HS3.4 | Deep learning in hydrology
Deep learning in hydrology
Co-organized by ESSI1/NP4
Convener: Frederik KratzertECSECS | Co-conveners: Anna PölzECSECS, Basil KraftECSECS, Daniel KlotzECSECS, Martin GauchECSECS
Orals
| Fri, 19 Apr, 14:00–15:45 (CEST), 16:15–18:00 (CEST)
 
Room 2.31
Posters on site
| Attendance Thu, 18 Apr, 10:45–12:30 (CEST) | Display Thu, 18 Apr, 08:30–12:30
 
Hall A
Posters virtual
| Attendance Thu, 18 Apr, 14:00–15:45 (CEST) | Display Thu, 18 Apr, 08:30–18:00
 
vHall A
Orals |
Fri, 14:00
Thu, 10:45
Thu, 14:00
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. We welcome abstracts 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) Probing, exploring and improving our understanding of the (internal) states/representations of deep learning models to improve models and/or gain system insights.

(3) Understanding the reliability of deep learning, e.g., under non-stationarity and climate change.

(4) Modeling human behavior and impacts on the hydrological cycle.

(5) Deep Learning approaches for extreme event analysis, detection, and mitigation.

(6) Natural Language Processing in support of models and/or modeling workflows.

(7) Uncertainty estimation for and with Deep Learning.

(8) Applications of Large Language Models (e.g. ChatGPT, Bard, etc.) in the context of hydrology.

(9) Advances towards foundational models in the context of hydrology and Earth Sciences more generally.

(10) Exploration of different optimization strategies, such as self-supervised learning, unsupervised learning, and reinforcement learning.

Orals: Fri, 19 Apr | Room 2.31

Chairpersons: Frederik Kratzert, Daniel Klotz, Martin Gauch
14:00–14:05
New modelling approaches
14:05–14:15
|
EGU24-6846
|
HS3.4
|
ECS
|
On-site presentation
Peter Nelemans, Roberto Bentivoglio, Joost Buitink, Ali Meshgi, Markus Hrachowitz, Ruben Dahm, and Riccardo Taormina

Fully distributed hydrological models take into account the spatial variability of a catchment, allowing for a more accurate representation of its heterogeneity, and assessing its hydrological response at multiple locations. However, physics-based fully distributed models can be time-consuming when it comes to model runtime and calibration, especially for large-scale catchments. On the other hand, deep learning models have shown great potential in the field of hydrological modelling, outperforming lumped rainfall-runoff conceptual models, and improving prediction in ungauged basins via catchment transferability. Despite these advances, the field still lacks a multivariable, fully distributed hydrological deep learning model capable of generalizing to unseen catchments. To address the aforementioned challenges associated with physics-based distributed models and deep learning models, we explore the possibility of developing a fully distributed deep learning model by using Graph Neural Networks (GNN), an extension of deep learning methods to non-Euclidean topologies including graphs and meshes.

We develop a surrogate model of wflow_sbm, a fully distributed, physics-based hydrological model, by exploiting the similarities between its underlying functioning and GNNs. The GNN uses the same input as wflow_sbm: distributed static parameters based on physical characteristics of the catchment and gridded dynamic forcings. The GNN is trained to produce the same output as wflow_sbm, predicting multiple gridded variables related to rainfall-runoff, such as streamflow, actual evapotranspiration, subsurface flow, saturated and unsaturated groundwater storage, snow storage, and runoff. We show that our GNN model achieves high performance in unseen catchments, indicating that GNNs are a viable option for fully distributed multivariable hydrological models capable of generalizing to unseen regions. Furthermore, the GNN model achieves a significant computational speedup compared to wflow_sbm. We will continue this research, using the GNN-based surrogate models as pre-trained backbones to be fine-tuned with measured data, ensuring accurate model adaptation, and enhancing their practical applicability in diverse hydrological scenarios.

How to cite: Nelemans, P., Bentivoglio, R., Buitink, J., Meshgi, A., Hrachowitz, M., Dahm, R., and Taormina, R.: Towards Fully Distributed Rainfall-Runoff Modelling with Graph Neural Networks, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-6846, https://doi.org/10.5194/egusphere-egu24-6846, 2024.

14:15–14:25
|
EGU24-2939
|
HS3.4
|
ECS
|
On-site presentation
Mohammad Sina Jahangir and John Quilty

The significance of probabilistic hydrological forecasting has grown in recent years, offering crucial insights for risk-based decision-making and effective flood management. This study explores generative deep learning models, specifically the conditional variational autoencoder (CVAE), for probabilistic streamflow forecasting. This innovative approach is applied for forecasting streamflow one to seven days (s) ahead in 75 Canadian basins included in the open-source Canadian model parameter experiment (CANOPEX) database. CVAE is compared against two benchmark quantile-based deep learning models: the quantile-based encoder-decoder (ED) and the quantile-based CVAE (QCVAE).

Over 9000 deep learning models are developed based on different input variables, basin characteristics, and model structures and evaluated regarding point forecast accuracy and forecast reliability. Results highlight CVAE‘s superior reliability, showing a median reliability of 92.49% compared to 87.35% for ED and 84.59% for QCVAE (considering a desired 90% confidence level). However, quantile-based forecast models exhibit marginally better point forecasts, as evidenced by Kling-Gupta efficiency (KGE), with a median KGE of 0.90 for ED and QCVAE (compared to 0.88 for CVAE). Notably, the CVAE model provides reliable probabilistic forecasts in basins with low point forecast accuracy.

The developed generative deep learning models can be used as a benchmark for probabilistic streamflow forecasting due to the use of the open-source CANOPEX dataset. Overall, the results of this study contribute to the expanding field of generative deep learning models in hydrological forecasting, offering a general framework that applies to forecasting other hydrological variables as well (precipitation and soil moisture).

How to cite: Jahangir, M. S. and Quilty, J.: Probabilistic streamflow forecasting using generative deep learning models, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-2939, https://doi.org/10.5194/egusphere-egu24-2939, 2024.

14:25–14:35
|
EGU24-20636
|
HS3.4
|
ECS
|
Virtual presentation
Jiangtao Liu, Chaopeng Shen, and Tadd Bindas

Accurate modeling of various hydrological variables is important for water resource management, flood forecasting, and pest control. Deep learning models, especially Long Short-Term Memory (LSTM) models based on Recurrent Neural Network (RNN) structures, have shown significant success in simulating streamflow, soil moisture, and model parameter assessment. With the development of large language models (LLMs) based on attention mechanisms, such as ChatGPT and Bard, we have observed significant advancements in fields like natural language processing (NLP), computer vision (CV), and time series prediction. Despite achieving advancements across various domains, the application of attention-based models in hydrology remains relatively limited, with LSTM models maintaining a dominant position in this field. This study evaluates the performance of 18 state-of-the-art attention-based models and their variants in hydrology. We focus on their performance in streamflow, soil moisture, snowmelt, and dissolved oxygen (DO) datasets, comparing them to LSTM models in both long-term and short-term regression and forecasting. We also examine these models' performance in spatial cross-validation. Our findings indicate that while LSTM models maintain strong competitiveness in various hydrological datasets, Attention models offer potential advantages in specific metrics and time lengths, providing valuable insights into applying attention-based models in hydrology. Finally, we discuss the potential applications of foundation models and how these methods can contribute to the sustainable use of water resources and the challenges of climate change.

How to cite: Liu, J., Shen, C., and Bindas, T.: Can Attention Models Surpass LSTM in Hydrology?, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-20636, https://doi.org/10.5194/egusphere-egu24-20636, 2024.

14:35–14:42
Benchmarking
14:42–14:52
|
EGU24-18154
|
HS3.4
|
ECS
|
Virtual presentation
Daneti Arun Sourya and Maheswaran Rathinasamy

Streamflow simulation or rainfall-runoff modelling has been a topic of research for the past few decades which has resulted in a plethora of modelling approaches ranging from physics models to empirical or data driven approaches. There are many physics-based (PB) models available to estimate streamflow, but still there exists uncertainty in model outputs due to incomplete representations of physical processes. Further, with advancements in machine learning (ML) concepts, there have been several attempts but with no/little physical consistency. As a result, models based on ML algorithms may be unreliable if applied to provide future hydroclimate projections where climates and land use patterns are outside the range of training data. 

Here we test blended models built by combining PB model state variables (specifically soil moisture) with ML algorithms on their ability to simulate streamflow in 671 catchments representing diverse conditions across the conterminous United States.

For this purpose, we develop a suite of blended hydrological models by pairing different PB models (Catchment Wetness Index, Catchment Moisture Deficit, GR4J, Australian Water Balance, Single-bucket Soil Moisture Accounting, and Sacramento Soil Moisture Accounting models) with different ML methods such as Long Short Term Memory network (LSTM), eXtreme Gradient Boosting (XGB).

The results indicate that the blended models provide significant improvement in catchments where PB models are underperforming. Furthermore, the accuracy of streamflow estimation is improved in catchments where the ML models failed to estimate streamflow accurately.

How to cite: Sourya, D. A. and Rathinasamy, M.: Can Blended Model Improve Streamflow Simulation In Diverse Catchments ?, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-18154, https://doi.org/10.5194/egusphere-egu24-18154, 2024.

14:52–15:02
|
EGU24-18762
|
HS3.4
|
On-site presentation
Elizabeth Lewis, Ben Smith, Stephen Birkinshaw, Helen He, and David Pritchard

National scale hydrological models are required for many types of water sector applications, for example water resources planning. Existing UK national-scale model frameworks are based on conceptual numerical schemes, with an emerging trend towards incorporating deep learning models. Existing literature has shown that groundwater/surface water interactions are key for accurately representing future flows, and these processes are most accurately represented with physically-based hydrological models.

In response to this, our study undertakes a comparative analysis of three national model frameworks (Neural Hydrology, HBV, SHETRAN) to investigate the necessity for physically-based hydrological modelling. The models were run with the full ensemble of bias-corrected UKCP18 12km RCM data which enabled a direct comparison of future flow projections. We show that whilst many national frameworks perform well for the historical period, physically-based models can give substantially different projections of future flows, particularly low flows. Moreover, our study illustrates that the physically-based model exhibits a consistent trajectory in Budyko space between the baseline and future simulations, a characteristic not shared by conceptual and deep learning models. To provide context for these results, we incorporate insights from other national model frameworks, including the eFlag project.

How to cite: Lewis, E., Smith, B., Birkinshaw, S., He, H., and Pritchard, D.: Benchmarking hydrological models for national scale climate impact assessment, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-18762, https://doi.org/10.5194/egusphere-egu24-18762, 2024.

15:02–15:12
|
EGU24-6432
|
HS3.4
|
ECS
|
On-site presentation
Hemant Servia, Frauke Albrecht, Samuel Saxe, Nicolas Bierti, Masatoshi Kawasaki, and Shun Kurihara

In addressing the challenge of streamflow prediction in ungauged basins, this study leveraged deep learning (DL) models, especially long short-term memory (LSTM) networks, to predict streamflow for pseudo ungauged basins in Japan. The motivation stems from the recognized limitations of traditional hydrological models in transferring their performance beyond the calibrated basins. Recent research suggests that DL models, especially those trained on multiple catchments, demonstrate improved predictive capabilities utilizing the concept of streamflow regionalization. However, the majority of these studies were confined to geographic regions within the United States.

For this study, a total number of 211 catchments were delineated and investigated, distributed across all four primary islands of Japan (Kyushu - 32, Shikoku - 13, Honshu - 127, and Hokkaido - 39) encompassing a comprehensive sample of hydrological systems within the region. The catchments were obtained corresponding to the streamflow observation points and their combined area represented more than 43% of Japan's total land area, after accounting for overlaps. After cleaning and refining the streamflow dataset, the remaining catchments (198) were divided into training (~70%), validation (~20%), and holdout test (~10%) sets. A combination of dynamic (time-varying) and static (constant) variables were obtained on a daily basis corresponding to the daily streamflow data and provided to the models as input features. However, the final model accorded higher significance to dynamic features in comparison to the static ones. Although the models were trained on daily time steps, the results were aggregated to monthly timescale. The main evaluation metrics included the Nash-Sutcliffe Efficiency (NSE) and Pearson’s correlation coefficient (r). The final model achieved a median NSE of 0.96, 0.83, & 0.78, and a median correlation of 0.98, 0.92, & 0.91 corresponding to the training, validation, and test catchments, respectively. For the validation catchments, 90% exhibited NSE values greater than 0.50, and 97% demonstrated a correlation surpassing 0.70. Correspondingly, these proportions were observed at 77% and 91% for the test catchments.

The results presented in this study demonstrate the feasibility and efficacy of developing a data-driven model for streamflow prediction in ungauged basins utilizing streamflow regionalization. The final model exhibits commendable performance, as evidenced by high NSE and correlation coefficients across the majority of the catchments. Importantly, the model's ability to generalize to unseen data is highlighted by its remarkable performance on the holdout test set, with only a few instances of lower NSE values (< 0.50) and correlation coefficients (< 0.70).

How to cite: Servia, H., Albrecht, F., Saxe, S., Bierti, N., Kawasaki, M., and Kurihara, S.: A Step Towards Global Hydrologic Modelling: Accurate Streamflow Predictions In Pseudo-Ungauged Basins of Japan, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-6432, https://doi.org/10.5194/egusphere-egu24-6432, 2024.

15:12–15:19
Improving training
15:19–15:29
|
EGU24-16474
|
HS3.4
|
ECS
|
On-site presentation
Bob E Saint Fleur, Eric Gaume, Michaël Savary, Nicolas Akil, and Dominique Theriez

In recent years, machine learning models, particularly Long Short-Term Memory (LSTM), have proven to be effective alternatives for rainfall-runoff modeling, surpassing traditional hydrological modeling approaches 1. These models have predominantly been implemented and evaluated for rainfall-runoff simulations. However, operational hydrology often requires short- and mid-term forecasts. To be effective, such forecasts must consider past observed values of the predicted variables, requiring a data assimilation procedure 2,3,4. This presentation will evaluate several approaches based on the combination of open-source machine learning tools and data assimilation strategies for short- and mid-term discharge forecasting of flood and/or drought events. The evaluation is based on the rich and well-documented CAMELS dataset 5,6,7. The tested approaches include: (1) coupling pre-trained LSTMs on the CAMELS database with a Multilayer Perceptron (MLP) for prediction error corrections, (2) direct discharge MLP forecasting models specific for each lead time, including past observed discharges as input variables, and (3) option 2, including the LSTM-predicted discharges as input variables. In the absence of historical archives of weather forecasts (rainfall, temperatures, etc.), the different forecasting approaches will be tested in two configurations: (1) weather forecasts assumed to be perfect (using observed meteorological variables over the forecast horizon in place of predicted variables or ensembles) and (2) use of ensembles reflecting climatological variability over the forecast horizons for meteorological variables ensembles made up of time series randomly selected from the past. The forecast horizons considered range from 1 to 10 days, and the results are analyzed in light of the time of concentration of the watersheds.

 

References

1. Kratzert F, Klotz D, Brenner C, Schulz K, Herrnegger M. Rainfall–runoff modelling using Long Short-Term Memory (LSTM) networks. Hydrol Earth Syst Sci. 2018;22(11):6005-6022. doi:10.5194/hess-22-6005-2018

2. Bourgin F, Ramos MH, Thirel G, Andréassian V. Investigating the interactions between data assimilation and post-processing in hydrological ensemble forecasting. J Hydrol (Amst). 2014;519:2775-2784. doi:10.1016/j.jhydrol.2014.07.054

3. Boucher M ‐A., Quilty J, Adamowski J. Data Assimilation for Streamflow Forecasting Using Extreme Learning Machines and Multilayer Perceptrons. Water Resour Res. 2020;56(6). doi:10.1029/2019WR026226

4. Piazzi G, Thirel G, Perrin C, Delaigue O. Sequential Data Assimilation for Streamflow Forecasting: Assessing the Sensitivity to Uncertainties and Updated Variables of a Conceptual Hydrological Model at Basin Scale. Water Resour Res. 2021;57(4). doi:10.1029/2020WR028390

5. Newman AJ, Clark MP, Sampson K, et al. Development of a large-sample watershed-scale hydrometeorological data set for the contiguous USA: data set characteristics and assessment of regional variability in hydrologic model performance. Hydrol Earth Syst Sci. 2015;19(1):209-223. doi:10.5194/hess-19-209-2015

6. Kratzert, F. (2019). Pretrained models + simulations for our HESSD submission "Towards learning universal, regional, and local hydrological behaviors via machine learning applied to large-sample datasets", HydroShare, https://doi.org/10.4211/hs.83ea5312635e44dc824eeb99eda12f06

7. Kratzert, F. (2019). CAMELS Extended Maurer Forcing Data, HydroShare, https://doi.org/10.4211/hs.17c896843cf940339c3c3496d0c1c077

How to cite: Saint Fleur, B. E., Gaume, E., Savary, M., Akil, N., and Theriez, D.: Short- and mid-term discharge forecasts combining machine learning and data assimilation for operational purpose, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-16474, https://doi.org/10.5194/egusphere-egu24-16474, 2024.

15:29–15:39
|
EGU24-9190
|
HS3.4
|
ECS
|
On-site presentation
Alberto Bassi, Antonietta Mira, Marvin Höge, Fabrizio Fenicia, and Carlo Albert

By employing Machine Learning techniques on the US-CAMELS dataset, we discern a minimal number of streamflow features. Together with meteorological forcing, these features enable an approximate reconstruction of the entire streamflow time-series. This task is achieved through the application of an explicit noise conditional autoencoder, wherein the meteorological forcing is inputted to the decoder to encourage the encoder to learn streamflow features exclusively related to landscape properties. The optimal number of encoded features is determined with an intrinsic dimension estimator. The accuracy of reconstruction is then compared with models that take a subset of static catchment attributes (both climate and landscape attributes) in addition to meteorological forcing variables. Our findings suggest that attributes gathered by experts encompass nearly all pertinent information regarding the input/output relationship. This information can be succinctly summarized with merely three independent streamflow features. These features exhibit a strong correlation with the baseflow index and aridity indicators, aligning with the observation that predicting streamflow in dry catchments or with a high baseflow index is more challenging. Furthermore, correlation analysis underscores the significance of soil-related and vegetation attributes. These learned features can also be associated with parameters in conceptual hydrological models such as the GR model family.

How to cite: Bassi, A., Mira, A., Höge, M., Fenicia, F., and Albert, C.: Learning Catchment Features with Autoencoders, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-9190, https://doi.org/10.5194/egusphere-egu24-9190, 2024.

15:39–15:45
Coffee break
Chairpersons: Basil Kraft, Anna Pölz, Frederik Kratzert
Spatio-temporal flood prediction
16:15–16:25
|
EGU24-566
|
HS3.4
|
ECS
|
On-site presentation
Matteo Pianforini, Susanna Dazzi, Andrea Pilzer, and Renato Vacondio

Among the non-structural strategies for mitigating the huge economic losses and casualties caused by floods, the implementation of early-warning systems based on real-time forecasting of flood maps is one of the most effective. The high computational cost associated with two-dimensional (2D) hydrodynamic models, however, prevents their practical application in this context. To overcome this drawback, “data-driven” models are gaining considerable popularity due to their high computational efficiency for predictions. In this work, we introduce a novel surrogate model based on the Transformer architecture, named FloodSformer (FS), that efficiently predicts the temporal evolution of inundation maps, with the aim of providing real-time flood forecasts. The FS model combines an encoder-decoder (2D Convolutional Neural Network) with a Transformer block that handles temporal information. This architecture extracts the spatiotemporal information from a sequence of consecutive water depth maps and predicts the water depth map at one subsequent instant. An autoregressive procedure, based on the trained surrogate model, is employed to forecast tens of future maps.

As a case study, we investigated the hypothetical inundation due to the collapse of the flood-control dam on the Parma River (Italy). Due to the absence of real inundation maps, the training/testing dataset for the FS model was generated from numerical simulations performed through a 2D shallow‐water code (PARFLOOD). Results show that the FS model is able to recursively forecast the next 90 water depth maps (corresponding to 3 hours for this case study, in which maps are sampled at 2-minute intervals) in less than 1 minute. This is achieved while maintaining an accuracy deemed entirely acceptable for real-time applications: the average Root Mean Square Error (RMSE) is about 10 cm, and the differences between ground-truth and predicted maps are generally lower than 25 cm in the floodable area for the first 60 predicted frames. In conclusion, the short computational time and the good accuracy ensured by the autoregressive procedure make the FS model suitable for early-warning systems.

How to cite: Pianforini, M., Dazzi, S., Pilzer, A., and Vacondio, R.: A Transformer-Based Data-Driven Model for Real-Time Spatio-Temporal Flood Prediction, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-566, https://doi.org/10.5194/egusphere-egu24-566, 2024.

16:25–16:35
|
EGU24-8102
|
HS3.4
|
ECS
|
On-site presentation
Tabea Cache, Milton S. Gomez, Jovan Blagojević, Tom Beucler, João P. Leitão, and Nadav Peleg

Predicting future flood hazards in a changing climate requires adopting a stochastic framework due to the multiple sources of uncertainties (e.g., from climate change scenarios, climate models, or natural variability). This requires performing multiple flood inundation simulations which are computationally costly. Data-driven models can help overcome this issue as they can emulate urban flood maps considerably faster than traditional flood simulation models. However, their lack of generalizability to both terrain and rainfall events still limits their application. Additionally, these models face the challenge of not having sufficient training data. This led state-of-the-art models to adopt a patch-based framework, where the study area is first divided into local patches (i.e., broken into smaller terrain images) that are subsequently merged to reconstruct the whole study area prediction. The main drawback of this method is that the model is blind to the surroundings of the local patch. To overcome this bottleneck, we developed a new deep learning model that includes patches' contextual information while keeping high-resolution information of the local patch. We trained and tested the model in the city of Zurich, at spatial resolution of 1 m. The evaluation focused on 1-hour rainfall events at 5 min temporal resolution and encompassing extreme precipitation return periods from 2- to 100-year. The results show that the proposed CNN-attention model outperforms the state-of-the-art patch-based urban flood emulator. First, our model can faithfully represent flood depths for a wide range of extreme rainfall events (peak rainfall intensities ranging from 42.5 mm h-1 to 161.4 mm h-1). Second, the model's terrain generalizability was assessed in distinct urban settings, namely Luzern and Singapore. Our model accurately identifies water accumulation locations, which constitutes an improvement compared to current models. Using transfer learning, the model was successfully retrained in the new cities, requiring only a single rainfall event to adapt the model to new terrains while preserving adaptability across diverse rainfall conditions. Our results suggest that by integrating contextual terrain information with local terrain patches, our proposed model effectively generates high-resolution urban pluvial flood maps, demonstrating applicability across varied terrains and rainfall events.

How to cite: Cache, T., Gomez, M. S., Blagojević, J., Beucler, T., Leitão, J. P., and Peleg, N.: Improving the Generalizability of Urban Pluvial Flood Emulators by Contextualizing High-Resolution Patches, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-8102, https://doi.org/10.5194/egusphere-egu24-8102, 2024.

16:35–16:45
|
EGU24-20907
|
HS3.4
|
On-site presentation
Julian Hofmann and Adrian Holt

The domain of spatial flood prediction is dominated by hydrodynamic models, which, while robust and adaptable, are often constrained by computational requirements and slow processing times. To address these limitations, the integration of Deep Learning (DL) models has emerged as a promising solution, offering the potential for rapid prediction capabilities, while maintaining a high output quality. However, a critical challenge with DL models lies in their requirement for retraining for each new domain area, based on the outputs of hydrodynamic simulations generated for that specific region. This need for domain-specific retraining hampers the scalability and quick deployment of DL models in diverse settings. Our research focuses on bridging this gap by developing a fully generalized DL model for flood prediction.

FloodWaive's approach pivots on creating a DL model that can predict flood events rapidly and accurately across various regions without requiring retraining for each new domain area. The model is trained on a rich dataset derived from numerous hydrodynamic simulations, encompassing a wide spectrum of topographical conditions. This training is designed to enable the model to generalize its predictive capabilities across different domains and weather patterns, thus overcoming the traditional limitation of DL models in this field.

Initial findings from the development phase are promising, showcasing the model's capability to process complex data and provide quick, accurate flood predictions. The success of this fully generalized DL modeling approach could revolutionize applications of flood predictions such as flood forecasting and risk analysis. Regarding the later, real-time evaluation of flood protection measures could become a reality. This would empower urban planners, emergency response teams, and environmental agencies with the ability to make informed decisions quickly, potentially saving lives and reducing economic losses.

While this project is still in its developmental stages, the preliminary results point towards a significant leap in flood forecasting technology. The ultimate goal is to offer a universally deployable, real-time flood prediction tool, significantly enhancing our ability to mitigate the impact of floods worldwide.

  

How to cite: Hofmann, J. and Holt, A.: Revolutionizing Flood Forecasting with a Generalized Deep Learning Model, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-20907, https://doi.org/10.5194/egusphere-egu24-20907, 2024.

16:45–16:52
Temperature
16:52–17:02
|
EGU24-18073
|
HS3.4
|
ECS
|
On-site presentation
Ryan S. Padrón, Massimiliano Zappa, and Konrad Bogner

Stream water temperatures influence aquatic biodiversity, agriculture, tourism, electricity production, and water quality. Therefore, stakeholders would benefit from an operational forecasting service that would support timely action. Deep Learning methods are well-suited for this task as they can provide probabilistic forecasts at individual stations of a monitoring network. Here we train and evaluate several state-of-the-art models using 10 years of data from 55 stations across Switzerland. Static features (e.g. station coordinates, catchment mean elevation, area, and glacierized fraction), time indices, meteorological and/or hydrological observations from the past 64 days, and their ensemble forecasts for the following 32 days are included as predictors in the models to estimate daily maximum water temperature for the next 32 days. We find that the Temporal Fusion Transformer (TFT) model performs best for all lead times with a cumulative rank probability score (CRPS) of 0.73 ºC averaged over all stations, lead times and 90 forecasts distributed over 1 full year. The TFT is followed by the Recurrent Neural Network (CRPS = 0.77 ºC), Neural Hierarchical Interpolation for Time Series (CRPS = 0.80 ºC), and Multi-layer Perceptron (CRPS = 0.85 ºC). All models outperform the benchmark ARX model. When factoring out the uncertainty stemming from the meteorological ensemble forecasts by using observations instead, the TFT improves to a CRPS of 0.43 ºC, and it remains the best of all models. In addition, the TFT model identifies air temperature and time of the year as the most relevant predictors. Furthermore, its attention feature suggests a dominant response to more recent information in the summer, and to information from the previous month during spring and autumn. Currently, daily maximum water temperature probabilistic forecasts are produced twice per week and made available at https://drought.ch/de/allgemeine-lage/wassertemperatur/fliessgewaesser-1-1.html. 

How to cite: Padrón, R. S., Zappa, M., and Bogner, K.: Operational stream water temperature forecasting with a temporal fusion transformer model, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-18073, https://doi.org/10.5194/egusphere-egu24-18073, 2024.

17:02–17:12
|
EGU24-9446
|
HS3.4
|
ECS
|
On-site presentation
Longhao Wang, Yongqiang Zhang, and Xuanze Zhang

Sea surface temperature (SST) is a critical parameter in the global ocean-atmospheric system, exerting a substantial impact on climate change and extreme weather events like droughts and floods. The precise forecasting of future SSTs is thus vital for identifying such weather anomalies. Here we present a novel three-dimensional (3D) neural network model based on self-attention mechanisms and Swin-Transformer for mid-term SST predictions. This model, integrating both climatic and temporal features, employs self-attention to proficiently capture the temporal dynamics and global patterns in SST. This approach significantly enhances the model's capability to detect and analyze spatiotemporal changes, offering a more nuanced understanding of SST variations. Trained on 59 years of global monthly ERA5-Land reanalysis data, our model demonstrates strong deterministic forecast capabilities in the test period. It employs a convolution strategy and global attention mechanism, resulting in faster and more accurate training compared to traditional methods, such as Convolutional Neural Network with Long short-term memory (CNN-LSTM). The effectiveness of this SST prediction model highlights its potential for extensive multidimensional modelling applications in geosciences.

How to cite: Wang, L., Zhang, Y., and Zhang, X.: Skilful prediction of mid-term sea surface temperature using 3D self-attention-based neural network, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-9446, https://doi.org/10.5194/egusphere-egu24-9446, 2024.

17:12–17:18
Miscellaneous
17:18–17:28
|
EGU24-17543
|
HS3.4
|
On-site presentation
Pascal Horton, Markus Mosimann, Severin Kaderli, Olivia Martius, Andreas Paul Zischg, and Daniel Steinfeld

Surface water floods are responsible for a substantial amount of damage to buildings, yet they have received less attention than fluvial floods. Nowadays, both research and insurance companies are increasingly focusing on these phenomena to enhance knowledge and prevention efforts. This study builds upon pluvial-related damage data provided by the Swiss Mobiliar Insurance Company and the Building Insurance of Canton Zurich (GVZ) with the goal of developing a data-driven model for predicting potential damages in future precipitation events.

This work is a continuation of a previous method applied to Swiss data, relying on thresholds based on the quantiles of precipitation intensity and event volume, which, however, resulted in an excessive number of false alarms. First, a logistic regression has been assessed using different characteristics of the precipitation event. Subsequently, a random forest was established, incorporating terrain attributes to better characterize local conditions. Finally, a deep learning model was developed to account for the spatio-temporal properties of the precipitation fields on a domain larger than the targeted 1 km cell. The deep learning model comprises a convolutional neural network (CNN) for 4D precipitation data and subsequent dense layers, incorporating static attributes. The model has been applied to predict the probability of damage occurrence, as well as the damage degree quantified by the number of claims relative to the number of insured buildings.

How to cite: Horton, P., Mosimann, M., Kaderli, S., Martius, O., Zischg, A. P., and Steinfeld, D.: Deep-learning-based prediction of damages related to surface water floods for impact-based warning, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-17543, https://doi.org/10.5194/egusphere-egu24-17543, 2024.

17:28–17:38
|
EGU24-15248
|
HS3.4
|
ECS
|
On-site presentation
Radosław Szostak, Mirosław Zimnoch, Przemysław Wachniew, Marcin Pietroń, and Paweł Ćwiąkała

Unmanned aerial vehicle (UAV) photogrammetry allows the generation of orthophoto and digital surface model (DSM) rasters of a terrain. However, DSMs of water bodies mapped using this technique often reveal distortions in the water surface, thereby impeding the accurate sampling of water surface elevation (WSE) from DSMs. This study investigates the capability of deep neural networks to accommodate the aforementioned perturbations and effectively estimate WSE from photogrammetric rasters. Convolutional neural networks (CNNs) were employed for this purpose. Three regression approaches utilizing CNNs were explored: i) direct regression employing an encoder, ii) prediction of the weight mask using an encoder-decoder architecture, subsequently used to sample values from the photogrammetric DSM, and iii) a solution based on the fusion of the two approaches. The dataset employed in this study comprises data collected from five case studies of small lowland streams in Poland and Denmark, consisting of 322 DSM and orthophoto raster samples. Each sample corresponds to a 10 by 10 meter area of the stream channel and adjacent land. A grid search was employed to identify the optimal combination of encoder, mask generation architecture, and batch size among multiple candidates. Solutions were evaluated using two cross-validation methods: stratified k-fold cross-validation, where validation subsets maintained the same proportion of samples from all case studies, and leave-one-case-out cross-validation, where the validation dataset originates entirely from a single case study, and the training set consists of samples from other case studies. The proposed solution was compared with existing methods for measuring water levels in small streams using a drone. The results indicate that the solution outperforms previous photogrammetry-based methods and is second only to the radar-based method, which is considered the most accurate method available.

This research was funded by National Science Centre, Poland, project WATERLINE (2020/02/Y/ST10/00065), under the CHISTERA IV programme of the EU Horizon 2020 (Grant no 857925).

How to cite: Szostak, R., Zimnoch, M., Wachniew, P., Pietroń, M., and Ćwiąkała, P.: Estimation of Small Stream Water Surface Elevation Using UAV Photogrammetry and Deep Learning, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-15248, https://doi.org/10.5194/egusphere-egu24-15248, 2024.

17:38–17:48
|
EGU24-811
|
HS3.4
|
ECS
|
On-site presentation
Hyekyeng Jung, Chris Soulsby, and Dörthe Tetzlaff

Water flows and related mixing dynamics in the unsaturated zone are difficult to measure directly, so stable water isotope tracers have been used successfully to quantify flux and storage dynamics and to constrain process-based hydrological models as proxy data. In this study, a data-driven model based on deep learning was adapted to interpolate and extrapolate spatio-temporal isotope signals of δ18O and δ2H in soil water in three dimensions. Further, this was also used to help quantify evapotranspiration and groundwater recharge processes in the unsaturated zone. To consider both spatial and temporal dependencies of water isotope signals in the model design, the output space was decomposed into temporal basis functions and spatial coefficients using singular value decomposition. Then, temporal functions and spatial coefficients were predicted separately by specialized deep learning models in interdependencies among target data, such as the LSTM model and convolutional neural network. Finally, the predictions by the models were integrated and analyzed post-hoc using XAI tools.

Such an integrated framework has the potential to improve understanding of model behavior based on features (e.g., climate, hydrological component) connected to either temporal or spatial information. Furthermore, the model can serve as a surrogate model for process-based hydrological models, improving the use of process-based models as learning tools.

How to cite: Jung, H., Soulsby, C., and Tetzlaff, D.: A deep learning approach for spatio-temporal prediction of stable water isotopes in soil moisture, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-811, https://doi.org/10.5194/egusphere-egu24-811, 2024.

17:48–18:00

Posters on site: Thu, 18 Apr, 10:45–12:30 | Hall A

Display time: Thu, 18 Apr 08:30–Thu, 18 Apr 12:30
Chairpersons: Daniel Klotz, Martin Gauch, Basil Kraft
A.56
|
EGU24-571
|
HS3.4
|
ECS
|
|
Farzad Hosseini Hossein Abadi, Cristina Prieto Sierra, Grey Nearing, Cesar Alvarez Diaz, and Martin Gauch

Abstract

Hydrological modeling of flashy catchments, susceptible to floods, represents a significant practical challenge.  Recent application of deep learning, specifically Long Short-Term Memory networks (LSTMs), have demonstrated notable capability in delivering accurate hydrological predictions at daily and hourly time intervals (Gauch et al., 2021; Kratzert et al., 2018).

In this study, we leverage a multi-timescale LSTM (MTS-LSTM (Gauch et al., 2021)) model to predict hydrographs in flashy catchments at hourly time scales. Our primary focus is to investigate the influence of model hyperparameters on the performance of regional streamflow models. We present methodological advancements using a practical application to predict streamflow in 40 catchments within the Basque Country (North of Spain).

Our findings show that 1) hourly and daily streamflow predictions exhibit high accuracy, with Nash-Sutcliffe Efficiency (NSE) reaching values as high as 0.941 and 0.966 for daily and hourly data, respectively; and 2) hyperparameters associated with the length of the input sequence exert a substantial influence on the performance of a regional model. Consistently optimal regional values, following a systematic hyperparameter tuning, were identified as 3 years for daily data and 12 weeks for hourly data. Principal component analysis (PCA) shows that the first principal component explains 12.36% of the variance among the 12 hyperparameters. Within this set of hyperparameters, the input sequence lengths for hourly data exhibit the highest load in PC1, with a value of -0.523; the load of the input sequence length for daily data is also very high (-0.36). This suggests that these hyperparameters strongly contribute to the model performance.

Furthermore, when utilizing a catchment-scale magnifier to determine optimal hyperparameter settings for each catchment, distinctive sequence lengths emerge for individual basins. This underscores the necessity of customizing input sequence lengths based on the “uniqueness of the place” (Beven, 2020), suggesting that each catchment may demand specific hydrologically meaningful daily and hourly input sequence lengths tailored to its unique characteristics. In essence, the true input sequence length of a catchment may encapsulate hydrological information pertaining to water transit over short and long-term periods within the basin. Notably, the regional daily sequence length aligns with the highest local daily sequence values across all catchments.

In summary, our investigation stresses the critical role of the input sequence length as a hyperparameter in LSTM networks. More broadly, this work is a step towards a better understanding and achieving accurate hourly predictions using deep learning models.

 

Keywords

Hydrological modeling; Streamflow Prediction; LSTM networks; Hyperparameters configurations; Input sequence lengths

 

References:

Beven, K. (2020). Deep learning, hydrological processes and the uniqueness of place. Hydrological Processes, 34(16), 3608–3613. doi:10.1002/hyp.13805

Gauch, M., Kratzert, F., Klotz, D., Nearing, G., Lin, J., and Hochreiter, S. (2021). Rainfall–runoff prediction at multiple timescales with a single Long Short-Term Memory network, Hydrol. Earth Syst. Sci., 25, 2045–2062, DOI:10.5194/hess-25-2045-2021.

Kratzert, F., Klotz, D., Brenner, C., Schulz, K., & Herrnegger, M. (2018). Rainfall--runoff modelling using Long Short-Term Memory (LSTM) networks. Hydrology and Earth System Sciences, 22(11), 6005–6022. DOI:10.5194/hess-22-6005-2018.

 

How to cite: Hosseini Hossein Abadi, F., Prieto Sierra, C., Nearing, G., Alvarez Diaz, C., and Gauch, M.: Hydrological Significance of Input Sequence Lengths in LSTM-Based Streamflow Prediction, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-571, https://doi.org/10.5194/egusphere-egu24-571, 2024.

A.57
|
EGU24-1497
|
HS3.4
|
ECS
Operational low-flow forecasting using Long Short-Term Memory networks
(withdrawn)
Jing Deng, Anaïs Couasnon, Ruben Dahm, Markus Hrachowitz, Klaas-Jan van Heeringen, Hans Korving, Albrecht Weerts, and Riccardo Taormina
A.58
|
EGU24-2872
|
HS3.4
|
ECS
|
Arash Rahi, Mehdi Rahmati, Jacopo Dari, and Renato Morbidelli

This research examines the effectiveness of Long Short-Term Memory (LSTM) models in predicting runoff coefficient (Rc) within the Rur basin at the Stah outlet (Germany) during the period from 1961 to 2021; monthly data of temperature (T), precipitation (P), soil water storage (SWS), and total evaporation (ETA) are used as an input. Because of the complexity in predicting undecomposed Rc time series due to noise, a novel approach incorporating discrete wavelet transform (DWT) to decompose the original Rc at five levels is proposed.

The investigation identifies overfitting challenges at level-1, gradually mitigated in subsequent decomposition levels, particularly in level-2, while other levels remain tuned. Reconstructing Rc using modelled decomposition coefficients yielded Nash-Sutcliffe efficacy (NSE) values of 0.88, 0.79, and 0.74 for the training, validation, and test sets, respectively. Comparative analysis highlights that modelling undecomposed Rc with LSTM yields to a minor accuracy, emphasizing the pivotal role of decomposition techniques in tandem with LSTM for enhanced model performances.

This study provides novel insights to address challenges related to noise effects and temporal dependencies in Rc modelling; through a comprehensive analysis of the interplay between atmospheric conditions and observed data, the research contributes in advancing predictive modelling in hydrology.

How to cite: Rahi, A., Rahmati, M., Dari, J., and Morbidelli, R.: Runoff coefficient modelling using Long Short-Term Memory (LSTM) in the Rur catchment, Germany, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-2872, https://doi.org/10.5194/egusphere-egu24-2872, 2024.

A.59
|
EGU24-4812
|
HS3.4
|
ECS
Wenyu Ouyang, Xuezhi Gu, Lei Ye, Xiaoning Liu, and Chi Zhang

Despite advances in hydrological Deep Learning (DL) models using Single Task Learning (STL), the intricate relationships among multiple hydrological components and model inputs might not be comprehensively encapsulated. This study employed a Long Short-Term Memory (LSTM) neural network and the CAMELS dataset to develop a Multi-Task Learning (MTL) model, predicting streamflow and evapotranspiration across multiple basins. An optimal multi-task loss weight ratio was determined manually during the validation phase for all 591 selected basins with streamflow data-gaps under 5%. During test period, MTL showed median Nash-Sutcliffe Efficiency predictions for streamflow and evapotranspiration at 0.69 and 0.92, consistent with two STL models. The MTL's strength appeared when predicting the non-target variable, surface soil moisture, using probes derived from LSTM cell states—representative of the internal DL model workings. This prediction showed a median correlation coefficient of 0.90, surpassing the 0.88 and 0.89 achieved by the streamflow and evapotranspiration STL models, respectively. This outcome suggests that MTL models could reveal additional rules aligned with hydrological processes through the inherent correlations among multiple hydrological variables, thereby enhancing their reliability. We termed this as "variable synergy," where MTL can simultaneously predict varied targets with comparable STL performance, augmented by its robust internal representation. Harnessing this, MTL promises enhanced predictions for high-cost observational variables and a comprehensive hydrological model.

How to cite: Ouyang, W., Gu, X., Ye, L., Liu, X., and Zhang, C.: Exploring Variable Synergy in Multi-Task Deep Learning for Hydrological Modeling, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-4812, https://doi.org/10.5194/egusphere-egu24-4812, 2024.

A.60
|
EGU24-5625
|
HS3.4
|
ECS
Mohammad Afsar and Yan Jin

Understanding and predicting salinity dynamics in marshlands is crucial for effective management and biodiversity conservation. Those ecosystems are facing increasing threats due to climate change and anthropogenic activities as they are located at the interface of terrestrial and aquatic environments. Here we present a cutting-edge study on the forecasting of salinity changes in St. Jones Reserve (39.10 N, 75.44 W) in Dover, DE utilizing a recurrent neural network-based Long Short-Term Memory (LSTM) model. Our overall objectives are to enhance the accuracy and robustness of salinity forecasts and facilitate informed decision-making for marshland conservation and sustainable land use. Model input features include measured groundwater and salinity levels as measured by Solinst 3001 Levelogger LTC 5 and 3001 Barologger 5 sensors at four sites along the transects. All the sensors collected data at a 15-minute interval from March 2022. The key contribution of this research includes the development of a novel LSTM architecture optimized marshland environments, as well as the integration of advanced feature engineering techniques to enhance model performance. The methodology is validated through extensive experimentation, utilizing historical salinity data from diverse marshland locations worldwide. Results demonstrate the model’s ability to outperform existing forecasting methods, providing timely and accurate predictions of salinity changes. Furthermore, the research explores the potential for real-time monitoring applications, aiding land managers and policymakers in implementing strategies to mitigate the impact of salinity fluctuations on marshland ecosystems.

How to cite: Afsar, M. and Jin, Y.: Forecast Salinity Changes in Coastal Wetland Using Deep Learning-based LSTM Model, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-5625, https://doi.org/10.5194/egusphere-egu24-5625, 2024.

A.61
|
EGU24-7186
|
HS3.4
Dongkyun Kim

This research created a deep neural network (DNN)-based hydrologic model for an urban watershed in South Korea using multiple LSTM (long short-term memory) units and a fully connected layer. The model utilized 10-minute intervals of radar-gauge composite precipitation and temperature data across 239 grid cells, each 1 km in resolution, to simulate watershed flow discharge every 10 minutes. It showed high accuracy during both the calibration (2013–2016) and validation (2017–2019) periods, with Nash–Sutcliffe efficiency coefficient values of 0.99 and 0.67, respectively. Key findings include: 1) the DNN model's runoff–precipitation ratio map closely matched the imperviousness ratio map from land cover data, demonstrating the model's ability to learn precipitation partitioning without prior hydrological information; 2) it effectively mimicked soil moisture-dependent runoff processes, crucial for continuous hydrologic models; and 3) the LSTM units displayed varying temporal responses to precipitation, with units near the watershed outlet responding faster, indicating the model's capability to differentiate between hydrological components like direct runoff and groundwater-driven baseflow.

How to cite: Kim, D.: Exploring How Machines Model Water Flow: Predicting Small-Scale Watershed Behavior in a Distributed Setting, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-7186, https://doi.org/10.5194/egusphere-egu24-7186, 2024.

A.62
|
EGU24-9573
|
HS3.4
|
ECS
|
Safa Mohammed and Ahmed Nasr

Traditional hydrological models have long served as the standard for predicting streamflow across temporal and spatial domains. However, a persistent challenge in modelling lies in mitigating bias inherent in streamflow estimation due to both random and systemic errors in the employed model. Removal of this bias is pivotal for effective water resources management and resilience against extreme events, especially amidst evolving climate conditions. An innovative solution to address this challenge involves the integration of hydrological models with deep learning methods, known as hybridisation. Long Short-Term Memory networks (LSTM), have emerged as a promising and efficient approach to enhancing streamflow estimation. This study focuses on coupling LSTM with a physically distributed model, Wflow_sbm, to serve as a post-processor aimed at reducing modelling errors. The coupled Wflow_sbm-LSTM model was applied to the Boyne catchment in Ireland, utilising a dataset spanning two decades, divided into training, validation, and testing sets to ensure robust model evaluation. Predictive performance was rigorously assessed using metrics like Modified Kling-Gupta Efficiency (MKGE) and Nash-Sutcliffe Efficiency (NSE), with observed streamflow discharges as the target variable. Results demonstrated that the coupled model outperformed the best-calibrated Wflow_sbm model in the study catchment based on the performance measures. The enhanced prediction of extreme events by the coupled Wflow_sbm-LSTM model strengthens the case for its integration into an operational river flow forecasting framework. Significantly, Wflow is endorsed by the National Flood Forecast Warning Service (NFFWS) in Ireland as a recommended model for streamflow simulations, specifically designed for fluvial flood forecasting. Consequently, our proposed Wflow_sbm-LSTM coupled model presents a compelling opportunity for integration into the NFFWS. With demonstrated potential to achieve precise streamflow estimations, this integration holds promise for significantly enhancing the accuracy and effectiveness of flood predictions in Ireland.

How to cite: Mohammed, S. and Nasr, A.: Advancing Streamflow Modelling: Bias Removal in Physically-Based Models with the Long Short-Term Memory Networks (LSTM) Algorithm, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-9573, https://doi.org/10.5194/egusphere-egu24-9573, 2024.

A.63
|
EGU24-10506
|
HS3.4
|
Neha Vinod, Arathy Nair Geetha Raveendran, Adarsh Sankaran, and Anandu Kochukattil Ajith

In response to the critical demand for accurate streamflow predictions in hydrology, this study introduces a Sparse Autoencoder-based Long Short-Term Memory (SA-LSTM) framework applied to daily streamflow data from three-gauge stations within the Greater Pamba River Basin of Kerala, India, which was the worst affected region by the devastating floods of 2018. The SA-LSTM model addresses the challenge of feature selection from an extensive set of corresponding 1 to 7 days lagged climatic variables, such as precipitation, maximum and minimum temperatures, by incorporating a sparsity constraint. This constraint strategically guides the autoencoder to focus on the most influential features for the prediction analysis. The prediction process involves training the SA-LSTM model on historical streamflow data and climatic variables, allowing the model to learn intricate patterns and relationships. Furthermore, this study includes a comparative analysis featuring the Random Forest (RF)-LSTM model, where the RF model is employed for feature extraction, and a separate LSTM model is used for streamflow prediction. While the RF-LSTM combination demonstrates competitive performance, it is noteworthy that the SA-LSTM model consistently outperforms in terms of predictive accuracy. Rigorous evaluation metrics, including Correlation Coefficient (R2), Root Mean Square Error (RMSE), Mean Square Error (MSE), and Mean Absolute Error (MAE), highlight the SA-LSTM's forecasting accuracy across the three stations. Notably, the R2 values surpass 0.85, RMSE values remain under 12 cubic meters per second (m³/s), MSE values are below 70 (m³/s), and MAE values approach 8 m³/s. The detailed comparison between the above models underscores the superior capabilities of the SA-LSTM framework in capturing complex temporal patterns, emphasizing its potential for advancing hydrological modeling and flood risk management in flood-prone regions.

 

Key words : Streamflow, LSTM, Sparse Autoencoder, Flood, Greater Pamba

How to cite: Vinod, N., Geetha Raveendran, A. N., Sankaran, A., and Kochukattil Ajith, A.: Enhancing Hydrological Predictions: Feature-Driven Streamflow Forecasting with Sparse Autoencoder-based Long Short-Term Memory Networks, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-10506, https://doi.org/10.5194/egusphere-egu24-10506, 2024.

A.64
|
EGU24-11506
|
HS3.4
|
ECS
Laura Soncin, Claudia Bertini, Schalk Jan van Andel, Elena Ridolfi, Francesco Napolitano, Fabio Russo, and Celia Ramos Sánchez

The increased variability of water resources and the escalating water consumption contribute to the risk of stress and water scarcity in reservoirs that are typically designed based on historical conditions. Therefore, it is relevant to provide accurate forecasts of reservoir inflow to optimize sustainable water management as conditions change, especially during extreme events, such as flooding and drought. However, accurate forecasting the inflow is not straightforward, due the uncertainty of the hydrological inputs and the strong non-linearity of the system. Numerous recent studies have employed approaches based on Machine Learning (ML) techniques, such as Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), and Random Forest (RF), with successful examples of providing skilful site-specific predictions. In particular, LSTM have emerged among the pool of ML models for their performance in simulating rainfall-runoff processes, thanks to their ability to learn long-term dependencies from time series. 
Here we propose an LSTM-based approach for inflow prediction in the Barrios de Luna reservoir, located in the Spanish part of the Douro River Basin. The reservoir has a dual role, as its water is used for irrigation during dry summer periods, and its storage volume is used to mitigate floods. Therefore, in order to operate the reservoir in the short-term, Barrios de Luna reservoir operators need accurate forecast to support water management decisions in the daily and weekly time horizons. In our work, we explore the potential of a LSTM model to predict inflow in the reservoir at varying lead times, ranging from 1 day up to 4 weeks. Initially, we use as inputs past inflow, precipitation and temperature observations, and then we include meteorological forecasts of precipitation and temperature from ECMWF Extended Range. For the latter experiments, different configurations of the LSTM are tested, i.e. training the model with observations and forecasts together and training the model with observations only and fine tune it with forecasts.
Our preliminary results show that precipitation, temperature and inflow observations are all crucial inputs to the LSTM for predicting inflow, and meteorological forecast inputs seem to improve performance for the longer lead-times of one week up to a month.
Predictions developed will contribute to the Douro case study of the CLImate INTelligence (CLINT) H2020 project.

How to cite: Soncin, L., Bertini, C., van Andel, S. J., Ridolfi, E., Napolitano, F., Russo, F., and Ramos Sánchez, C.: Forecasting reservoir inflows with Long Short-Term Memory models, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-11506, https://doi.org/10.5194/egusphere-egu24-11506, 2024.

A.65
|
EGU24-11768
|
HS3.4
|
ECS
Manuel Traub, Fedor Scholz, Thomas Scholten, Christiane Zarfl, and Martin V. Butz

In the era of big data, managing and storing large-scale meteorological datasets is a critical challenge. We focus on high-resolution rainfall data, which is crucial to atmospheric sciences, climate research, and real-time weather forecasting. This study introduces a deep learning-based approach to compress the German Radar-Online-Aneichung (RADOLAN) rainfall dataset. We achieve a compression ratio of 200:1 while maintaining a minimal mean squared reconstruction error (MSE). Our method combines a convolutional autoencoder with a novel binarization mechanism, to compress data from a resolution of 900x900 pixels at 32-bit depth to 180x180 pixels at 4-bit depth. Leveraging the ConvNeXt architecture (Zhuang Liu, et al., 'A ConvNet for the 2020s'), our method learns a convolutional autoencoder for enhanced meteorological data compression. ConvNeXt introduces key architectural modifications, such as revised layer normalization and expanded receptive fields, taking inspiration from Vision Transformer to form a modern ConvNet. Our novel binarization mechanism, pivotal for achieving the high compression ratio, operates by dynamically quantizing the latent space representations using a novel magnitude specific noise injection technique. This quantization not only reduces the data size but also preserves crucial meteorological information as our low reconstruction MSE demonstrates. Beyond rainfall data, our approach shows promise for other types of high-resolution meteorological datasets, such as temperature, humidity, etc. Adapting our method to these modalities could further streamline the data management processes in meteorological deep learning scenarios and thus facilitate efficient storage and processing of diverse meteorological datasets.

How to cite: Traub, M., Scholz, F., Scholten, T., Zarfl, C., and Butz, M. V.: High-Efficiency Rainfall Data Compression Using Binarized Convolutional Autoencoder, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-11768, https://doi.org/10.5194/egusphere-egu24-11768, 2024.

A.66
|
EGU24-12293
|
HS3.4
|
ECS
|
Everett Snieder and Usman Khan

Machine learning has extensively been applied to for flow forecasting in gauged basins. Increasingly, models generating forecasts in some basin(s) of interest are trained using data from beyond the study region. With increasingly large hydrological datasets, a new challenge emerges: given some region of interest, how do you select which basins to include among the training dataset?

There is currently little guidance on selecting data from outside the basin(s) under study. An intuitive approach might be to select data from neighbouring basins, or basins with similar hydrological characteristics. However, a growing body of research suggests that including hydrologically dissimilar basins can in fact produce greater improvements to model generalisation. In this study, we use clustering as a simple yet effective method for identifying temporal and spatial hydrological diversity within a large hydrological dataset. The clustering results are used to generate information-rich subsets of data, that are used for model training. We compare the effects that basin subsets, that represent various hydrological characteristics, have on model generalisation.
Our study shows that data within individual basins, and between hydrologically similar basins, contain high degrees of redundancy. In such cases, training data can be heavily undersampled with no adverse effects – or even moderate improvements to model performance. We also show that spatial hydrological diversity can hugely benefit model training, providing improved generalisation and a regularisation effect.

How to cite: Snieder, E. and Khan, U.: Towards improved spatio-temporal selection of training data for LSTM-based flow forecasting models in Canadian basins, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-12293, https://doi.org/10.5194/egusphere-egu24-12293, 2024.

A.67
|
EGU24-13353
|
HS3.4
|
Mostafa Saberian and Vidya Samadi

Accurately predicting streamflow poses a considerable challenge particularly for intense storm events occurring across complex river systems. To tackle this issue, we developed multiple deep neural network models including a Long Short-term Memory (LSTM) and Neural Hierarchical Interpolation for Time Series Forecasting (N-HiTS) to predict short duration (1-hour) flood hydrographs. LSTM excels in preserving prolonged dependencies in structured time series data, while N-HiTS introduces an innovative deep neural architecture characterized by backward and forward residual links and a deep stack of fully connected layers. In addition, N-HiTS employs a combination of multi-rate sampling and multi-scale synthesis of predictions, resulting in a hierarchical forecasting structure that reduces computational requirements and enhances accuracy. Our goal was to evaluate the robustness and effectiveness of these advanced algorithms by comparing them with the National Water Model (NWM) forecast, across a large and complex river system i.e., the Wateree River Basin in South Carolina, USA. The models were trained and tested using precipitation, temperature, humidity, and solar radiation data during the periods 01/01/2009 to 09/30/2022 and 10/1/2022 to 01/01/2024, respectively.  Analysis suggests that N-HiTS showcased state-of-the-art performance and enhanced hourly flood forecasting accuracy by approximately 10% compared to LSTM and NWM with a negligible difference in computational costs. N-HiTS was able to more accurately forecast time to peak and peak rate values of hourly flood hydrographs compared to the LSTM and NWM. Our extensive experiments revealed the importance of multi-rate input sampling and hierarchical interpolation approaches designed within the N-HiTS model that drastically improved the flood forecasting and interpretability of the predictions.

How to cite: Saberian, M. and Samadi, V.: Flood Prediction Using Deep Neural Networks Across a Large and Complex River System , EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-13353, https://doi.org/10.5194/egusphere-egu24-13353, 2024.

A.68
|
EGU24-13848
|
HS3.4
Jieun Kang, Gayeong Lee, Sooho Park, Chunggil Jung, and Juyeong Yu

We propose a hybrid deep learning model that combines long short-term memory networks (LSTMs) to capture both spatial and temporal dependencies in the river system. The LSTM component processes spatial information derived from topographical data and river network characteristics, allowing the model to understand the physical layout of the river basin. Simultaneously, the LSTM component exploits temporal patterns in historical dam release and rainfall data, enabling the model to discern the dynamics of flood propagation. In comparison of previous study, previous results accepted only hydrological models such as HECRAS, FLDWAV, FLUMEN. But, this study accept combination of HECRAS and Deep Learning algorithm, LSTM. The goal of this study is to predict the river highest level and travel time by dam release 3 to 6 hours in advance throughout the Seomjin river basin. In order to achieve, this study conducted hydrological modeling (HECRAS) and developed a deep learning algorithm (LSTM). Afterward, the developed model combining HECRAS and LSTM was verified at six flood alert stations. Finally, the models will provide the river highest level and travel time information up to 6 hours in advance at six flood alert stations. To train and validate the model, we compile a comprehensive dataset of historical dam release events and corresponding flood travel times from a range of river basins. The dataset includes various hydrological and meteorological features to ensure the model's robustness in handling diverse scenarios. The deep learning model is then trained using a subset of the data and validated against unseen events to assess its generalization capabilities. Preliminary results indicate that the hybrid HECRAS-LSTM model outperforms traditional hydrological models in predicting flood travel times. The model exhibits improved accuracy, particularly in cases of complex river geometries and extreme weather events. Additionally, the model demonstrates its potential for real-time forecasting, as it can efficiently process and assimilate incoming data. In conclusion, our study showcases the effectiveness of using a hybrid HECRAS-LSTM model for forecasting flood travel time by dam release. By leveraging the power of deep learning, we pave the way for more precise and reliable flood predictions, contributing to the overall resilience and safety of communities located downstream of dam-controlled river systems.

How to cite: Kang, J., Lee, G., Park, S., Jung, C., and Yu, J.: The Development of Forecasting System Flood Travel Time by Dam Release for Supplying Flood Information Using Deep Learning at Flood Alert Stations in the Seomjin River Basin, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-13848, https://doi.org/10.5194/egusphere-egu24-13848, 2024.

A.69
|
EGU24-14815
|
HS3.4
|
ECS
Quantifying the Effect of Additional Training Data When Using Machine Learning to Predict Streamflow in Ungauged Basins
(withdrawn)
Alexander Werenka and Steven Weijs
A.70
|
EGU24-15073
|
HS3.4
A Hybrid Deep Learning Framework to Generate Locally Relevant Streamflow from Large Scale Hydrological Models 
(withdrawn)
Bhanu Magotra and Manabendra Saharia
A.71
|
EGU24-16234
|
HS3.4
|
ECS
Grith Martinsen, Niels Agertoft, and Phillip Aarestrup

The utilization of data-driven models in hydrology has witnessed a significant increase in recent years. The open-source philosophy underpinning much of the code developed and research being conducted has facilitated widespread access to the the hydrological community to sophisticated machine learning models and technology (Reichstein 2019). These data driven approaches to hydrological modelling has witnessed growing interest after multiple studies has shown how machine-learning models were able to outperform nationwide traditional physics-based hydrological models (Kratzerts et al. 2019). The latter often demands substantial man-hours for development, calibration and fine-tuning to accurately represent relevant hydrological processes.

In this national-scale explorative study we undertake an in-depth examination of Danish catchment hydrology. Our objective is to understand what processes and dynamics are well captured by a purely data driven model without physical constraints, namely the Entity-Aware Long Short-Term Model (EA-LSTM). The model code was developed by Kratzerts et al. (2019) and the analysis build on top of a newly published national CAMELS data set covering 301 catchments in Denmark (Koch and Schneider, 2022), with an average resolution of 130 km2.

Denmark, spanning an area of around 43 000 km2, demonstrates a relatively high data coverage. Presently more than 400 stations record water level measurements in the Danish stream network, while a network of 243 stations have collected meteorological data since 2011. These datasets maintained by the Danish Environmental Protection Agency and the Danish Meteorological Institute, respectively, and are publicly available.

Despite Denmark’s data abundance, Koch and Schneider (2022) demonstrated that the data-driven EA-LSTM model, trained with the CAMELS dataset for Denmark (from now on referred to as the DK-LSTM) were not able to outperform the traditional physics-based hydrological model, against which it was benchmarked. Consequently, performance of the DK-LSTM model could be increased by pre-training it with simulations from a national physics-based model indicating that dominating hydrological processes are not described by the readily available input data in the CAMELS dataset.

This study conducts a comprehensive analysis of Danish catchment hydrology aiming to explore three aspects: 1) the common characteristics of the catchments where the DK-LSTM performs well or encounters challenges, 2) the identification of hydrological characteristics, that exhibit improvement when informing the data-driven model with physics-based model simulations, and 3) an exploration of whether the aforementioned findings can guide us in determining necessary physical constraints and/or input variables that explains the hydrological processes for the data-driven model approach at a national scale, using the example of DK-LSTM.

 

Koch, J., and Schneider, R. Long short-term memory networks enhance rainfall-runoff modelling at the national scale of Denmark. GEUS Bulletin49. https://doi.org/10.34194/geusb.v49.8292, 2022.

Kratzert, F., Klotz, D., Shalev, G., Klambauer, G., Hochreiter, S., and Nearing, G.: Towards learning universal, regional, and local hydrological behaviors via machine learning applied to large-sample datasets, Hydrol. Earth Syst. Sci., 23, 5089-5110, https://doi.org/10.5194/hess-23-5089-2019, 2019.

Reichstein, M., Camps-Valls, G., Stevens, B. et al. Deep learning and process understanding for data-driven Earth system science. Nature 566, 195–204. https://doi.org/10.1038/s41586-019-0912-1, 2019.

How to cite: Martinsen, G., Agertoft, N., and Aarestrup, P.: A bottom-up approach to identify important hydrological processes by evaluating a national scale EA-LSTM model for Denmark, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-16234, https://doi.org/10.5194/egusphere-egu24-16234, 2024.

A.72
|
EGU24-17502
|
HS3.4
|
ECS
Marvin Höge, Florian Wenk, Andreas Scheidegger, Carlo Albert, and Andreas Frömelt

Neural Ordinary Differential Equations (ODEs) fuse neural networks with a mechanistic equation framework. This hybrid structure offers both traceability of model states and processes, as it is typical for physics-based models, and the ability of machine learning to encode new functional relations. Neural ODE models have demonstrated high potential in hydrologic predictions and scientific investigation of the related process in the hydrologic cycle, i.e. tasks of water quantity estimation (Höge et al., 2022).

This explicit representation of state variables is key to water quality modelling. There, we typically have several interrelated state variables like nitrate, nitrite, phosphorous, organic matter,…  Traditionally, these states are modelled based on mechanistic kinetic rate expressions that are often only rough approximations of the underlying dynamics. At the same time, this domain of water research suffers from data scarcity and therefore solely data-driven methods struggle to provide accurate predictions reliably. We show how to improve predictions of state dynamics and to foster knowledge gain about the processes in such interrelated systems with multiple states using Neural ODEs. 

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., Wenk, F., Scheidegger, A., Albert, C., and Frömelt, A.: Towards improved Water Quality Modelling using Neural ODE models, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-17502, https://doi.org/10.5194/egusphere-egu24-17502, 2024.

A.73
|
EGU24-21431
|
HS3.4
Miguel Barrios, Henry Rubiano, and Cristian Guevara-Ochoa

One of the latent difficulties in the fields of climatology, meteorology, and hydrology is the scarce rainfall information available due to the limited or nonexistent instrumentation of river basins, especially in developing countries where the establishment and maintenance of equipment entail high costs relative to the available budget. Hence, the importance of generating alternatives that seek to improve spatial precipitation estimation has been increasing, given the advances in the implementation of computational algorithms that involve Machine Learning techniques. In this study, a multitask convolutional neural network was implemented, composed of an encoder-decoder architecture (U-Net), which simultaneously estimates the probability of rain through a classification model and the precipitation rate through a regression model at a spatial resolution of 2 km2 and a temporal resolution of 10 minutes. The input modalities included data from rain gauge stations, weather radar, and satellite information (GOES 16). For model training,  validation, and testing, a dataset was consolidated with 3 months of information (February to April 2021) with a distribution of 70/15/15 percent, covering the effective coverage range of the Munchique weather radar located in the Andean region of Colombia. The obtained results show a Probability of Detection (POD) of 0.59 and a False Alarm Rate (FAR) of 0.39. Regarding precipitation rate estimation, it is assessed with a Root Mean  Square Error (RMSE) of 1.13 mm/10min. This research highlights the significant capability of deep learning algorithms in reconstructing and reproducing the spatial pattern of rainfall in tropical regions with limited instrumentation. However, there is a need to continue strengthening climatological monitoring networks to achieve significant spatial representativeness, thereby reducing potential biases in model estimations. 

How to cite: Barrios, M., Rubiano, H., and Guevara-Ochoa, C.: Implementation of deep learning algorithms in the sub-hourly rainfall fields estimation from remote sensors and rainfall gauge information in the tropical Andes, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-21431, https://doi.org/10.5194/egusphere-egu24-21431, 2024.

Posters virtual: Thu, 18 Apr, 14:00–15:45 | vHall A

Display time: Thu, 18 Apr 08:30–Thu, 18 Apr 18:00
Chairpersons: Frederik Kratzert, Daniel Klotz, Anna Pölz
vA.11
|
EGU24-14765
|
HS3.4
|
ECS
|
|
Akanksha Soni, Surajit Deb Barma, and Amai Mahesha

This study investigates the efficacy of Multi-Layer Perceptron (MLP) models in groundwater level modeling, specifically emphasizing the pivotal role of input data quality, particularly precipitation data. Unlike prior research that primarily focused on regional datasets like those from the India Meteorological Department (IMD), our research explores the integration of global precipitation data, specifically leveraging the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG) dataset for MLP-based modeling. The assessment was conducted using two wells in Dakshina Kannada, evaluating four MLP models (GA-MLP, EFO-MLP, PSO-MLP, AAEO-MLP) with IMERG and IMD precipitation data. Performance metrics were employed, including mean absolute error, root mean square error, normalized Nash-Sutcliffe efficiency, and Pearson's correlation index. The study also includes convergence analysis and stability assessments, revealing the significant impact of the precipitation dataset on model performance. Noteworthy findings include the superior performance of the AAEO-MLP model in training with IMD data and the GA-MLP model's outperformance in testing at the Bajpe well with both datasets. The stability of the GA-MLP model, indicated by the lowest standard deviation values in convergence analysis, underscores its reliability. Moreover, transitioning to the IMERG dataset improved model performance and reduced variability, providing valuable insights into the strengths and limitations of MLP models in groundwater-level modeling. These results advance the precision and dependability of groundwater level forecasts, thereby supporting more effective strategies for international groundwater resource management.

How to cite: Soni, A., Barma, S. D., and Mahesha, A.: Optimizing Groundwater Forecasting: Comparative Analysis of MLP Models Using Global and Regional Precipitation Data, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-14765, https://doi.org/10.5194/egusphere-egu24-14765, 2024.