HS5.4

EDI
Water resources policy and management - forecast and control methods

Highly varying hydro-climatological conditions, multi-party decision-making contexts, and the dynamic interconnection between water and other critical infrastructures create a wealth of challenges and opportunities for water resources planning and management. For example, reservoir operators must account for a number of time-varying drivers, such as the downstream users’ demands, short- and long-term water availability, electricity prices, and the share of power supplied by wind and solar technologies. In this context, adaptive and robust management solutions are paramount to the reliability and resilience of water resources systems. To this purpose, emerging work is focusing on the development of models and algorithms that adapt short-term decisions to newly available information, often issued in the form of weather or streamflow forecasts, or extracted from observational data collected via pervasive sensor networks, remote sensing, cyberinfrastructure, or crowdsourcing.

In this session, we solicit novel contributions related to improved multi-sectoral forecasts (e.g., water availability and demand, energy and crop prices), novel data analytics and machine learning tools for processing observational data, and real-time control solutions taking advantage of this new information. Examples include: 1) approaches for incorporating additional information within control problems; 2) methods for characterizing the effect of forecast uncertainty on the decision-making process; 3) integration of information with users’ preferences, behavioral uncertainty, and institutional setting; 4) studies on the scalability and robustness of optimal control algorithms. We welcome real-world examples on the successful application of these methods into decision-making practice.

Convener: Charles Rougé | Co-conveners: Louise CrochemoreECSECS, Matteo Giuliani, Stefano Galelli
Presentations
| Wed, 25 May, 10:20–11:40 (CEST)
 
Room 2.17

Presentations: Wed, 25 May | Room 2.17

Chairpersons: Louise Crochemore, Matteo Giuliani
10:20–10:30
|
EGU22-10682
|
solicited
|
Presentation form not yet defined
Alden Keefe Sampson, David Lambl, Lauren Gulland, Mostafa Elkurdy, Phillip Butcher, and Laura Read

Accurate streamflow forecasts equip water managers to adapt to changing flow regimes and constraints, increase water supply reliability, reduce flood risk, and maximize revenue. Over the 2021 water year, the Upstream Tech team took part in a live, 1-10 day ahead streamflow forecasting competition using our flow forecast system, HydroForecast. The competition was a chance to objectively compare operational forecasts using a range of modeling approaches from national agencies, hydropower utilities’ in-house teams, private forecasters and individual modelers at 19 sites in North America. HydroForecast outperformed both statistical and conceptual models and won the competition. We evaluate HydroForecast’s performance relative to other models to identify its strengths and areas for further research by region, season, and forecast horizon. We also share what our theory-guided machine learning approach to hydrologic modeling means in practice for HydroForecast, focusing on the key facets of our approach which contribute most to our accuracy. Finally, we describe the largest opportunities for further forecast accuracy gains we identified in this competition and some of the research efforts we are working on to meet those opportunities.

How to cite: Sampson, A. K., Lambl, D., Gulland, L., Elkurdy, M., Butcher, P., and Read, L.: Rumble on the River: Analyzing Model Performance in a Year-Long, Live Streamflow Forecasting Competition, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10682, https://doi.org/10.5194/egusphere-egu22-10682, 2022.

10:30–10:35
|
EGU22-9553
|
Virtual presentation
Manuel Pulido-Velazquez and Hector Macian-Sorribes

The use of artificial intelligence is growing in many science areas boosted by an unprecedent increase in data availability and the improvements in computer hardware. Its application in Earth system sciences is particularly relevant due to the existence of complex behavioural patterns whose reproduction with traditional methods is challenging. Its use in seasonal forecasting is favoured by the existence of large amounts of open meteorological data.

This study showcases the potential of artificial intelligence to downscale and post-process seasonal meteorological forecasts in semiarid river basin. The artificial intelligence methodology used is fuzzy logic. Daily raw seasonal forecasts correspond to the ECMWF SEAS5 seasonal forecasts from the Copernicus Climate Change Service (CS3), available at a 1º grid, while daily ERA5 reanalysis, available at a 0.25º grid through the C3S, is employed as observational data. The meteorological variables used are precipitation; 2-meter mean, minimum and maximum temperatures; incident shortwave solar radiation and wind speed.

The artificial intelligence algorithm is coded in a Python script. The script requires the coordinates of a target grid (that may coincide or not with the grid of observational data). For each point it performs the post-processing with the following process: 1) extracts the observational and forecast data for the closest point available; 2) computes the cdf’s of both datasets per month; 3) builds and trains fuzzy logic systems to match the forecasts cdfs to the observational cdfs; and 4) obtains the post-processed forecasts for the target grid provided. The script admits any meteorological variable, seasonal forecasting system from the C3S and observational dataset (it has been successfully tested with ERA5Land and the Spain02 gridded dataset).

The script has been applied to the semi-arid upper Tagus and Segura river basins. The Segura river basin suffers a severe overexploitation alleviated by a water transfer from the upper Tagus. The fuzzy logic systems chosen were Sugeno of order 1, with two inputs: the raw meteorological forecasts and the month of the year. The same grid as ERA5 was considered, and for each point the fuzzy logic systems were trained so that the forecasts monthly cdfs match the ones from ERA5. The training process took on average 20 minutes per point and variable with a standard computer, and results show that the post-processed cdfs closely match the ERA5 cdfs. Furthermore, the skill of the post-processed forecasts was evaluated using the Mean Absolute Error (MAE) and compared to the skill of raw forecasts to assess the adequacy of the post-processing.

Acknowledgements:  This study has received funding from the European Union’s Horizon 2020 research and innovation programme under the GoNEXUS project (grant agreement No 101003722); the eGROUNDWATER project (GA n. 1921), part of the PRIMA programme supported by the European Union’s Horizon 2020 research and innovation programme; and the subvencions del Programa per a la promoció de la investigación científica, el desenvolupament tecnològic i la innovació a la Comunitat Valenciana (PROMETEO) under the WATER4CAST project.

How to cite: Pulido-Velazquez, M. and Macian-Sorribes, H.: Post-processing of seasonal forecasts in a semi-arid river basin through artificial intelligence, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-9553, https://doi.org/10.5194/egusphere-egu22-9553, 2022.

10:35–10:40
|
EGU22-8805
|
ECS
|
Presentation form not yet defined
Gal Perelman and Barak Fishbain

In many studies arises the need to generate synthetic data sets. Such data can answer different needs as data imputation, non-stationary systems analysis, Monte Carlo simulations, training of data-driven models, uncertainty analysis and more. Previous efforts to generate synthetic data focused mostly on statistical methods which did not maintain the statistical moments of the original dataset, while producing a large number of random different time series. Here, a novel method is developed, based on signal processing and discrete Fourier transform (DFT) theory. The method allows to generate synthetic time series signals with similar statistical moments of any given signal. Moreover, the method allows control on the correlation level between the original and the synthesized signals. We also provide mathematical proofs that our method maintains the first two statistical moments. The method is illustrated on two different datasets showing that also the third and fourth moments are kept. Figure 1 shows, in blue, a true water demand time-series taken from a real-life system. For this signal, 50 synthesized signals are generated with increasing correlation levels - from top, with the lowest correlation, to bottom, presenting the highest correlations between the original and synthesized signals.

                                                                                                                    

                                                                Figure 1 – Domestic water demand signal with different correlation level

How to cite: Perelman, G. and Fishbain, B.: Synthesizing water-related time series for simulation studies while maintaining the original signal’s statistical moments, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8805, https://doi.org/10.5194/egusphere-egu22-8805, 2022.

10:40–10:45
|
EGU22-9677
|
Presentation form not yet defined
Francesca Pianosi, Andres Peñuela-Fernandez, and Charles Rougé

Over the last decade, weather and hence hydrological forecasts with lead time up to six-seven months (so called “seasonal” forecasts) have become increasingly available. One of the key intended purposes of these forecast products is to support water resources management, particularly for water supply or other operational objectives that require a lead time longer than decadal. However, examples of water agencies that have formally embedded these products in their operational practice are hardly reported to date. This is often traced back to the uncertainty and inaccuracy that affect these forecast products, particularly outside the tropics, and many have concluded that skill must increase before seasonal forecasts can be used in operational management (Jackson-Blake et al. 2021). However, a handful of simulation studies have recently suggested that forecasts value (that is, the enhancement of management performance when forecasts are integrated into decision-making procedures) may exceed their skill (that is, the ability to predict inflows correctly). This is an interesting perspective and calls for more studies to investigate the relationship between skill and value for water management, so to understand when and how value could be extracted from forecasts despite their limited skill.

With this motivation, in a previous simulation study (Penuela et al. 2020) we evaluated the potential of seasonal forecasts for improving the operations of a pumped-storage supply system in the UK. We found that the forecast value was only loosely related to skill, and that operational priorities (that is, the relative weight given to the two objectives of saving energy and reaching full capacity at the end of the filling season) and hydrological conditions (the initial reservoir storage and the overall inflow volume over the filling season) determined the forecast value more than its skill.

In this work, we use the same case study to further explore the skill-value relationship by comprehensively assessing and comparing ensemble of forecasts with different skill. First, we use a novel technique to generate synthetic ensembles of weather forecasts with similar characteristics to the original one (provided by the ECMWF seasonal forecasting systems SEAS5, in our case) and artificially increased skill – up to the ‘perfect’ forecast where all ensemble members coincide with observations. Second, for each synthetic ensemble, we generate the corresponding hydrological forecasts through a conceptual rainfall-runoff model. Last, through a nested optimisation-simulation procedure, we reconstruct the reservoir operations that would have resulted from (optimally) using those hydrological forecasts over a 11-years simulation period. We then compare resulting performances in terms of storage conservation and energy costs (the forecast ‘value’) as forecast skill increases. This helps us shed some more light on the skill-value relationship, and identify thresholds (if they exist) below which forecasts are not useful, or conversely, above which further improving skill does not significantly increase value.

Peñuela et al. (2020) Assessing the value of seasonal hydrological forecasts for improving water resource management: insights from a pilot application in the UK, HESS, 24. https://doi.org/10.5194/hess-24-6059-2020

Jackson-Blake et al. (2021) Opportunities for seasonal forecasting to support water management outside the tropics, HESSD, In review. https://doi.org/10.5194/hess-2021-443

How to cite: Pianosi, F., Peñuela-Fernandez, A., and Rougé, C.: On the relationship between forecast skill and value for water management, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-9677, https://doi.org/10.5194/egusphere-egu22-9677, 2022.

10:45–10:50
|
EGU22-5956
|
ECS
|
On-site presentation
Nicola Crippa, Guang Yang, Manolis Grillakis, Aristeidis Koutroulis, and Matteo Giuliani

Population expansion and socio-economic development have been increasing the pressure on water resources, which are often exploited in a non-renewable way. Besides, climate change can modify hydroclimatic patterns and exaggerate freshwater water stress. Flexible operation of existing water reservoirs is one of the most cost-effective ways to mitigate water-related stress, by storing water when it is abundant and releasing it when droughts persist. In this context, medium - to long-term hydroclimatic forecasts are set to cover a central role for properly informing reservoir decision-making and operation policy design. State-of-the-art hydroclimatic forecast products are indeed becoming more and more skillful at seasonal or longer lead-times, especially in regions characterized by climate teleconnections or by predictable hydrological behavior, such as baseflow- or snow-dominance. Nevertheless, the link between forecast skill and forecast value, i.e. the performance improvement obtained thanks to the use of the forecasts, is neither easily predictable nor necessarily positive. Each system indeed requires specific forecasts according to its characteristics, such as climate and hydrological regime, size of the reservoir, management objectives, and the skill of existing forecast systems do not necessarily translate into a significant gain in system performance.

In this work, we quantify the value of seasonal forecast information in informing the operations of the Faneromeni reservoir on the Crete island. The reservoir is primarily used to provide water to an important agricultural district during the dry summer season. Current operation of this reservoir is based on the available storage at the beginning of the irrigation season, which, in normal conditions, allows the supply of the irrigation demand if the reservoir is completely full; otherwise, the reservoir releases are modulated according to the storage shortage. We instead investigate alternative policies for the operations of the Faneromeni reservoir by using the Evolutionary Multi Objectives Direct Policy Search (EMODPS) method, which allows the design of flexible rules to cope with the variability of the hydrologic conditions as well as to include forecast information for conditioning operational decisions.

Preliminary results show that EMODPS policies can improve the existing operation of the Faneromeni reservoir. Moreover, these solutions also allow to mitigate the negative impacts of climate change and flexibly adapt the reservoir operations to the projected hydroclimatic conditions.

 

This work is supported by the STREAM project funded by the Prince Albert II of Monaco Foundation, grant number 2981.

How to cite: Crippa, N., Yang, G., Grillakis, M., Koutroulis, A., and Giuliani, M.: Assessing the value of seasonal forecasts in informing reservoir operations in water-stressed Mediterranean basins, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5956, https://doi.org/10.5194/egusphere-egu22-5956, 2022.

10:50–10:55
|
EGU22-9611
|
On-site presentation
Yongshin Lee, Francesca Pianosi, Miguel Rico-Ramirez, and Andres Peñuela

Increased intensity and frequency of extreme weather events predicted in a warmer climate can cause damages to human life, properties and the natural environment. Also, these changes pose significant challenges to water resources management. For example, in South-Korea, there were prolonged droughts lasting more than 3 years from 2013 to 2015 whereas, record-breaking monsoon caused severe flood damages in 2020.  Nowadays the construction of new water infrastructure such as large reservoirs is almost impossible in many developed countries due to social and environmental objections and therefore the operation of existing reservoirs is extremely important. In order to improve the performance of reservoir operations, we need to make better use of reliable weather forecasting information.

There have been noteworthy advances in seasonal climate forecasts over the last decade. Seasonal forecasts are long-term meteorological forecasts (1 to 7 months) that could be a game changer in reservoir operation and adaptation to climate change once it is demonstrated that they provide reliable information in the water sector. However, so far seasonal forecasts have never been used in practice and simulation experiments similar to those reported in the scientific literature for other regions, such as Europe or the US, have not been conducted for South Korea.

Therefore, assessing the value of seasonal forecasts in reservoir operation is highly significant matter. In this study, we will try to demonstrate their value by comparing, via model simulation, the use of forecasts with the use of diverse scenarios, including deterministic low inflow scenarios (or worst cases), Ensemble Streamflow Prediction (ESP) and perfect forecast conditions (where observations are used in place of forecasts). The analysis will be carried out for 6 different reservoirs having different catchment sizes  from 95.4km2 to 1,584km2 to determine any link between forecasts value and catchment characteristics and draw general guidelines for future forecasts use. The results from each scenario will be compared in terms of ‘skill’, representing the forecast accuracy, and ‘value’, representing the final effect on reservoir operation such as water resources availability and flood prediction. This study aims at understanding how valuable the seasonal forecasts can be and how to apply them to have better performance in practice.

How to cite: Lee, Y., Pianosi, F., Rico-Ramirez, M., and Peñuela, A.: Assessment of seasonal forecasts for reservoir operation in South Korea, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-9611, https://doi.org/10.5194/egusphere-egu22-9611, 2022.

10:55–11:00
|
EGU22-13084
|
ECS
|
Virtual presentation
Richard Laugesen, Mark Thyer, David McInerney, and Dmitri Kavetski

Forecasts have the potential to improve decision-making but have not been widely evaluated because current forecast value methods have critical limitations. The ubiquitous Relative Economic Value (REV) is limited to binary decisions, cost-loss economic model, and risk neutral decision-makers. Expected Utility Theory can flexibly model more real-world decisions, but its application in forecasting has been limited and the findings are difficult to compare. To enable a systematic comparison of these methods a new metric, Relative Utility Value (RUV), is developed based on Expected Utility Theory. It has the same interpretation as REV but is more flexible and able to handle a wider range of real-world decisions because all aspects of the decision-context are user-defined. Also, when specific assumptions are imposed it is shown that REV and RUV are equivalent. We demonstrate the key differences and similarities between the methods with a case study using probabilistic subseasonal streamflow forecasts in a catchment in the Southern Murray-Darling Basin of Australia. This showed that for most decision-makers the ensemble forecasts were more valuable than a reference climatology for all lead-times (max 30 days), decision types (binary, multi-categorical, and continuous-flow), and levels of risk aversion. Risk aversion had a mixed impact across the different decision-types and the key driver was found to be the specific decision thresholds relative to the damage function. The generality of RUV makes it applicable to any domain where forecast information is used for making decisions, and the flexibility enables forecast assessment tailored to specific decisions and decision-makers. It complements forecast verification and enables assessment of forecast systems through the lens of customer impact.

How to cite: Laugesen, R., Thyer, M., McInerney, D., and Kavetski, D.: A flexible approach for evaluating the value of probabilistic forecasts for different decision types and risk averse decision-makers, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-13084, https://doi.org/10.5194/egusphere-egu22-13084, 2022.

11:00–11:05
|
EGU22-3103
|
ECS
|
Virtual presentation
Liam Ekblad, Jonathan Herman, Scott Steinschneider, Matteo Giuliani, and Andrea Castelletti

Water infrastructure operations can adapt to both short-term variability and long-term change. Studies that have leveraged climate information to reoperate infrastructure have yet to explore the direct use of spatially distributed information in operating policy training, which could enable learning from weather patterns associated with emerging risks—for example, flood and drought events associated with atmospheric rivers or high-pressure ridges, respectively, which result from co-occurring weather and climate patterns on multiple timescales. This study investigates the potential for spatial projections from large-ensemble climate models to directly inform reservoir operating policies using a deep reinforcement learning strategy, aiming to discover flexible, climate-informed policies without prior dimension reduction, which could cause loss of information. The approach is demonstrated for Folsom Reservoir in California. We investigate how learned policies interpret spatial climate information by connecting flood control and water supply shortage operations to the sensitivity and salience patterns associated with the input images. To assess the extent to which trained policies generalize to possible future climates, policies trained on historical data are tested on held-out scenarios drawn from the same period, and their performance is compared to flood and shortage scenarios drawn from a future period. Trained policies are robust to the variability present across climate model ensembles, demonstrate value in identifying spatial climate patterns for operations, and maintain the flexibility to dynamically adapt to climate change as it occurs, illustrating a broad benefit to global infrastructure systems facing climate risks.

How to cite: Ekblad, L., Herman, J., Steinschneider, S., Giuliani, M., and Castelletti, A.: Connecting spatial climate information to infrastructure operations using deep reinforcement learning, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3103, https://doi.org/10.5194/egusphere-egu22-3103, 2022.

11:05–11:10
|
EGU22-5846
|
ECS
|
On-site presentation
Andrea Ficchì, Federico Staffa, Raffaele Giuseppe Cestari, Simone Formentin, and Andrea Castelletti

Hydro-meteorological forecasts are more and more easily available with improving skill over longer timescales and with higher spatiotemporal resolutions. Their uncertainties are commonly represented by ensemble prediction systems which are now dominant at the global and continental scale, while short-term deterministic forecasts are still used, especially in some local contexts. For effective water resources management, it is critical to understand how to use the wealth of information provided by different forecast systems over multiple timescales to help meet the water demand of key competing socio-economic sectors, while reducing short-term impacts and bringing the controlled systems to desirable states in the long term. Real-time control schemes of water reservoirs like Model Predictive Control (MPC) can help meet these goals, by providing a flexible framework to use forecasts proactively and satisfy multiple competing objectives while respecting operational constraints.

In this study, we propose a new nested multi-stage stochastic MPC framework integrating the use of both deterministic and ensemble hydrological forecasts over multiple timescales from short-term (60 hours) to seasonal (7 months ahead). We demonstrate the performance of this real-time controller for the Lake Como system, located in the Italian Alps, where a large lake is regulated mainly for irrigation supply and flood control. First, seasonal ensemble forecasts are used to solve a Tree-Based MPC (TB-MPC) problem optimising the reservoir management over several months, by adopting a tree structure to summarise the ensemble information including the resolution of uncertainty in time. Second, the decisions identified so far are used to condition daily operations over a month using sub-seasonal probabilistic forecasts (up to 46 days) under the same TB-MPC approach. Third, the decisions for the first few days are then further adapted to optimize operations three days ahead using deterministic short-term forecasts with MPC. The sub-seasonal and seasonal ensemble (re-)forecasts used are those produced by the European Flood Awareness System (EFAS) from the Copernicus Emergency Management Service which uses ensemble meteorological forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF). While EFAS is uncalibrated for the study area, we apply bias-correction techniques to improve the agreement of forecasts with local observations and allow their use for resolving the water-balance within MPC. The short-term 60-h forecasts come from a locally calibrated hydrological model (TOPKAPI) using deterministic weather forecasts from the COSMO (Consortium for Small Scale Modelling) model. The skill of all these forecasts is assessed, as well as the ensemble spread–error relationship for EFAS at different lead times. To evaluate the value of the forecasts we compare the performance of the real-time MPC controller with different benchmarks including perfect forecasts, climatology, and persistence. Finally, we investigate the link between forecast skill and value for reservoir operation, and we compare the performance of the nested MPC framework integrating multi-timescale forecasts with the MPC using single forecasts.

How to cite: Ficchì, A., Staffa, F., Cestari, R. G., Formentin, S., and Castelletti, A.: Integrated real-time control of water reservoirs with deterministic and probabilistic multi-timescale forecasts: Application to the Lake Como , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5846, https://doi.org/10.5194/egusphere-egu22-5846, 2022.

11:10–11:15
|
EGU22-5783
|
Virtual presentation
Andrea Castelletti, Dennis Zanutto, Andrea Ficchì, and Matteo Giuliani

Given the ever-growing accuracy of forecast products over different lead times, it’s increasingly important to understand how to efficiently select and use the most valuable information to support adaptive and robust management of water resources under changing hydroclimatic conditions. In this study, we investigate how the most valuable information from multi-scale forecasts can be selected and used to inform the optimal operation of multipurpose water reservoirs. Our framework combines Input Variable Selection algorithms supporting the selection of the most informative policy inputs, including different forecast variables over diverse lead times, with the Evolutionary Multi-Objective Direct Policy Search method for designing Pareto optimal control policies conditioned on forecast information.

We test this approach on the Lake Como system, a regulated lake in Northern Italy which is operated for preventing floods along the lake shores, providing irrigation supply to downstream users and avoiding low lake levels. Our approach allows the identification of the best subset or combination of variables and metrics extracted from a suite of forecast products. In particular, the performance of the system is evaluated using short-term local deterministic forecasts as well as sub-seasonal and seasonal large-scale ensemble forecasts provided by the European Flood Awareness System (EFAS), part of the Copernicus Emergency Management Service. The candidate variables proposed as inputs for the IVS include different statistics extracted from these forecasts, including the accumulated inflow up to different lead times, the maximum daily flow over different temporal scales and spatial domains, the ensemble forecast variance and some skill scores. The performance of the designed forecast-informed operating policies is contrasted against various benchmarks, including perfect forecasts and the climatology. Beside improving the operating policy performance, results are expected to provide insights about the intrinsic bias of forecast products and to highlight the role of forecast uncertainty in policy design.

How to cite: Castelletti, A., Zanutto, D., Ficchì, A., and Giuliani, M.: Extracting the most valuable information from multi-timescale hydrological forecasts for informing the operation of multipurpose water systems , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5783, https://doi.org/10.5194/egusphere-egu22-5783, 2022.

11:15–11:20
|
EGU22-12889
|
Virtual presentation
Cristina Aguilar, Rafael Pimentel, Sergio Vela, Eva Contreras, Fátima Moreno, and María José Polo

The operation of hydropower systems relies on the hydraulic conditions of the water system and their management is carried out based on certain operation rules (e.g., turbine minimum and maximum discharge) and environmental flows requirements. The effects of global change in hydropower systems are varied as expected changes in the meteorological forcing agents determine considerable modifications in the water flows that affect the amount of available water and thus, the production and profitability from hydropower plants. In this context, flash-flood events are especially relevant as quick management decisions need to be applied to minimize not only the potential damages downstream, but also the energy production losses connected to a conservative approach. The optimization of the decision-making process under flash flood events has two main challenges in these areas. On the one hand, there are several meteorological forecasting systems at different spatiotemporal scales currently available. However, the greater uncertainty linked to the rapid response time of these headwater catchments limits their use. On the other hand, the insufficient number of control points with available real time measurements (i.e., precipitation gauges and water level controls) makes it difficult to create early warning systems with an appropriate uncertainty quantification.

This study presents the definition of an early warning system that forecasts the inflow into a headwater Mediterranean catchment with a fast hydrological response. The Cala dam (59 hm3) was selected as the pilot reservoir with hydroelectric production as its main use, but also with irrigation and leisure demands. The contributing catchment (535 km2) is a good example of Mediterranean conditions in southern Spain, with agroforestry uses and a quick response to intense precipitation events due to steep slopes, shallow soils and groundwater redistribution, which does not favor the modulation of the hydrological response. The warning system was built based on the current operational rules of the reservoir. Once the flood event starts, the use of real time information about the water volume stored in the reservoir and the inflow in the next hour estimated using a Bayesian approach based on antecedent precipitation and other water flows states in the catchment, constitute the hydrological indicators to base the decision on, together with the generation of thresholds and requirements of the hydropower system. This methodological scheme could be easily transferable into other Mediterranean catchments with similar characteristics. Moreover, the development of these tools as decision support systems in the decision-making process is essential and allows the incorporation of advanced plans to adapt to global warming.

 

This work has been funded by the project FEDER UCO-1381239 Herramienta de pronóstico estocástico de caudal para gestión de centrales hidroeléctricas en cuencas mediterráneas a distintas escalas temporales, with the economic collaboration of the European Funding for Rural Development (FEDER) and the Office for Economy, Knowledge, Enterprises and University of the Andalusian Regional Government.

How to cite: Aguilar, C., Pimentel, R., Vela, S., Contreras, E., Moreno, F., and Polo, M. J.: Flash flood early-warning system in a Mediterranean reservoir at operational scales for hydropower production, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12889, https://doi.org/10.5194/egusphere-egu22-12889, 2022.

11:20–11:25
|
EGU22-710
|
ECS
|
Virtual presentation
Qingwen Lu

Floodwater conservation in reservoir flood control capacity will lead to additional flood control risk for reservoir operation during flood seasons. Coupling the meteorological and hydrological uncertainties, the probability density function of reservoir initial flood regulation water level is derived to quantify the uncertainty in floodwater conservation through an analytical method. In reservoir flood control operation, the uncertainty of initial water level being above the designed flood limited water level and the uncertainty of inflow caused by flood forecast error are two main risk factors. This study developed a dynamic and intelligent risk prediction and diagnosis model for reservoir flood regulation under two-dimensional uncertainties based on Bayesian network. Three modules are included: Bayesian network structure learning, parameter learning, and probability inference. The nodes of Bayesian network are determined and the network structure is established with expert knowledge; the parameter learning is conducted with the training samples obtained from Monte Carlo simulation. Thereafter, through the prior probability inference without posterior information and the posterior probability inference with given posterior information, the variation of flood risk is analyzed under single-factor uncertainty and two-factors uncertainties. The model is applied to Xianghongdian Reservoir in China using a flood of 100 years return period. Results indicate: the risk resulted from inflow uncertainty is greater than that of the uncertainty of initial water level; there is a certain complementarity between the uncertainties of inflow and initial water level, and the combined risk is between the results of two single-factor risk levels. Moreover, Bayesian Network is able to conduct bi-directional inferences and infer the probability distribution of any other node, which has practical value for risk assessment and control of reservoir flood control operation.

How to cite: Lu, Q.: Risk analysis for reservoir flood control operation considering two-dimensional uncertainties based on Bayesian network, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-710, https://doi.org/10.5194/egusphere-egu22-710, 2022.

11:25–11:30
|
EGU22-1072
|
ECS
|
Virtual presentation
Carlotta Valerio, Matteo Giuliani, Andrea Castelletti, Alberto Garrido, and Lucia De Stefano

Interbasin water transfers (IBWT) are often conceived as solutions to balance freshwater's uneven spatial and temporal distribution. Climate change, increasing water demand and water quality deterioration are expected to further increase the importance of water transfer schemes in future. At the same time, IBWT are often controversial and raise concerns about their social, environmental and economic impacts.

The Tagus-Segura aqueduct (TSA) in Spain is among the major IBWT projects existing in the world. It is designed to transfer a maximum of 650 hm3/year from the Entrepeñas and Buendía dams in the Tagus headwaters river basin to the Segura river basin for irrigation and urban water supply purposes. The reduction of the natural runoff registered since the 80ies, the implementation of a non-optimal operating rule and, finally, the degradation of the Tagus river ecosystems have generated strong, still unsolved tensions between donor and receiving regions.

In this study, we employ the Evolutionary Multiobjective Direct Policy Search (EMODPS) to optimize the operation of the TSA with respect to four potentially conflicting objectives: the Tagus (i) and the Segura water demands (ii); hydropower production downstream of the Entrepeñas and Buendía dams (iii) and the social-economic benefit of the population living on the shores of the reservoirs (iv). The release decision parameters and the operating rule parameters are jointly optimized, thus allowing the exploration of trade-offs between objectives and the definition of an operating rule that could benefit all the stakeholders involved. We tested the optimization under several scenarios, with the aim to assess the effect of the implementation of different environmental flows in the Tagus river on the TSA operations.

By applying a state-of-art method such as the EMODPS to the TSA case, this work contributes to the intense ongoing debate on the present and future of this controversial water transfer in Spain. We also explore the potential of the EMODPS approach to guide the design of efficient and sustainable operating rules of water transfers, with the ultimate goal of mitigating tensions between recipient and donor regions and seeking to fulfil the environmental needs in the donor basin.

How to cite: Valerio, C., Giuliani, M., Castelletti, A., Garrido, A., and De Stefano, L.: Optimizing the operating rule of a controversial interbasin water transfer: the Tagus-Segura aqueduct (Spain), EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-1072, https://doi.org/10.5194/egusphere-egu22-1072, 2022.

11:30–11:35
|
EGU22-9742
|
Presentation form not yet defined
Jan Kwakkel and Jazmin Zatarian-Salazar

Direct policy search (DPS) is increasingly being used to flexibly design policies for multi-objective reservoir control. DPS is a promising approach that can easily find policies for many heterogenous objective functions, particularly when coupling global approximators with evolutionary algorithms. Nonetheless, specifying the topology and the family of global approximators, i.e., radial basis functions, is usually done by trial and error for practical applications and is often not reported in the literature. How does the selected family of radial basis functions affect the quality of the resulting control policies? Does the chosen family influence the search behavior of the evolutionary algorithms? Can we formulate recommendations for which families are more or less suitable in general, or given the characteristics of the control problem? We test a suite of radial basis functions to address these questions for finding Pareto optimal reservoir control policies using an established reference case. This reference case is the Conowingo reservoir, a transboundary water body in the Susquehanna River Basin in North-East US. The reservoir needs to meet multiple competing water needs for hydropower production, environmental flows, recreation, cooling water for Peach Bottom atomic power plant, and urban water supply for Baltimore, MD, and Chester, PA. To optimize the Pareto optimal reservoir control policies, we use the e-NSGA2 algorithm. Our study shows the effect of using different families of radial basis functions, particularly their impact on the recommended reservoir operations, the resulting tradeoffs across the different sectors, and the search behavior of the evolutionary algorithm.

How to cite: Kwakkel, J. and Zatarian-Salazar, J.: What family of Radial Basis Functions to use in Direct Policy Search? A comparative analysis , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-9742, https://doi.org/10.5194/egusphere-egu22-9742, 2022.

11:35–11:40
|
EGU22-13011
|
ECS
|
Virtual presentation
Sadegh Sadeghi Tabas and Vidya Samadi

The increasing stress on water resource systems has prompted researchers to look for ways to improve the performance of reservoir operations. Changes in demand, various hydrological inputs, and new environmental stresses are among issues that water managers face. These concerns have sparked interest in applying different techniques to determine reservoir operation policy to improve reservoir system performance. As the resolution of analysis rises, it becomes more difficult to effectively represent a real-world system using currently available approaches for determining the best reservoir operation policy. One of the challenges is the "curse of dimensionality," which occurs when the discretization of the state and action spaces becomes finer or when more state or action variables are taken into account. Because of the dimensionality curse, the number of state-action variables is limited, rendering dynamic programming (DP) and stochastic DP (SDP) ineffective in handling complex reservoir optimization issues. Reinforcement learning (RL) is one way to overcome the aforementioned curses of stochastic optimization of water resources systems. RL is a well-known and influential technique in machine learning research that can solve a wide range of optimization and simulation challenges. In this study, a novel continuous-action deep RL algorithm called Deep Deterministic Policy Gradients (DDPG) is applied to solve the DP problem for the Folsom Reservoir system located in California, US. Without requiring any model simplifications or surrendering any of the critical characteristics of DP, the employed continuous action-space RL method effectively overcomes dimensionality concerns. The system model employs an iterative learning method that takes into account delayed rewards without requiring an explicit probabilistic model of hydrologic processes, and it can learn the best actions that maximize total expected reward by interacting with a simulated environment. This research is funded by the US Geological Survey.

How to cite: Sadeghi Tabas, S. and Samadi, V.: Sustainable Reservoir Operation and Control Using a Deep Reinforcement Learning Policy Gradient Method, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-13011, https://doi.org/10.5194/egusphere-egu22-13011, 2022.