S8

The skill of hydro-meteorological forecasts usually drops to zero at horizons beyond 10 to 14 days and predictions of daily values with a lead-time of more than two weeks deceive an unrealistic reliability and accuracy. However, the advances of ensemble forecast systems, the aggregation of daily information to weeks and the classification of variables (e.g. classifying the runoff in terciles, i.e.  below, above or within “normal” conditions in view of climatology) extend the skill of forecasts up to three weeks. With the help of post-processing methods, the skill can even be further extended up to four weeks.

Machine Learning (ML) based post-processing methods are becoming more and more popular. Several ML methods have been compared for classifying hydrological forecasts into terciles ranging from Multinomial to complex Deep Neural Networks models (DNN).  For the verification of the predicted classes and their associated probabilities the accuracy measure and the cross-entropy (log-loss) have been calculated.  The models have been applied to monthly forecasts for Switzerland divided into 300 catchments and run operationally with a 500mx500m resolution (www.drought.ch).

All the analyzed ML methods showed good results and are able to improve the accuracy and the cross-entropy equally well. The focus of further investigations has been laid on the Gaussian Process (GP) model. This method has the advantage that class probabilities can be calculated directly.  Furthermore, several studies highlighted recently the linkage between the GP and Bayesian DNN. The results of this study confirmed these similarities favoring the less complex GP model.  In order to take the uncertainty of the measurements and simulations into account, a multi-label classification (MLC) method has been introduced. Contrary to classical classification methods, the classes in MLC are not mutually exclusive. Thus, it avoids the sharp assignment to discrete classes, but allows that for example a runoff value measured (possibly with some error) at the border of one class could belong to the neighboring classes at the same time. The MLC method applied with a GP model showed the best results regarding accuracy and cross-entropy for all 300 catchments and significantly improves the skill over the whole forecast horizon of four weeks.

How to cite: Bogner, K., Chang, A. Y.-Y., and Zappa, M.: Challenges of machine learning based post-processing methods for sub-seasonal forecasts, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-104, https://doi.org/10.5194/iahs2022-104, 2022.

The scientific advancement in hydrological modeling along with the progress in numerical weather forecasting has allowed the generation of useful hydro-climate services providing skillful simulations and reliable forecasts. To meet the needs from the climate-dependent socio-economic sectors, such as energy production and public water supply, there is a need to quantify the predictability of hydrological extremes. In this study, seasonal hydrological reforecasts were generated across about 35,400 basins in Europe using the E-HYPE hydrological model for the period 1993-2015. The service setup uses seasonal predictions of daily mean precipitation and temperature from the fifth-generation seasonal forecasting system of the European Centre for Medium-Range Weather Forecasts (ECMWF SEAS5). Hydrological forecast predictability was analyzed with respect to the simulated streamflow climatology. The assessment was conducted on both high and low streamflow and therefore the Brier Skill Score (BSS) was used for 10th, 90th and 95th weekly percentiles for different initializations and lead times. Results show that overall hydrological extremes over Europe are well predicted in terms of BSS, with spatial and temporal variability depending on the initialization month and lead time. The results contribute to identifying different geographical areas and times where the seasonal hydrological forecasts provide an added-value on the long-term predictability of, and hence preparedness to, flooding and droughts, which consequently benefit regional and even national decision-making in various socio-economic sectors.

How to cite: Du, Y., Pechlivanidis, I., and Clemenzi, I.: How does the seasonal forecast quality of hydrological extremes vary in space and time?, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-119, https://doi.org/10.5194/iahs2022-119, 2022.

Hydropower accounts for nearly half of Sweden’s electrical energy production, however the natural availability of water in the rivers is asymmetrically distributed throughout the year. As much as 70% of the annual streamflow is generated during the relatively short spring flood period (i.e., May – July) and the situation is further complicated by the fact that this occurs just when the energy demand begins to fall as the summer approaches. To offset these issues, operators must store as much of this excess streamflow in reservoirs for later use while still maintaining adequate storage capacity for flood control. Seasonal forecasts (up to 7 months ahead) of reservoir inflows are vital for the operational effectiveness of such endeavours.

To avoid the complexity and known biases associated with dynamic GCM-based seasonal forecasting approaches while improving the forecast skill, an analogue-weighted Ensemble Streamflow Prediction approach (A-ESP) was implemented in 84 sub-catchments across seven of the largest hydropower producing river systems in Sweden. The approach uses hydrological weather regimes (HWR) to select analogues from the of historical ensemble of meteorological data to force the hydrological models. Here, HWRs are classified large-scale atmospheric regimes used to describe “average” variability of local climate that results in precipitation events of similar frequency and magnitude. The selected analogues are pooled with the traditional ESP ensemble to make up the A-ESP seasonal forecast. We assess the forecast skill with respect to volume error and frequency of improvement using the ESP approach as the benchmark.

The results show that HWRs are a relatively simple yet useful tool for improving the forecast skill by offering objective criteria for analogue selection. Compared to the traditional ESP approach, the HWR-based A-ESP approach is able to reduce the inflow volume forecast errors by between 2-15% for all months and outperforms the ESP between 53-70% of the time. These results suggest that the analogue hypothesis to seasonal forecasting is still relevant and that using HWR for selection is effective. The comparative simplicity of this approach would offer operators improved forecasts that are both relatively computationally cheap and easy to implement, especially for operators using ESP-based forecast systems.    

How to cite: Yang, W., Foster, K., and Pechlivanidis, I. G.: Enhancement of seasonal hydrological forecasting with analogue selection of historical years, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-200, https://doi.org/10.5194/iahs2022-200, 2022.

Stochastic methods for synthetic streamflow simulations have traditionally supported operational management and planning of surface water systems in both short- and long-term futures. Here we focus on a particular strain of stochastic streamflow generators that use copulas, a generic statistical framework for formulating interdependencies, to resample streamflow series at single and multiple sites. Such stochastic simulators are based on a series of conditional probabilities that are inferred from joint probabilities between streamflow series in time and space. We discuss the core algorithm behind such stochastic samplers and provide a set of practical guidelines across a range of timescales, flow regimes and catchment characteristics on when and how these schemes should and can be developed. We then provide a generalized framework for altering the parameters of these samplers so that streamflow series can be generated under changing conditions, whether such changes are initiated by climate or human interventions. By informing these samplers with weather and climate indices, we show how the quality of streamflow projections can be significantly improved across a range of temporal scales and highlight the potential of such climatic-informed streamflow samplers for short-term predictions, particularly during high flow seasons.

How to cite: Nazemi, A. and Zaerpour, M.: Toward a unified stochastic framework for projection and prediction of streamflow under changing conditions, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-291, https://doi.org/10.5194/iahs2022-291, 2022.

An increasing number of countries and river basins worldwide are confronted with droughts. Inevitably, this leads to an increasing pressure on the available water resources and associated drought risks. In order to prevent severe drought impacts, such as food insecurity and possibly even conflicts on water use, timely and focused drought mitigation measures need to be taken. Here, early stage drought forecasts can be of great value, especially when they include local sector-specific information.

Over the past years Deltares developed the Global Flood Forecasting and Information System (GLOFFIS). Up until recent its focus was on floods but now seasonal forecasts have been included to provide drought early warning information. GLOFFIS is based on the wflow framework for hydrological modelling which is embedded within the Delft-FEWS forecast production system that is applied in many countries worldwide.

Wflow relies on so-called pedo-transfer functions that translate input base maps such as land use and soil type to estimate model parameter values. Herewith the model requires little calibration and can easily be implemented for additional basins. The set-up of GLOFFIS is global, yet for the drought forecasting component it was decided to include only high-resolution basin scale models since many global initiatives already exist and the resolution of global models generally lacks granularity to inform decision making. The system is flexible and contains workflows to be easily extended with hydrological models for new basins on demand. The drought forecasts are informed by means of standard drought indicators like discharge anomalies, soil moisture deficit, evaporation deficit indices. This list can be extended depending on the local need. So far hydrological models have been included for the Paraná, Lempa, Rhine and Niger river basins. The application for the Parana will be demonstrated as this basin in South-America currently experiences a severe drought.

How to cite: Sperna Weiland, F., Trambauer, P., Hendriks, D., den Toom, M., Weerts, A., and Verkade, J.: Aiming for globally available but locally relevant seasonal drought risk forecasts with GLOFFIS, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-415, https://doi.org/10.5194/iahs2022-415, 2022.

In France, groundwater is an important resource for industry, irrigation, and drinking water. As the dry periods become more frequent and longer due to the changing climate, it is crucial to forecast water table level evolution for the forthcoming months. Regional groundwater flow models can be useful to reach this objective (e.g., Mackay et al., 2015). Downscaled atmospheric seasonal forecasts are generally used to feed hydrological surface models, which provide surface conditions (drainage, runoff) to groundwater flow models.

Our application aims at using the Aqui-FR hydro-geological modelling platform for assessing the groundwater level for the forthcoming months. Aqui-FR was developed in order to gather in a single numerical tool several regional hydrogeological models covering much of the French metropolitan area (Vergnes et al., 2019). It allows to simulate the evolution of groundwater resources on short to long term periods.

Since the beginning of 2020, a real time prototype has been set up. Every month, it provides a status of the water table and a monthly forecast for the next six months (see figure) using the atmospheric seasonal forecast produced at Météo-France (Voldoire et al., 2019). These results are posted on an experimental website accessible to a group of beta-users. Some of them rely on this tool to help them in their decision making in a drought alert situation. For the first time in 2021, a hydro-geological seasonal forecast bulletin has been produced and provided to the national committee for drought monitoring and anticipation.

How to cite: Willemet, J.-M., Munier, S., Besson, F., Etchevers, P., Le Moigne, P., Rousset, F., Soubeyroux, J.-M., Viel, C., Habets, F., Ackerer, P., Amraoui, N., de Dreuzy, J.-R., Gallois, N., Magand, C., Thiery, D., and Vergnes, J.-P.: Aqui-FR: towards a hydro-geological seasonal forecasting system for metropolitan France, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-525, https://doi.org/10.5194/iahs2022-525, 2022.

How to cite: Portele, T. C., Lorenz, C., Dibrani, B., Laux, P., Bliefernicht, J., and Kunstmann, H.: Seasonal forecasts offer economic benefit for hydrological decision making in semi‐arid regions, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-546, https://doi.org/10.5194/iahs2022-546, 2022.

Hydrological drought conditions have negative implications for environmental and socio-economic activities in France. Operational drought forecasts with adequate timing and quality can guide decision-makers to prepare and make informed decisions on water management. Available ensemble meteorological forecasting systems, used to force hydrological models, differ in configuration, resolution, timescale, release frequency, and accuracy. These differences in atmospheric forecasting systems limit their usefulness for operational drought forecasts. To address this challenge, we proposed an approach to produce seamless atmospheric forecasts (SeamF). SeamF were constructed from the extended- and long-range ensemble prediction systems (ECMWF and Météo-France, respectively) using a random concatenation member by member method in a first step.

The new SeamF system provides a 144-day outlook of hourly and daily precipitation and temperature variables, updated once a week, suitable for hydrological modelling over the country. In addition to the practicability of these integrated forecasts, as compared to stand-alone extended- or long-range forecasts, the level of deterministic and probabilistic verifications metrics suggests a promising improvement in hydrological applications. This approach builds a basis for more sophisticated seamless forecast production and application for hydrological decision-making in France.

How to cite: Gbangou, T., Viel, C., Rousset, F., Willemet, J.-M., Etchevers, P., Munier, S., Ramos, M.-H., and Bourgin, F.: Seamless meteorological forecast production and evaluation towards hydrological decision-making in France : CIPRHES projet, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-579, https://doi.org/10.5194/iahs2022-579, 2022.

Much of northern Tunisia regularly experiences extremes of drought and flooding, with high rainfall variability. Development of reliable and accurate seasonal rainfall forecasts can provide valuable information to help mitigate some of the outcome of floods and enhance water management, particularly for agriculture. Ensemble monthly rainfall forecasts are carried out over horizons ranging from 1 to 6 months using a hybrid wavelet neural network model. The hybrid model called MWD-NARX based on a non-linear autoregressive network with exogenous inputs (NARX) coupled to multiresolution wavelet decomposition (MWD) is developed in this work. First, The MWD is used to decompose the data into different components on various time scale. Then to predict each precipitation decomposition the NARX ensemble model is employed. For an operational forecasting, the forecasts obtained from the decompositions are summed to represent the true precipitation forecast value. The outcomes of MWD-NARX are compared with Artificial Neural Networks (ANN). The seasonal forecasts of average precipitation by sub-basins of the Medjerda river basin are carried out. Large scale climate teleconnection indicators of ENSO, PDO, NAO and Mediterranean Oscillation were used as inputs to the model. The results indicate that exogenous inputs like climatic indices clearly improves the accuracy of forecasts in terms of the coefficient R² on 82% of SBVs compared to a model that uses only climate indices as inputs with 1 month delay time. It increases then the forecast lead-time up to 6 months. The same conclusion is made when compared to an ANN. The correlation coefficient between observed and forecasted monthly precipitation is ranging from 0.5 to 0.8. It was also found that the MWD-NARX underestimates the extremes. The spatial variability of the quality of the forecasts depends mainly on the local effect of precipitation more than on the quality of the hydrological data observed on the forecasts. It can be concluded that exogenous inputs like climate indices can add some additional information to enhance monthly precipitation forecasts at longer lead-times. The forecasting model coupled to data pre-processing method made it possible to produce very satisfactory forecasts of non-stationary data by extracting significant modes of variability.

How to cite: Ouachani, R., Bargaoui, Z., and Ouarda, T.: Seasonal Precipitation forecasting with large scale climate predictors: A hybrid  wavelet multiresolution -NARX scheme, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-732, https://doi.org/10.5194/iahs2022-732, 2022.

The use of artificial intelligence models to simulate and forecast groundwater has become more and more popular since the early 1990ies. These models able to forecast groundwater levels are increasingly being used to support decision making for water resources management and climate change impact local assessments. This study presents a comparison of different algorithms to simulate and forecast the piezometry of the Astian sands aquifer in the Occitanie region (France) at sub seasonal to seasonal scales. The aim is to set up a transferable methodology allowing quick implementation on different aquifers, to lay the foundations for a tool for water resources management focusing on droughts.

Different models were tested in this study with prediction horizons from a few weeks and up to the hydrological year. Models used include (1) Multi-layer Perceptron (MLP), (2) recurrent models such as Long short-term memory models, Gated Recurrent Unit (GRU), Non-Linear Regressive Neural Networks (NARX), Dual Stage Attention Based Recurrent Networks (DA-RNN); (3) a Temporal Convolutional Network (TCN); (4) a Reservoir Computing Network (Echo State Network). Pre-processing methods were tested to increase performance and attempt to better reproduce higher frequencies, especially the Ensemble Empirical Modal Decomposition (EEMD). Input variables were the daily rainfall data, temperature data and piezometry. To validate the model and assess the performance of the simulations, different metrics were used (Nash Sutcliffe Efficiency, Kling Gupta Efficiency, Coefficient of persistency). The loss function was a combination of RMSE and NSE, the optimizer used was Adam, the training was done on two thirds of the time series and validation and testing on the remaining data.

Several models achieved high performances: DA-RNN, the GRU, MLP and NARX (NSE>0.91 and Cp>0.6 at 20 days). Predictions were carried out iteratively with a 20-days window, reinjecting the predicted data and re-training the model (adaptative model) until a prediction of more than 400 days was reached (using observed rainfall and temperature). The stability and robustness of the model is further investigated through additional cross validation, estimation of confidence intervals, and simulations in different hydrogeological contexts, to aim toward a transferable tool for various aquifers.

How to cite: Duran, L., Mouni, R., Duperron, V., Maisonnasse, L., Artigue, G., Pistre, S., and Johannet, A.: Prediction of water level variations in aquifers using recurrent and convolutional neural networks, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-750, https://doi.org/10.5194/iahs2022-750, 2022.

For an optimized use of water resources for irrigation or power generation, knowledge about their expected availability in the coming months is essential. The European Centre for Medium-Range Weather Forecasts (ECMWF) issues monthly ensemble forecasts for the next seven months (SEAS5 model). In order to be able to use these for Germany, we present an analysis of the temperature and precipitation forecast quality for Germany with a resolution on district level. ERA5 reanalysis data from ECMWF and raster observation data from the German Weather Service are used as a basis for comparison. To increase the performance, two bias corrections were tested, a mean value correction and a correction of both the mean value and the standard deviation. Using several quality parameters such as the Continuous Ranked Probability Skill Score (CRPSS), the Brier Skill Score (BSS) and the Receiver Operating Characteristic (ROC) curves, it is shown that the uncorrected SEAS5 seasonal forecasts at monthly resolution provide a significantly increased performance compared to purely climatological forecasts, mainly in the first Lead Month. The reliability (RV) averages 78% across years, months, regions, and variables, with the first forecast month providing a particularly good RV of 86%. All prediction months have positive CRPSS for the precipitation at all times for all regions, and the bias correction to the mean also improves the skill by an average of 33% based on the CRPSS. In addition, both bias corrections significantly improve the prediction quality of the above-average extremes (> 90th percentile) for temperatures and precipitation. Thus, the SEAS5 seasonal forecasts, together with subsequent bias correction procedures, open up the possibility of estimating expected water availability at the county level with forecast periods of several months.

How to cite: Weber, J. N., Lorenz, C., Portele, T., and Kunstmann, H.: Predictive performance of bias-corrected seasonal SEAS5 forecasts for Germany, IAHS-AISH Scientific Assembly 2022, Montpellier, France, 29 May–3 Jun 2022, IAHS2022-751, https://doi.org/10.5194/iahs2022-751, 2022.