HS4.3 | Probabilistic hydro-meteorological forecasts: ensembles, assimilation, predictive uncertainty, verification and decision making
Probabilistic hydro-meteorological forecasts: ensembles, assimilation, predictive uncertainty, verification and decision making
Convener: Ruben ImhoffECSECS | Co-conveners: Annie Yuan-Yuan ChangECSECS, Shaun Harrigan, Schalk Jan van Andel, Kolbjorn Engeland
| Tue, 16 Apr, 14:00–15:45 (CEST)
Room 2.15
Posters on site
| Attendance Mon, 15 Apr, 10:45–12:30 (CEST) | Display Mon, 15 Apr, 08:30–12:30
Hall A
Posters virtual
| Attendance Mon, 15 Apr, 14:00–15:45 (CEST) | Display Mon, 15 Apr, 08:30–18:00
vHall A
Orals |
Tue, 14:00
Mon, 10:45
Mon, 14:00
This session brings together scientists, forecasters, practitioners and stakeholders interested in exploring the use of ensemble hydro-meteorological forecast and data assimilation techniques in hydrological applications: e.g., flood control and warning, reservoir operation for hydropower and water supply, transportation, and agricultural management. It will address the understanding of sources of predictability and quantification and reduction of predictive uncertainty of hydrological extremes in deterministic and ensemble hydrological forecasting. Uncertainty estimation in operational forecasting systems is becoming a more common practice. However, a significant research challenge and central interest of this session is to understand the sources of predictability and development of approaches, methods and techniques to enhance predictability (e.g. accuracy, reliability etc.) and quantify and reduce predictive uncertainty in general. Ensemble data assimilation, NWP preprocessing, multi-model approaches or hydrological postprocessing can provide important ways of improving the quality (e.g. accuracy, reliability) and increasing the value (e.g. impact, usability) of deterministic and ensemble hydrological forecasts. The models involved with the methods for predictive uncertainty, data assimilation, post-processing and decision-making may include machine learning models, ANNs, catchment models, runoff routing models, groundwater models, coupled meteorological-hydrological models as well as combinations (multimodel) of these. Demonstrations of the sources of predictability and subsequent quantification and reduction in predictive uncertainty at different scales through improved representation of model process (physics, parameterization, numerical solution, data support and calibration) and error, forcing and initial state are of special interest to the session.

Orals: Tue, 16 Apr | Room 2.15

Chairpersons: Ruben Imhoff, Annie Yuan-Yuan Chang, Kolbjorn Engeland
Development of techniques and workflows for hydro-meteorological ensemble forecasting systems
On-site presentation
Michael Wagner and Jens Grundmann

Runoff in small catchments tend to response quickly to heavy precipitation input. The potential disastrous consequences demand a reliable precipitation forecast and flood early warning for an appropriate flood defense management. Numerical weather models provide relatively long lead times that allow early warnings of heavy precipitation. Further, meteorological ensembles consider the uncertainty of the forecast. Feeding this data to a hydrological model propagates the meteorological information and its uncertainty to catchment discharge time series.

Within the scope of the project HoWa-PRO (funded by the Federal Ministry of Education and Research, Germany) we propose a flood early warning system for the example of three catchments in the Free State of Saxony. The so-called sentinel watches meteorological ensemble forecasts of the German Weather Service (DWD). If a specific precipitation criterion is surpassed, the sentinel starts collecting and concatenating various precipitation products. For operational use, a combination of radar, nowcast, and (ensemble) forecast data is created (Radolan-RW, Radolan-RV, Icon-D2-EPS, Icon-EU). Besides this renowned precipitation products, we set up a second hydrologic ensemble forecasting system using prototypic data of upcoming products for precipitation observation and forecasting. Here we combine (1) observed radar data assimilated to precipitation gauges and commercial microwave links (pyRADMAN), and (2) the seamless prediction data SINFONY-INTENSE. The latter is a combination of nowcasting and numerical forecast ensembles. Both data products are delivered some minutes earlier than the classic data. The sentinel evaluates the concatenated precipitation data in the catchments according to further criteria for heavy precipitation events. If a criterion is met, the hydrological model is started with the formerly concatenated full ensemble precipitation data. The results are used in a prototypic web demonstrator to depict the current flood situation in the covered catchments. An easy to grasp traffic light scheme and – if needed or wanted – additional information including the uncertainty range facilitate quick decisions and actions of the flood defense management in the appropriate region.

The sentinel scales well with additional catchments which can be simulated in parallel. Currently, the sentinels for both data versions (operational and upcoming precipitation products) are invoked each 30 min, shortly after new observed data is delivered. The used WeatherDataHarmonizer library (Wagner and Grundmann, 2023) ensures a temporally, spatially, and formally homogeneous precipitation data set with a lead time of maximum 180 h, a time resolution of 15 min, and a spatial resolution of about 1 km. Each component of the sentinel is robust in a sense of handling missing operational data or machine faults.

Additionally to the technical aspects, we present results of operational hydrologic ensemble forecasts for selected events and catchments and compare the performance of both systems.

Wagner, M. and Grundmann, J.: Precipitation Data Harmonizer: Harmonizing radar, nowcast, and forecast precipitation data for hydrological applications, EGU General Assembly 2023, Vienna, Austria, 24–28 Apr 2023, EGU23-8978, https://doi.org/10.5194/egusphere-egu23-8978, 2023.

How to cite: Wagner, M. and Grundmann, J.: Operational Hydrological Ensemble Forecasts in Small Catchments – Implementing Seamless Precipitation Predictions, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-7372, https://doi.org/10.5194/egusphere-egu24-7372, 2024.

On-site presentation
Céline Cattoën, Maria-Helena Ramos, Daniela Peredo, Stuart Moore, and Trevor Carey-Smith

High-resolution ensemble weather forecasts are essential for forecasting floods in complex topography with fast-responding catchments. However, they are computationally expensive. This study explores a cost-effective alternative via the use of time-lagged ensembles, defined as ensemble forecasts from the same model initialised at different times but verifiying at the same time.

We evaluate lagged ensemble products, with varying configurations and proportions of lagged members, constructed from convective-scale numerical weather predictions (NWP) ensembles to forecast extreme flood events for fast-responding catchments. We compare four lagged products for two extreme event case studies in France and New Zealand. Lagged NWP ensemble products are used to drive hydrological models and evaluate flood forecasts across a range of performance evaluations based on traditional event-based metrics and user-focused strategies. We construct forecast diagrams and associated metrics based on the Brier Score for varying flood threshold severities to evaluate anticipatory and overly alarmist predictions.  

Comparisons with a control burst ensemble (without any lagging) reveal that flood forecasts derived from lagged convective-scale NWP ensembles have the potential to better capture extreme flood events, albeit with some limitations. Benefits vary with time and lagged product configuration, but generally, lagged ensemble products match or surpass their control counterparts, particularly in spread-skill, anticipatory prediction of a severe flood threshold, and consistency of forecasts -- critical to retaining trust from the emergency management sector. However, lagged products tend to increase overly alarmist predictions at early forecast ranges but become a more effective strategy at longer ranges over a larger region.

Utilising forecast diagram metrics based on the Brier Score allows the evaluation of multiple basins and ensemble products, incorporating an end-user-focused perspective for decision-making, such as anticipation of flood exceedance thresholds. The results provide valuable insights into lagged convective-scale weather ensembles' potential benefits and limitations in enhancing flood forecasting accuracy and reliability.

How to cite: Cattoën, C., Ramos, M.-H., Peredo, D., Moore, S., and Carey-Smith, T.: Assessing lagged convective-scale weather ensembles for improved flood forecasting in fast-responding catchments , EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-1543, https://doi.org/10.5194/egusphere-egu24-1543, 2024.

Virtual presentation
Belinda Trotta

We introduce Neighbourhood Ensemble Copula Coupling, a technique for post-processing ensemble precipitation forecasts to produce physically realistic, well-calibrated scenarios.

Modern precipitation forecasting typically uses ensemble forecasts produced by numerical weather prediction (NWP) models. These ensembles aim to represent the range of probable weather outcomes, and enable us to derive probabilistic predictions (for example, we may predict a 20% chance of at least 10 mm rainfall at a particular location). Because NWP models employ a simplified representation of the atmospheric dynamics, and model processes at a coarse scale, the probabilities derived from the ensemble must be calibrated to accurately describe the probability distribution at the specific forecast locations.

Moreover, for many applications, such as flood prediction, as well as a probabilistic prediction of rainfall at each location, it is also useful to know the correlations between different locations. A river is most likely to flood when there is high rainfall at several nearby locations, so the probability of a flood depends on the joint probability distribution of rainfall at these locations. This is not easy to calculate, since the individual distributions are not independent. For example, if there is a rain band and we are unsure how fast it will move, we may know that at a particular forecast time it will rain either in location A or location B, but not both simultaneously. Thus for hydrology applications, scenario-based forecasts are often more useful than probability forecasts, but we still require the ensemble of scenarios to be well-calibrated; that is, the distribution of scenarios at a location should approximate the true expected probability distribution.

One popular approach to this problem is Ensemble Copula Coupling (ECC). Given an NWP ensemble forecast, and a probabilistic forecast derived from it, ECC is a method to derive a calibrated ensemble forecast by arranging quantiles of the probabilistic forecast in the order specified by the original ensemble. This process improves the statistical accuracy of the ensemble; in other words, the distribution of the calibrated ensemble members at each grid point more closely approximates the true expected distribution. However, the trade-off is that the individual members are not as physically realistic as the original ensemble, with noisy variation among neighbouring grid points. Also, depending on the calibration method, extremes in the original ensemble are sometimes muted: in particular, reliability calibration, a simple and widely used non-parametric probability calibration method, suffers from this problem. Neighbourhood Ensemble Copula Coupling (N-ECC) is a simple modification of ECC designed to address these drawbacks. Testing N-ECC with the calibrated probability forecasts produced by reliability calibration shows that, compared to standard ECC, our method produces forecasts which are less noisy and more visually plausible, and which also have improved statistical properties. Specifically, the forecast is sharper, so that extremes are better predicted, and the continuous rank probability score (CRPS) is also slightly improved.

How to cite: Trotta, B.: Producing calibrated ensemble precipitation forecasts using Neighbourhood Ensemble Copula Coupling, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-14291, https://doi.org/10.5194/egusphere-egu24-14291, 2024.

On-site presentation
Lingfeng Li, Huan Wu, Lujiang Lu, and Weitian Chen

Sub-seasonal to seasonal (S2S) weather forecasting is widely regarded as a big challenge because of less predictability than short-term and seasonal forecasts, which significantly impedes the streamflow forecasts at the same time scale. In this study, we propose an integrated numerical and statistical approach to improve the accuracy of S2S streamflow forecasting based on a physically based hydrological model, i.e., the Dominant river tracing-Routing Integrated with VIC Environment (DRIVE) model with a Bayesian joint probability (BJP) model. The DRIVE model provides S2S streamflow simulations by leveraging physical processes, while the BJP model could mitigate the issue of over-fitted flood peaks and partially correct under-fitted flows. The main strategy to reduce the streamflow prediction uncertainty is through optimizing the integration of hydrological model simulations and statistical predictions. We applied the integrated DRIVE-BJP model to the Pear River Basin during the 2020-2022 time period and performed the validation according to observations at 24 hydrological stations located within the river basin. The results show the proposed ensemble approach yields significant improvements compared to the single BJP or DRIVE model. This study of fusion of the DRIVE and BJP models to enhance the sub-seasonal flood prediction, showing promising practical values in flood early warning and water resource management.

Key words: S2S, ensemble streamflow forecast, DRIVE model, BJP model, Pearl River basin

How to cite: Li, L., Wu, H., Lu, L., and Chen, W.: Ensemble streamflow probability prediction at the sub-seasonal to seasonal (S2S) timescale, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-9909, https://doi.org/10.5194/egusphere-egu24-9909, 2024.

On-site presentation
James Bennett, Ilias Pechlivanidis, Yiheng Du, Marie-Amélie Boucher, and Fredrik Wetterhall

Creating forecast systems that add value across spatial scales and time horizons is crucial for a variety of fields, from meteorology and climate science to business and public policy. The priorities for developing such systems may vary depending on the specific domain and objectives. The international community of practice of the Hydrological Ensemble Prediction Experiment (HEPEX) has been seeking to advance the science and practice of hydrological ensemble prediction and its use in impact- and risk-based decision-making. Over the years, HEPEX has been promoting knowledge utilising cutting-edge techniques and data to innovate hydrological forecasting methods, products and systems, and improve services for users in the water-related sectors. During the 2023 HEPEX workshop in Norrköping, Sweden, (https://hepex.org.au/hepex-workshop-2023-forecasting-across-spatial-scales-and-time-horizons/), the community proposed and elaborated on the key priorities for (co-)creating hydrological forecast systems that are broadly applicable and can add value for local/regional decision-making. The notes from five breakout groups (about 10 participants in each group) were collected and analysed, while the proposed efforts and ways forward were classified and prioritised. Here, we present the outcomes from these breakout discussions, while we support the identified priorities providing backgrounds on the science needed, the HEPEX contribution towards these priorities, and the path forward for contributing to the United Nations Early Warnings for All (EW4All) initiative.

How to cite: Bennett, J., Pechlivanidis, I., Du, Y., Boucher, M.-A., and Wetterhall, F.: What are the top priorities for (co-) creating hydrological forecast systems that add value across spatial scales and time horizons? Outcomes from the 2023 HEPEX workshop, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-20930, https://doi.org/10.5194/egusphere-egu24-20930, 2024.

Ensemble verification methods and hydro-meteorological forecasting system evaluation
On-site presentation
Evaluation and uncertainty assessment of a real-time regional ensemble flow forecasting system in Catalonia
Xinyu Li, Marc Berenguer, Carles Corral, Shinju Park, and Daniel Sempere-Torres
On-site presentation
Maliko Tanguy, Shaun Harrigan, Corentin Carton De Wiart, Michel Wortmann, Thomas Haiden, and Christel Prudhomme

Operational hydrological forecasting systems play a vital role for effective decision-making in disaster management and resource planning. This study addresses the critical need to conduct a fair and realistic assessment of the skill in one such system: the European Flood Awareness System (EFAS). As an integral component of the Copernicus Emergency Management Service (CEMS), EFAS undergoes monthly forecast verification, spanning lead times from 6 hours to 10 days. This verification process provides users with a valuable insight into the system’s performance in predicting streamflow for the preceding month.  

Motivated by the importance of providing stakeholders with trustworthy information, our research focuses on a thorough examination of benchmark forecasts to evaluate EFAS performance. The choice of benchmark forecasts significantly influences the perceived accuracy of the system, and using benchmarks that are too easy to beat can lead to artificially inflated skill. Therefore, the primary objective of this work is to pinpoint the most suitable benchmark, serving as a robust reference for assessing the true capabilities of EFAS. This will then feed into the development of a ‘headline score’ which is a unique value of a key metric representative of a geographical domain that enables to track performance evolution.

The study employs various benchmark forecasts, including persistence forecast, climatology, and the previous day’s forecast, using the Continuous Ranked Probability Skill Score (CRPSS) for skill assessment. Expanding on previous findings that identified persistence forecast as the most suitable for short lead times and climatology for longer lead times, this work refines and extends these results. We specifically examine the influence of catchment characteristics on the selection of the optimal benchmark at different lead times for operational forecasting evaluation. By uncovering the most robust benchmark, our study contributes to a more accurate understanding of EFAS capabilities, ultimately enhancing the overall performance assessment of EFAS. The nuanced insights gained from this focused examination serve as a step toward refining the methodology and criteria employed to develop new ‘headline scores’, instrumental in evaluating the evolution of the system’s forecasting skill.

How to cite: Tanguy, M., Harrigan, S., Carton De Wiart, C., Wortmann, M., Haiden, T., and Prudhomme, C.: Forecast Verification in Operational Hydrological Forecasting: A Detailed Benchmark Analysis for EFAS, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-9172, https://doi.org/10.5194/egusphere-egu24-9172, 2024.

On-site presentation
Ubolya Wanthanaporn, Ronald Hutjes, and Iwan Supit

The potential use of European Centre for Medium-Range Weather Forecast (ECMWF) ensemble prediction system SEAS5 over Mainland Southeast Asia was evaluated. The evaluation spans 30 years (1985–2014), examining SEAS5's skill in predicting temperature and precipitation. Subsequently, SEAS5 data was used to force the Variable Infiltration Capacity (VIC) hydrological model for runoff and streamflow forecasts, as well as the WOrld FOod Studies (WOFOST) crop model for rice production forecasts. These hydrological and agricultural results were compared against the WFDE5-driven reanalysis using verification skill metrics at grid cells for each month. Furthermore, the hydrological results were compared against observed station data. The reanalysis of rice yield was also compared against FAO observations, but proved inconclusive. The findings reveal promising predictive capabilities for temperature beyond a 2-month forecast, while the skill of precipitation and streamflow forecasts extend to a 1-month. Noteworthy, strong seasonal and regional dependence occurs, with high forecast skills during the pre-monsoon (April–May) and post-monsoon (October–November). Year–to–year precipitation tercile plots highlight skill in predicting the anomalous seasonal conditions associated with ENSO. The significant streamflow skill at each initiation month and lead time corresponds to the forecasting skill of meteorological variables. Nevertheless, it is important to note that the skill level of discharge and runoff forecasts is generally lower compared to the skill in temperature and precipitation. For the rice prediction, SEAS5 exhibits high performance at the beginning of the rainy season, where strong seasonal climate predictions are observed. The model shows the ability to capture anomalous rice yields and consistent accuracy throughout a 1-month to 3-month forecast. However, limitations in skill are evident when rice planting times are delayed by one or two months during the rainy season, as well as when planting in the dry season. SEAS5 shows useful skills that can potentially be used for hydrological and agricultural anticipatory management. The results could already support an initial step to come to potential anticipatory (agro-)hydrological management and could be utilised as an input for an early warning system in various sectors.

How to cite: Wanthanaporn, U., Hutjes, R., and Supit, I.: Skill of the ECMWF SEAS5 ensemble prediction system in streamflow and rice yield forecasting for Mainland Southeast Asia , EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-1371, https://doi.org/10.5194/egusphere-egu24-1371, 2024.

On-site presentation
Celia Ramos Sánchez, Micha Werner, Lucia De Stefano, and Andrew Schepen

The potential of user-centric climate services to facilitate proactive drought management approaches is leading to increased efforts to develop and test climate services that include hydro-meteorological forecast products. In Spain, there is an interest to incorporate seasonal forecasts into drought management, including by integrating the information these provide into the system of drought indicators that are used in current operations. However, despite seasonal forecast information being available to users, their actual use to support operational decisions is limited. One aspect that fosters uptake is how credible users consider seasonal forecasts, including the quality of the forecasts. Additionally, the salience of the information provided is important, with information being credible at the spatial and temporal scales commensurate to those of the indicators used to support decisions. Bias-correction of seasonal forecasts has a crucial role in enhancing forecast quality, though there are several aspects intrinsic to bias-correction processes that may influence the degree of quality enhancement from the perspective of the users’ interests. An aspect that has so far received little attention is the spatial resolution of the forecast and the reference precipitation datasets that are applied.

Here, we examine the influence of spatial resolution of both the forecast and the reference precipitation datasets used in the bias-correction process on the skill of seasonal forecasting, from the perspective of the indicators used in operational drought management decisions. We apply bias-corrected daily rainfall forecasts (ECMWF System 5) for a region within the Spanish Douro River Basin. We use a Bayesian Joint Probability (BJP) modelling approach for bias correction and the Schaake Shuffle method to conserve the temporal and spatial correlations across lead times and sub-catchments, respectively (Schepen et al., 2018). We consider two resolutions of the forecast product (1° and 0.4°), and apply the bias correction at three nested spatial scales, namely independent sub-catchments, aggregated sub-catchments and the entire catchment. These spatial aggregations have been selected to match those of the indicators used in the drought management plan implemented by the Douro Basin Authority. Forecast quality is evaluated through a set of skill scores and metrics at the daily scale, as well as at the aggregated temporal scales of the indicators used. Our results show that the bias correction applied to the seasonal forecasts does improve the skill of the forecast information at the spatial and temporal scales that are relevant to the indicators used operationally. We also show the influence on the improvement of skill of choices made in selecting the spatial resolution of the forecasts themselves and at which bias correction is applied, and discuss how this can help inform the design of climate services to support operational drought management decisions.

How to cite: Ramos Sánchez, C., Werner, M., De Stefano, L., and Schepen, A.: Influence of spatial resolution on forecast quality of bias-corrected seasonal forecasts from a drought management perspective, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-17468, https://doi.org/10.5194/egusphere-egu24-17468, 2024.


Posters on site: Mon, 15 Apr, 10:45–12:30 | Hall A

Display time: Mon, 15 Apr 08:30–Mon, 15 Apr 12:30
Chairpersons: Schalk Jan van Andel, Shaun Harrigan, Ruben Imhoff
Ensemble verification methods and hydro-meteorological forecasting system evaluation
Louise Crochemore, Stefano Materia, and Elisa Delpiazzo and the H2020 CLARA project team

Assessing the information provided by co-produced climate services is a timely challenge given the continuously evolving scientific knowledge and its increasing translation to address societal needs. Here we propose a joint evaluation and verification framework to assess prototype services that provide seasonal forecast information based on the experience from the H2020 CLARA project. The quality and value of the forecasts generated by CLARA services were assessed for five climate services utilizing the Copernicus Climate Change Service seasonal forecasts and responding to knowledge needs from the water resources management, agriculture, and energy production sectors. This joint forecast verification and service evaluation highlights various skills and values across physical variables, services and sectors, as well as a need to bridge the gap between verification and user-oriented evaluation. We provide lessons learnt based on the service developers’ and users’ experience, and recommendations to deploy such verification and evaluation exercises. Lastly, we formalize a framework for joint verification and evaluation in service development, following a transdisciplinary (from data purveyors to service users) and interdisciplinary chain (climate, hydrology, economics, decision analysis). The diagnostics provided by the joint assessment improve on individual quality and skill assessment by bringing forth more dimensions and an in-depth analysis of the sources of service value. Service co-production should thus consider simultaneously and iteratively the quality of the hydro-climate information provided and its value for decision-making to inform robust enhancement strategies. Such effort, however, requires for prolonged collaborations between social and climate scientists, service developers and service users, beyond the lifespan of common research projects and for building communities that allow such long-term collaborations.

How to cite: Crochemore, L., Materia, S., and Delpiazzo, E. and the H2020 CLARA project team: Joint verification and evaluation of seasonal forecasts from climate services: Experience from the H2020 CLARA project, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-11481, https://doi.org/10.5194/egusphere-egu24-11481, 2024.

Parthkumar Modi, Jared Carbone, Hannah Kamen, Eric Small, Bill Szafranski, Cameron Wobus, and Ben Livneh

More than half of annual runoff across the montane regions of the western US and Europe originates as snowmelt, making knowledge of snowpack crucial to the quality of water supply forecasts. However, ongoing and projected warming is expected to reduce snow water equivalent (SWE) and alter snowmelt timing, thus impacting forecast skill and uncertainty. Rising temperatures are anticipated to reduce the fraction of future precipitation falling as snow by up to 30% in intermountain and continental regions of the western US. This will fundamentally alter the regional water cycle, and we posit that this will increase forecast errors and uncertainty, ultimately impacting the quality of decision-making that relies on water supply information. This research assesses the Relative Economic Value (REV) of water supply forecasts under changing snowpack regimes to understand the impact of forecast uncertainty on economic outcomes. Forecast errors will be rigorously estimated using statistical, physical, and machine learning models applied to 76 western US basins. Historical and projected future hydrology (2025-2050) will serve as a test bed for the analysis. Preliminary results over these basins suggest forecast errors on the order of +/- 25%, corresponding with changes in economic outcomes of up to +/- 15%. With findings from the proposed research, we hope to aid water entities by assessing the economic outcomes based on the skill of water supply forecasts, thereby exploring how forecast users can maximize productivity in response to changing climate conditions.

How to cite: Modi, P., Carbone, J., Kamen, H., Small, E., Szafranski, B., Wobus, C., and Livneh, B.: The impact of streamflow forecast errors on economic outcomes in future climates, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-13749, https://doi.org/10.5194/egusphere-egu24-13749, 2024.

Schalk Jan van Andel, Claudia Bertini, Celia Ramos Sánchez, Andrea Ficchì, Matteo Giuliani, Michiel Pezij, Dorien Lugt, Lucia De Stefano, Ilias Pechlivanidis, Micha Werner, and Andrea Castelletti

Sub-seasonal to seasonal (S2S) hydrometeorological ensemble predictions are being put to the test, given their potential to increase preparedness for, and improve management of, extreme events. Whereas forecast skill at the global, continental, and regional scales of S2S hydrometeorological predictions is relatively well-documented and regularly updated in the scientific literature for generic definitions of events, assessments of forecast skill at the catchment and local scale for local use-case definition of events have been reported to a relatively limited extent.

In this research, the forecast skill of S2S predictions is analysed for local preparedness and management of droughts. In case studies in the Netherlands, Spain and Italy; potential end users have been consulted on their definition of drought, warning thresholds, decision-making process, and potential mitigation actions. The result is a variety of meteorological, hydrological and agricultural use-case-specific drought event definitions.

For each of these case studies, a state-of-the-art multi-year (re-)forecast hydrometeorological dataset is downloaded or developed (e.g. for sub-seasonal and seasonal meteorological forecasts from ECMWF Extended Range and SEAS5 prediction systems, respectively, and for hydrological forecasts from the E-HYPE and GLOFAS hydrological systems), and its performance analysed for the user-driven drought definitions.

The results contribute to an improved understanding of the potential of state-of-the-art S2S predictions for local-scale drought preparedness and management, and identify aspects to focus on in further enhancing this potential.

How to cite: van Andel, S. J., Bertini, C., Ramos Sánchez, C., Ficchì, A., Giuliani, M., Pezij, M., Lugt, D., De Stefano, L., Pechlivanidis, I., Werner, M., and Castelletti, A.: Use-case specific performance assessment of sub-seasonal to seasonal drought predictions for local-scale applications in the Netherlands, Spain, and Italy, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-21041, https://doi.org/10.5194/egusphere-egu24-21041, 2024.

Mohamed Elghorab, Jens Grundmann, Michael Wagner, and Schalk Jan van Andel

The application of ensemble forecasting for predicting extreme weather events and floods, necessitates a thorough assessment of its performance and reliability. Consequently, evaluation emerges as a crucial step, offering valuable insights into the overall predictive skill of ensemble forecasts. Notably, limited attention has been given to evaluating the utility of ensemble forecasts as early warning tools in small catchments that are characterized by rapid hydrological processes and flash floods. This study addresses this gap by focusing on the performance analysis of the ensemble weather forecast provided by the Deutscher Wetterdienst (DWD) in ten small catchments distributed across three regions in Saxony, Germany. The findings aim to contribute to a deeper understanding of the effectiveness of ensemble forecasting in the context of early warning systems for small catchments.

To realize this research, a specialized computational tool particularly tailored for the structure and format of the forecast products was developed. It encompasses an arsenal of evaluation metrics, including contingency table-based metrics that examine forecasts from the perspective of extreme events (e.g., false and true alarm rates, accuracy, area under the ROC curve). Additionally, the tool incorporates metrics treating the entire forecast ensemble as a probability distribution, measuring its degree of conformity with the ground truth (e.g., CRPS & Brier score).

The evaluation exercise involved a numerical comparison between modelled forecasts and actual measured observations. For meteorological forecasts, the evaluation was conducted using the numerical weather prediction models ICON-D2-EPS/COSMO-D2-EPS against RADOLAN-RW radar observations. In the case of hydrological forecasts, the modelled runoff ensemble forecasts were compared against gauged catchment runoff measurements.

Upon a comprehensive evaluation of ensemble predictions from various perspectives, including accuracy and reliability, it is observed that the utilization of ensemble forecasting in small catchments demonstrates a satisfactory level of performance. Notably, the central region of Saxony exhibited slightly superior performance compared to the other two test regions. Furthermore, the results indicate a tendency for the ensemble average to consistently overestimate observations in both rainfall and runoff forecasts. It is also equally important to note that this tendency is primarily attributed to the mathematical approach employed in the analysis.

In conclusion, the collective findings of this research offer valuable insights into the practical application of ensemble forecasts for decision-making in each of the small catchments. Despite the overestimation tendency of the forecasts, the overall performance and reliability of ensemble forecasting in small catchments, especially in the central region of Saxony, suggest its potential as a valuable tool for informed decision-making in hydrological and meteorological contexts.

How to cite: Elghorab, M., Grundmann, J., Wagner, M., and van Andel, S. J.: Evaluation of Hydro-meteorological Ensemble Forecasts in Small Catchments in Saxony, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-16763, https://doi.org/10.5194/egusphere-egu24-16763, 2024.

Development of techniques and workflows for hydro-meteorological ensemble forecasting systems
Louise Arnal, Martyn P. Clark, Alain Pietroniro, Vincent Vionnet, David R. Casson, Paul H. Whitfield, Vincent Fortin, Andrew W. Wood, Wouter J. M. Knoben, Brandi W. Newton, and Colleen Walford

Seasonal streamflow forecasts provide key information for decision-making in sectors such as water supply management, hydropower generation, and irrigation scheduling. Principal component regression (PCR) stands as a well-established and widely used data-driven method for seasonal streamflow forecasting, offering advantages over more complex methods, including intuitive use of local data to represent key hydrological processes and low computational resource requirements.

We will present FROSTBYTE, a systematic and reproducible data-driven workflow for probabilistic seasonal streamflow forecasting in snow-fed river basins. FROSTBYTE is available on GitHub as a collection of Jupyter Notebooks, facilitating broader applications in cold regions and contributing to the ongoing advancement of methodologies. This structured workflow consists of five essential steps: 1) Regime classification and basins selection, 2) Streamflow pre-processing, 3) Snow Water Equivalent (SWE) pre-processing, 4) Forecasting using PCR, and 5) Hindcast verification. It was applied to 75 basins characterized by a snowmelt-driven regime and limited regulation across diverse North American geographies and climates. Ensemble hindcasts of winter to summer streamflow volumes were generated from 1979 to 2021, with initialization dates ranging from January 1st to September 1st. The hindcasts were evaluated with a user-oriented approach, tailored to offer insights for snow monitoring experts, forecasters, decision-makers, and workflow developers. Join us to learn more about FROSBYTE, and explore ways in which you can actively contribute to its development.

How to cite: Arnal, L., Clark, M. P., Pietroniro, A., Vionnet, V., Casson, D. R., Whitfield, P. H., Fortin, V., Wood, A. W., Knoben, W. J. M., Newton, B. W., and Walford, C.: FROSTBYTE: A reproducible data-driven workflow for probabilistic seasonal streamflow forecasting in snow-fed river basins across North America, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-13915, https://doi.org/10.5194/egusphere-egu24-13915, 2024.

Sakila Saminathan and Subhasis Mitra

The main objective of this study is to evaluate the efficacy of machine learning (ML) techniques in improving numerical weather prediction (NWP) based reference evapotranspiration (ETo) forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) at short to medium range time scale across different zones in the Indian region. The meteorological hindcasts from ECMWF are used to estimate ETo forecasts using the FAO Penman-Monteith equation. Thereafter, the raw forecasts are post-processed using two ML techniques: Support Vector Regression (SVR) and Extreme Gradient Boosting (XGBoost). The ML techniques are applied to rawETo forecast in order to improve its reliability and accuracy. The raw and ML post-processed ETo forecasts are assessed using deterministic evaluation metrics. Results highlight that ML post-processed ETo forecasts have superior skill than raw ETo forecasts. The highest improvement is reported in the Himalayan regions, and the XGBoost model outperformed the SVR model across all zones. The outcomes of this study has implications towards agricultural water management and irrigation scheduling over the Indian subcontinent.

How to cite: Saminathan, S. and Mitra, S.: Post-processing of short to medium range NWP based reference evapotranspiration forecasts using Machine Learning Techniques across the Indian subcontinent, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-12142, https://doi.org/10.5194/egusphere-egu24-12142, 2024.

Daniela Maldonado, Ximena Vargas, and Pablo Mendoza

Understanding the availability and distribution of water resources is crucial for efficient management. Chilean basins in the northern and central regions typically exhibit characteristics of a snow or mixed regime, with snowmelt runoff being the primary source of supply. Industries such as mining, agriculture and hydroelectric power generation experience peak demand during the snowmelt period. Determining the average and monthly distribution of runoff during this period is essential for effective planning.

Currently, both public and private institutions perform forecasts to provide valuable information to diverse users, including farmers and hydroelectric companies. This research aims to enhance the snowmelt forecast in snow-dominated or mixed basins integrating hydrometeorological forecasts and system states into a fuzzy model. The study focuses on the Choapa en Cuncumén, Maipo en el Manzano and Tinguiririca Bajo los Briones rivers.

The methodology establishes relationships between system inputs, such as precipitation, snow cover and temperature, and the output variable of snowmelt runoff. Using membership functions and fuzzy rules, the model is developed based on observations collected over fifteen-year period. The performance of each model is evaluated with the commonly employed metrics, through sensitivity analysis and cross- validation. A subsequent comparison with existing models allows us to draw conclusions about the effectiveness of fuzzy models in predicting snowmelt runoff.

The simulation yield Kling-Gupta Efficiency (KGE) values exceeding 0.5 in the first two basins and close to 0.4 in the last one, indicating an acceptable forecast relative to those produced by other institutions. This underscores varying model performance across the three basins, contingent on specific conditions and dependencies on model inputs. While a successful calibration is achieved, a detailed examination of water rights and streamflow naturalization, particularly in basins where this value holds significance, is essential.

How to cite: Maldonado, D., Vargas, X., and Mendoza, P.: Application of fuzzy logic for water supply forecasting in three Andean catchments of central Chile, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-4437, https://doi.org/10.5194/egusphere-egu24-4437, 2024.

Jeremy Rohmer, Eva Membrado, Sophie Lecacheux, Deborah Idier, Andrea Filippini, Rodrigo Pedreros, Alice Dalphinet, Denis Paradis, and David Ayache

Recent advances in high performance computing have enabled numerical weather prediction systems to move from deterministic to probabilistic forecasting using Ensemble Prediction Systems (EPS). While EPS are increasingly used to predict river flows and induced floods in several countries, it is only emerging for marine flooding. Despite ongoing efforts to develop new generations of high performance hydrodynamic models accounting for complex processes, the main challenge still remains the computer power required to run multiple simulations with a chain of models of increasing resolutions (from a hundred meters for water level at the coast, to a few meters for coastal waves and marine flooding). To overcome this limitation, the machine-learning-based metamodelling approach has made great progresses in this field of application.

Through a statistical analysis of pre-calculated training databases, metamodels can predict key flooding indicators (surge, discharge, water depth, etc.) at a given spatial locations of interest within reasonable time and computing resources while preserving the accuracy of full process models. Yet, some issues remain to push this approach toward operational applications: (1) the production of spatialized indicators with metamodels such as inland water depth maps, (2) the characterization of the cascading sources of uncertainties throughout the entire chain.

To address these difficulties, we use a set of numerical simulation results of about 200 flood maps computed on the Arcachon Lagoon (Gironde, France) for a large variety of randomly-generated metoceanic forcing conditions (surge, tide, wave and wind). On this basis, a metamodeling procedure is developed by combining a non-linear dimension reduction method relying on deep-learning-based autoencoders, denoted AE (to represent the very high-dimensional spatialized output) and on Gaussian process (Gp) regression (to model the link between the metoceanic forcing conditions and the flood response). Cross validation and comparison to historical real cases (such as storms Xynthia and Klaus) show satisfying predictive capability. However, the concern is the model uncertainty that affects the different steps of the whole metamodeling procedure. To quantify it, we rely on a stochastic approach that combines conditional Gp simulations with AE random responses using Monte Carlo Dropout method. In order to discuss predictive uncertainty to support decision-making for real-time forecasts, we compare the impact of metamodelling uncertainty with that induced by the variability of metoceanic forcing conditions which are modelled on the basis of the Meteo-France ensemble named PEARP "Prévision d'Ensemble ARPege" for recent storm events, as well as for synthetic marine inundation events.

How to cite: Rohmer, J., Membrado, E., Lecacheux, S., Idier, D., Filippini, A., Pedreros, R., Dalphinet, A., Paradis, D., and Ayache, D.: Ensemble forecasts of marine flood maps assisted by probabilistic machine learning techniques: Application at Arcachon Lagoon (France), EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-14945, https://doi.org/10.5194/egusphere-egu24-14945, 2024.

Ojiro Furuoka and Tomohito Yamada

Forecasting of natural phenomena is generally based on observation data, but it is impossible to measure everything perfectly, and there are uncertainties in the limits of observation and human perception. In runoff forecasting, there is also uncertainty in the rainfall information that uses as input data, and it is important for flood control to estimate the effect of this uncertainty on the output data, the runoff data.

Currently, rainfall in Japan is monitored by ground rain gauges and meteorological radar. Ground rain gauges measure rainfall on the ground, but they are installed at intervals of 10 to 20 km, and not all locations are measured seamlessly. Ground-based rain gauges have spatial uncertainties because they may miss cumulonimbus clouds with a horizontal scale of about 10 km that bring heavy rainfall, and because rainfall in mountainous areas is considered to be highly spatially variable due to the topography. On the other hand, radar observations use radio waves to measure the shape of raindrops in the sky and estimate rainfall on a 250 m mesh or 1 km mesh. There are limitations in accuracy, such as indirect rainfall estimation and rainfall attenuation behind strong rainfall areas. There are discrepancies in comparison with ground rain gauges at the same location, and radar must also accept uncertainties due to observation limitations. 

Uncertainty in rainfall data is not limited to observational uncertainty.Uncertainty also exists in the process of rainfall retention by soil until it flows into the river. In particular, while the degree of soil wetness prior to a rainfall event is considered to have a significant impact on the process of water retention, it is impossible to observe the wetness of the entire watershed, and there is inherent uncertainty as to how much rainfall will contribute to direct runoff. Against this background, it is necessary to convert the conventional deterministic model, which uniquely provides rainfall data and uniquely determines the runoff height, to a stochastic model that accounts for uncertainty.

For studies that consider uncertainty, a rainfall-runoff model that takes into account uncertainties due to observation limitations was proposed by Yamada et al. and a rainfall-runoff model that takes into account uncertainties due to observation limitations and initial soil moisture uncertainty was proposed by Intan and Yamada.

In this study, the theoretical development is described and a rainfall-runoff model with two uncertainties is applied to Kinugawa river basin to quantify the inflow.

How to cite: Furuoka, O. and Yamada, T.: A Study of Rainfall-Runoff Process considering two uncertainties in Basin with multiple dams, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-14893, https://doi.org/10.5194/egusphere-egu24-14893, 2024.

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

Display time: Mon, 15 Apr 08:30–Mon, 15 Apr 18:00
Chairpersons: Shaun Harrigan, Kolbjorn Engeland, Annie Yuan-Yuan Chang
Ensemble verification methods and hydro-meteorological forecasting system evaluation
Pierre Baguis, Joris Van den Bergh, Emmanuel Roulin, and Françoise Gellens-Meulenberghs

The flooding event of July 2021 is a natural disaster that resulted in fatalities and extensive damage to infrastructure. The flood was triggered by the extreme precipitation event of 13-16 July 2021. We investigate it for a selection of catchments of the Meuse River in Belgium using hydrological modelling and forecasting. 

We make use of a high resolution radar-based quantitative precipitation estimation of the event (RADFLOOD21), generated at the Royal Meteorological Institute of Belgium. This dataset is used to perform simulations with the hydrological model SCHEME in order to analyze the hydrometeorological conditions responsible for the flooding. Hydrological reforecasts are also performed, using the RADFLOOD21 data for initialization and precipitation hindcasts as input to a hydrological prediction system. Precipitation hindcasts are provided by numerical weather prediction (NWP) models, including the ECMWF ENS ensemble predictions. The main goal of the study is to evaluate the model and forecasting system performance in terms of river discharge for this exceptional precipitation event, and to investigate the impact of the new input data.

How to cite: Baguis, P., Van den Bergh, J., Roulin, E., and Gellens-Meulenberghs, F.: Hydrological modelling of the extreme flooding event of July 2021, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-10587, https://doi.org/10.5194/egusphere-egu24-10587, 2024.