, to overcome extrapolation and over-fitting issues of data-driven emulators, a Physics-Informed Neural Network model is adopted for the emulation of all two-dimensional hydrodynamic models which model fluid according the shallow water equations. This study introduces a novel approach to encoding the conservation of mass into a deep learning model, with additional terms included in the optimisation criterion, acting to regularise the model, avoid over-fitting and produce more physically consistent predictions by the emulator.

How to cite: Donnelly, J., Daneshkhah, A., and Abolfathi, S.: Physics-Informed Neural Networks for Statistical Emulation of Hydrodynamical Numerical Models, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5445, https://doi.org/10.5194/egusphere-egu23-5445, 2023.

A.59
|
EGU23-6313
|
ECS
|
Mathias Busk Dahl, Troels Norvin Vilhelmsen, Rasmus Bødker Madsen, and Thomas Mejer Hansen

Decision-making related to groundwater management often relies on results from a deterministic groundwater model representing one ‘optimal’ solution. However, such a single deterministic model lacks representation of subsurface uncertainties. The simplicity of such a model is appealing, as typically only one is needed, but comes with the risk of overlooking critical scenarios and possible adverse environmental effects. Instead, we argue, that groundwater management should be based on a probabilistic model that incorporates the uncertainty of the subsurface structures to the extent that it is known. If such a probabilistic model exists, it is, in principle, simple to propagate the uncertainties of the model parameter using multiple numerical simulations, to allow a quantitative and probabilistic base for decision-makers. However, in practice, such an approach can become computationally intractable. Thus, there is a need for quantifying and propagating the uncertainty numerical simulations and presenting outcomes without losing the speed of the deterministic approach.

This presentation provides a probabilistic approach to the specific groundwater modelling task of determining well recharge areas that accounts for the geological uncertainty associated with the model using a deep neural network. The results of such a task are often part of an investigation for new abstraction well locations and should, therefore, present all possible outcomes to give informative decision support. We advocate for the use of a probabilistic approach over a deterministic one by comparing results and presenting examples, where probabilistic solutions are essential for proper decision support. To overcome the significant increase in computation time, we argue that this problem can be solved using a probabilistic neural network trained on examples of model outputs. We present a way of training such a network and show how it performs in terms of speed and accuracy. Ultimately, this presentation aims to contribute with a method for incorporating model uncertainty in groundwater modelling without compromising the speed of the deterministic models.

How to cite: Busk Dahl, M., Norvin Vilhelmsen, T., Bødker Madsen, R., and Mejer Hansen, T.: Moving away from deterministic solutions: A probabilistic machine learning approach to account for geological model uncertainty in groundwater modelling, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-6313, https://doi.org/10.5194/egusphere-egu23-6313, 2023.

A.60
|
EGU23-7828
|
ECS
|
Claudia Bertini, Gerald Corzo, Schalk Jan van Andel, and Dimitri Solomatine

Water managers need accurate rainfall forecasts for a wide spectrum of applications, ranging from water resources evaluation and allocation, to flood and drought predictions. In the past years, several frameworks based on Artificial Intelligence have been developed to improve the traditional Numerical Weather Prediction (NWP) forecasts, thanks to their ability of learning from past data, unravelling hidden relationships among variables and handle large amounts of inputs. Among these approaches, Long Short-Term Memory (LSTM) models emerged for their ability to predict sequence data, and have been successfully used for rainfall and flow forecasting, mainly with short lead-times. In this study, we explore three different multi-variate LSTM-based models, i.e. vanilla LSTM, stacked LSTM and bidirectional LSTM, to forecast daily precipitation for the upcoming 30 days in the area of Rhine Delta, the Netherlands. We use both local atmospheric and global climate variables from the ERA-5 reanalysis dataset to predict rainfall, and we introduce a fuzzy index for the models to account for seasonality effects. The framework is developed within the H2020 project CLImate INTelligence (CLINT), and its outcomes have the potential to improve forecasting precipitation deficit in the study area.

How to cite: Bertini, C., Corzo, G., van Andel, S. J., and Solomatine, D.: Sub-seasonal daily precipitation forecasting based on Long Short-Term Memory (LSTM) models, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-7828, https://doi.org/10.5194/egusphere-egu23-7828, 2023.

A.61
|
EGU23-8746
|
ECS
Everett Snieder and Usman Khan

Recent years have seen an increase of deep learning applications for flow forecasting. Large-sample hydrological (LSH) studies typically try to predict the runoff of a catchment using some selection of hydrometeorological features from the respective catchment. One aspect of these models that has received little attention in LSH is the effect that data from upstream catchments has on model performance. The number of available and stations and distance between stations is highly variable between catchments, which creates a unique modelling challenge. Existing LSH studies either use some form of linear aggregation of upstream flows as input features or omit them altogether. The potential of upstream data to improve the performance of real-time flow forecasts has not yet been systematically evaluated on a large scale. The objective of our study is to evaluate methods for integrating upstream features for real-time, data-driven flow forecasting models. Our study uses a subset of Canadian catchments (n>150) from the HYSETS database. For each catchment, long-short term memory networks (LSTMs) are used to generate flow forecasts for lead times of 1 to 3 days. We evaluate methods for identifying, selecting, and integrating relevant upstream input features within a deep-learning modelling framework, which include using neighbouring upstream stations, using all upstream stations, and using all stations with embedded dimensionality reduction. Early results indicate that while the inclusion of upstream data often yields improvements in model performance, including too much upstream information can easily have detrimental effects.

How to cite: Snieder, E. and Khan, U.: A large sample study of the effects of upstream hydrometeorological input features for LSTM-based daily flow forecasting in Canadian catchments, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8746, https://doi.org/10.5194/egusphere-egu23-8746, 2023.

A.62
|
EGU23-15604
|
ECS
|
Anna Pölz, Julia Derx, Andreas Farnleitner, and Alfred Paul Blaschke

Karst springs provide drinking water for approximately 700 million people worldwide. Complex subsurface flow processes lead to challenges for modelling spring discharges. Machine learning (ML) models possess the ability to learn non-linear patterns and show promising results in forecasting dynamic spring discharge. We compare the performance of three ML models of varying complexity in forecasting karst spring discharges: the multivariate adaptive regression spline model (MARS), a feed-forward neural network (ANN) and a long short-term memory model (LSTM). The well-studied alpine karst spring LKAS2 in Austria is used as test case. We provide model explanations including feature attribution through Shapley additive explanations (SHAP), a method based on Shapley values. Our results show that the higher the model complexity, the higher the accuracy, based on the evaluated symmetric mean absolute percentage error of the three investigated models. With SHAP every prediction can be explained through each feature in each input time step. We found seasonal model differences. For example, snow influenced the model mostly in winter and spring. Analyzing the combinations of input time steps and features provided further insights into the model performance. For instance, the SHAP results showed that a high electrical conductivity in recent time steps, which indicates that the karst water is less diluted with precipitation, leads to a reduced discharge forecast. These feature attribution results coincide with physical processes within karst systems. Therefore, the introduced SHAP method can increase the confidence in ML model forecasts and emphasizes the raison d’être of complex and accurate deep learning models in hydrology. This allows the operator to better understand and evaluate the model’s output, which is essential for drinking water management.

How to cite: Pölz, A., Derx, J., Farnleitner, A., and Blaschke, A. P.: Forecasting discharges through explainable machine learning approaches at an alpine karst spring, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-15604, https://doi.org/10.5194/egusphere-egu23-15604, 2023.

A.63
|
EGU23-9726
|
ECS
|
Tanja Morgenstern, Jens Grundmann, and Niels Schütze

Floods are among the most frequently occurring natural disasters in Germany. Therefore, predicting their occurrence is a crucial task for efficient disaster management and for the protection of life, property, infrastructure and cultural assets. In recent years Deep Learning methods gained popularity on the research field on flood forecasting methods – Long Short-Term Memory (LSTM) networks being part of them.

Efficient disaster management needs a fine temporal resolution of runoff predictions. Past work at TU Dresden on LSTM networks shows certain challenges when using input data with hourly resolution, such as systematically poor timing in peak flow prediction (Pahner et al. (2019) and Morgenstern et al. (2021)). At times, disaster management even requires flood forecasts for hitherto unobserved catchments, so in total a regionally transferable rainfall-runoff model with a fine temporal resolution is needed. We derived the idea for a potential approach from Kratzert et al. (2019) and Fang et al. (2021): they demonstrate that LSTM networks for rainfall(R)-runoff(R)-modeling benefit from an integration of multiple diverse catchments in the training dataset instead of a strictly local dataset, as this allows the networks to learn universal hydrologic catchment behavior. However, their training dataset consists of daily resolution data.

Following this approach, in this study we train the LSTM networks using single catchments ("local network training") as well as combinations of diverse catchments in Saxony, Germany ("regional network training"). The training data (hourly resolution) consist of area averages of observed precipitation as well as of observed discharge at long-term observation gauges in Saxony. The gauges belong to small, fast-responding Saxon catchments and vary in their hydrological and geographical properties, which in turn are part of the network training as well.

We show the preliminary results and investigate the following questions:

  • With a finer temporal resolution than daily values, characteristics of flood waves become more pronounced. Concerning the detailed simulation of flood waves, do regional LSTM-based R-R-models enable more accurate and robust flow predictions compared to local LSTM-based R-R-models – especially for rare extreme events?
  • Are regional LSTM-based R-R-models – trained at this temporal resolution – able to generalize to unobserved areas or areas with discharge observations unsuitable for network training?

 

References

Fang, K., Kifer, D., Lawson, K., Feng, D., Shen, C. (2022). The Data Synergy Effects of Time-Series Deep Learning Models in Hydrology. In: Water Resources Research (58). DOI: 10.1029/2021WR029583

Kratzert, F., Klotz, D., Shalev, G., Klambauer, G., Hochreiter, S., Nearing, G. (2019). Towards learning universal, regional, and local hydrological behaviors via machine learning applied to large-sample datasets. Hydrology and Earth System Sciences (23), S. 5089–5110. DOI: 10.5194/hess-23-5089-2019

Morgenstern, T., Pahner, S., Mietrach, R., Schütze, N. (2021): Flood forecasting in small catchments using deep learning LSTM networks. DOI: 10.5194/egusphere-egu21-15072

Pahner, S., Mietrach, R., Schütze, N. (2019): Flood Forecasting in small catchments: a comparative application of long short-term memory networks and artificial neural networks. DOI: 10.13140/RG.2.2.36770.89286.

How to cite: Morgenstern, T., Grundmann, J., and Schütze, N.: Flood Forecasting with Deep Learning LSTM Networks: Local vs. Regional Network Training Based on Hourly Data, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9726, https://doi.org/10.5194/egusphere-egu23-9726, 2023.

A.64
|
EGU23-10317
|
ECS
Hyojeong Choi and Dongkyun Kim

Precipitation forecast models based on meteorological radar data using machine learning architectures accurately predict spatio-temporal progress of precipitation. However, these data-driven forecasting models tend to underestimate magnitude of extreme precipitation events because the training of them is based on the observed precipitation data in which the normal precipitation events are included significantly more than the rare extreme events. This study proposes a ConvLSTM-based precipitation nowcasting model that can accurately predict space-time field of extreme precipitation. First, precipitation events were classified into 5 subsets using the k-means clustering algorithm based their statistical properties such as mean, standard deviation, skewness, duration, and the calendar month at which the precipitation event occurred. Then, a ConvLSTM-based neural network was trained based on the subset containing extreme precipitation events (events with large mean, variance, and duration occurred in summer months). The model was trained and tested based on the 4km-10minute resolution radar-gauge composite precipitation field of central part of South Korea (200km x 200km) for the period of 2009-2015 and 2016-2020, respectively. The NSE of the model that was trained based on the whole precipitation data was 0.55 while the one trained based on the subset of extreme precipitation was 0.78 showing a significant improvement.

This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. NRF-2021R1A2C2003471). 

How to cite: Choi, H. and Kim, D.: A convolutional LSTM model with high accuracy to predict extreme precipitation space-time fields, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10317, https://doi.org/10.5194/egusphere-egu23-10317, 2023.

A.65
|
EGU23-12315
Tobias Krueger, Mark Somogyvari, Ute Fehrenbach, and Dieter Scherer

Process-based models are the standard tools today when trying to understand how physical systems work. There are situations however, when system understanding is not a primary focus and it is worth substituting existing process-based models with computationally more efficient meta-models (or emulators), i.e. proxies designed for specific applications. In our research we have explored potential data-driven meta-modeling approaches for applications in hydrology, designed to solve specific research questions.

In order to find a suitable meta-modeling approach, we have experimented with a set of different data-driven methods. We have employed a multi-fidelity modeling approach, where we gradually increased the complexity of our models. In total five different approaches were investigated: linear model with ordinary least squares regression, linear model with two different Bayesian methods (Hamiltonian Monte Carlo and transdimensional Monte Carlo) and two machine learning approaches (dense artificial neural network and long short-term memory (LSTM) neural network).

For method development the project case study of the Groß Glienicker Lake was used. This is a glacial lake near Berlin, with a strong negative trend in water levels in the last decades. Supported by the observation model from the Central European Refined analysis, we had a daily, high resolution meteorological dataset (precipitation and actual evapotranspiration) and lake level observations for 16 years.

All of the used models are designed similarly: they predict lake level changes one day ahead using precipitation and evapotranspiration data from the previous 70 days. This interval was selected after an extensive parameter test with the linear model. By predicting the change in stored water, we linearize the problem, and by using a longer time interval we allow the methods to automatically compensate for any lag or memory effects inside the catchment. The different methods are evaluated by comparing the fits between the observed and the reconstructed lake levels.

As expected, increasing the model and inversion complexity improves the quality of the reconstruction. Especially the use of nonlinear models was advantageous, the artificial neural network outperformed every other method. However, in the used example these improvements were relatively small – meaning that in practice the simplest linear method was advantageous due to its computational efficiency and robustness, and ease of use and interpretation.

In this presentation we discuss the challenges of data preparation and optimal model design (especially the memory of the hydrological system), while finding the hyperparameters of the specific methods themselves was relatively straight forward. Our results suggest that problem linearization should be a preferred first step in any meta-modeling application, as it helps the training of nonlinear models as well. We also discuss data requirements, because we found that the size of our dataset was too small for the most complex LSTM method, which yielded unstable results and learned spurious background trends.

How to cite: Krueger, T., Somogyvari, M., Fehrenbach, U., and Scherer, D.: Meta-modeling with data-driven methods in hydrology, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12315, https://doi.org/10.5194/egusphere-egu23-12315, 2023.

A.66
|
EGU23-13493
|
ECS
fadil boodoo, carole delenne, Renaud hostache, and julien freychet

Accurate streamflow forecasting can help minimizing the negative impacts of hydrological events such as floods and droughts. To address this challenge, we explore here artificial neural networks models (ANNs) for streamflow forecasting. These models, which have been proven successful in other fields, may offer improved accuracy and efficiency compared to traditional conceptually-based forecasting approaches.

The goal of this study is to compare the performance of a traditional conceptual rainfall-runoff (hydrological) model with an artificial neural network (ANN) model for streamflow forecasting. As a test case, we use the Severn catchment in the United Kingdom. The adopted ANN model has a long short-term memory (LSTM) architecture with two hidden layers, each with 256 neurons. The model is trained on a 25-year dataset from 1988 to 2013 and tested on a 3-year dataset (from 2014 to 2016). It is also validated on a 3-year dataset (from 2017 to 2020, 2019 being a particularly wet year), to assess its performance in extreme hydrological conditions. The study focuses on daily and hourly predictions.

To conduct this study, the conceptual hydrological model called Superflex is used as a benchmark. Both models are first evaluated using the Nash-Sutcliffe Efficiency (NSE) score. To enable a fair and accurate comparison, both models share the same inputs (i.e. meteorological forcings: total precipitation, daily maximum and minimum temperatures, daylight duration, mean surface downward short wave radiation flux, and vapor pressure). The ANN model was implemented using the Neuralhydrology library developed by F. Kratzert.

In our study, we found that LSTM model is able to provide more accurate one-day forecasts than the  hydrological model Superflex. For the daily predictions, the average NSE score using the LSTM model is 0.85 (with an average NSE score of 0.99 for training period, and 0.85 for validation period), which is higher than the NSE score of 0.74 achieved by the Superflex model (with a score of 0.84 for training period).

The hourly prediction using NSE with the superflex model had a score of 0.88, with a score of 0.7 during training. The LSTM model had an average NSE score of 0.87, with an average score of 0.99 during training and an average score of 0.85 during validation.

These results were obtained without adjusting the hyperparameters and by training the model only on data from the Severn watershed.The ANN model has demonstrated promising results compared to a state-of-the-art conceptual hydrological model in our studies. We will further compare both models using different training dataset periods, and different catchements. These additional tests will provide more information on the capabilities of the LSTM model and help to confirm its effectiveness.

How to cite: boodoo, F., delenne, C., hostache, R., and freychet, J.: Comparison of  a conceptual rainfall-runoff  model with an artificial neural network model for streamflow prediction, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-13493, https://doi.org/10.5194/egusphere-egu23-13493, 2023.

Posters virtual: Tue, 25 Apr, 08:30–10:15 | vHall HS

Chairpersons: Frederik Kratzert, Martin Gauch
vHS.8
|
EGU23-2382
Renjie Zhou and Yanyan Zhang

Having a continuous and complete karst discharge data record is necessary to understand hydrological behaviors of the karst aquifer and manage karst water resources. However, caused by many problems such as equipment errors and failure of observation, lots of hydrological and research dataset contains missing spring discharge values, which becomes a main barrier for further environmental and hydrological modeling and studies. In this work, a novel approach that integrates deep learning algorithms and ensemble empirical mode decomposition (EEMD) is developed to reconstruct missing karst spring discharge values with the local precipitation. EEMD is firstly employed to decompose the precipitation data, extract useful features, and remove noises. The decomposed precipitation components are then fed as input data to various deep learning models for performance comparison, including convolutional neural network (CNN), long short-term memory (LSTM), and hybrid CNN-LSTM models to reconstruct the missing discharge values. Root mean squared error (RMSE) and Nash–Sutcliffe efficiency coefficient (NSE), are calculated to evaluate the reconstruction performance as metrics. The models are validated with the spring discharge and precipitation data collected at Barton Spring in Texas. The reconstruction performance of various deep learning models with and without EEMD are compared and evaluated. The main conclusions can be summarized as: 1) by using EEMD, the integrated deep models significantly improve reconstruction performance and outperform the simple deep models; 2) among three integrated models, the LSTM-EEMD model obtains the best reconstruction results among three deep learning algorithms; 3) For models with monthly data, the reconstruction performance decreases greatly with the increase of missing rate: the best reconstruction results are obtained when the missing rate is low. If the missing rate was 50%, the reconstruction results become notably poorer. For models with daily data, the reconstruction performance is less impacted by the missing rate and the models can obtain satisfactory reconstruction results when missing rates range from 10% to 50%.

How to cite: Zhou, R. and Zhang, Y.: Reconstruct karst spring discharge data with hybrid deep learning models and ensemble empirical mode decomposition method, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2382, https://doi.org/10.5194/egusphere-egu23-2382, 2023.

vHS.9
|
EGU23-4177
|
ECS
Abdalla Mohammed and Gerald Corzo

Rainfall-runoff (RR) modeling remains a challenging task in the field of hydrology especially when it comes to regional scale hydrology. Recently, the Long Short-Term Memory (LSTM) - which is known for its ability to learn sequential and temporal relations - has been widely adopted in RR modeling. The Convolutional Neural Networks (CNN) have matured enough in computer vision tasks, and trials were conducted to use them in hydrological applications. Different combinations of CNN and LSTM have proved to work; however, questions remain about suitability of different model architectures, the input variables needed for the model and the interpretability of the learning process of the models for regional scale.

 

In this work we trained a sequential CNN-LSTM deep learning architecture to predict daily streamflow between 1980 and 2014, regionally and simultaneously, over 86 catchments from CAMELS dataset in the US. The model was forced using year-long spatially distributed (gridded) input with precipitation, maximum temperature and minimum temperature for each day, to predict one day streamflow. The model takes advantage of the CNN to encode the spatial patterns in the input tensor, and feed them to the LSTM for learning the temporal relations between them. The trained model was further fine-tuned to predict for 3 local sub-clusters of the 86 stations. This was made in order to test the significance of fine-tuning in the performance and model learning process. Also, to interpret the spatial patterns learning process, a perturbation was introduced in the gridded input data and the sensitivity of the model output to the perturbation was shown in spatial heat maps. Finally, to evaluate the performance of the model, different benchmark models were trained using -as possible- a similar training setup as for the CNN-LSTM model. These models are CNN without the LSTM part (regional model), LSTM without CNN part (regional model), simple single-layer ANN (regional model), and LSTM trained for individual stations (considered as state of the art). All of these benchmark models have been fined-tuned for the 3 clusters as well.

 

CNN-LSTM model, after being fine-tuned, performed well predicting daily streamflow over the test period with a median Nash-Sutcliffe efficiency (NSE) of 0.62 and 65% of the 86 stations with NSE > 0.6 outperforming all benchmark models that were trained regionally using the same training setup. The model also achieved a comparable performance as for the -state of the art- LSTM trained for individual stations. Fine-tuning improved the performance for all of the models during the test period. The CNN-LSTM model, was shown to be more sensitive to input perturbations near the stations in which the prediction is intended. This was even clearer for the fine-tuned model, indicating that the model is learning spatially relevant information from the input gridded data, and fine tuning is helping on guiding the model to focus more on the relevant input.  

 

This work shows the potential of CNN and LSTM for regional Rainfall-runoff modeling by capturing spatiotemporal patterns involved in RR process. The work, also, contributes toward more physically interpretable data-driven modeling paradigm.

How to cite: Mohammed, A. and Corzo, G.: Evaluation of regional Rainfall-Runoff modelling using convolutional long short-term memory:  CAMELS dataset in US as a case study., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4177, https://doi.org/10.5194/egusphere-egu23-4177, 2023.

vHS.10
|
EGU23-5199
|
ECS
|
Nicolas Weaver, Taha-Abderrahman El-Ouahabi, Thibault Hallouin, François Bourgin, Charles Perrin, and Vazken Andréassian

Machine learning models have recently gained popularity in hydrological modelling at the catchment scale, fuelled by the increasing availability of large-sample data sets and the increasing accessibility of deep learning frameworks, computing environments, and open-source tools. In particular, several large-sample studies at daily and monthly time scales across the globe showed successful applications of the LSTM architecture as a regional model learning of the hydrological behaviour at the catchment scale. Yet, a deeper understanding of how machine learning models close the water balance and how they deal with inter-catchment groundwater flows is needed to move towards better process understanding. We investigate the performance and behaviour of the LSTM architecture at a monthly time step on a large sample French data set coined CHAMEAU – following the CAMELS initiative. To provide additional information to the learning step of the LSTM, we use the parameter sets and fluxes from the conceptual GR2M model that has a dedicated formulation to deal with inter-catchment groundwater flows. We see this study as a contribution towards the development of hybrid hydrological models.

How to cite: Weaver, N., El-Ouahabi, T.-A., Hallouin, T., Bourgin, F., Perrin, C., and Andréassian, V.: How do machine learning models deal with inter-catchment groundwater flows?, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5199, https://doi.org/10.5194/egusphere-egu23-5199, 2023.

vHS.11
|
EGU23-15629
|
ECS
Nicole Tatjana Scherer, Muhammad Nabeel Usmann, Markus Disse, and Jingshui Huang

Most floods are caused by heavy rainfall events, including the disaster in the Simbach catchment in 2016. For the Simbach catchment, a study was already carried out using the conceptual Hydrologiska Byråns Vattenbalansavdelning (HBV) model to simulate the extreme event of 2016. While the calibration model performance is classified as very good, the overall validation is classified as unsatisfactory. Recent studies showed that data-driven models outperform benchmark rainfall-runoff models. A widely used data-driven model is the Long-Short-Term-Memory algorithm (LSTM). The main advantage of this algorithm is the ability to learn short-term as well as long term dependencies.

The objective of this work is to determine if a data-driven model outperforms the conceptual model. For this purpose, in a first step a LSTM model is setup and its results are compared with the results of the HBV model. It is assumed that the LSTM model outperforms the HBV model in training and validation but is not able to simulate the extreme event, because the extrapolation capabilities of Neuronal Networks are poor if they operate outside of their training range. In a second step, it is studied if the model performance can be improved by providing more features to the model. Therefore, different feature combinations are provided to the model. Furthermore, it is assumed that providing more data to the model will improve its performance. Therefore, in a third step more events are used for training and validation.

It was concluded that the LSTM model is able to simulate the rainfall-runoff process. A satisfactory overall model performance can be achieved using only precipitation as input data and a small training dataset of four events. But, as the HBV model, the LSTM model is not able to simulate the extreme event, because no extreme event is present within the training dataset. However, the LSTM model outperforms the HBV model, because the LSTM generalizes better. Furthermore, the model performance of the LSTM model using six events can be improved by providing additionally the soil moisture class as input data. Whereas providing more features to the model results in worse model performance. Providing more events to the model does not significantly improve its performance. However, the model improved especially for the event in June 2015. If the model is trained with more events having higher magnitude than the 2015 event, the event in 2015 is no longer classified as an out-of-sample event, resulting in better model performance. Providing the model more events and more input features does not significantly improve the model performance. 

The results show the potential and limitations using the LSTM model in modeling extreme events.

How to cite: Scherer, N. T., Usmann, M. N., Disse, M., and Huang, J.: Peak Hydrological Event Simulation with Deep Learning Algorithm, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-15629, https://doi.org/10.5194/egusphere-egu23-15629, 2023.