The 2021 summer flooding was an extremely rare event, driven by precipitation extremes that exceed Dutch design levels for flood protection of relatively small rivers and waterways. However, similar events in neighboring locations cannot be ruled out in the near future. The implications of such extreme rainfall amounts will vary by region, subject to local topography, water systems, and societal exposure. We explore the diversity of potential flood impacts induced by a similar event by constructing impact-oriented event storylines for characteristic water management regions in the Netherlands. The plausibility of the storylines is underlined by using physical evidence, proven impact-modelling concepts, and expert judgment successfully assessing the (sometimes unexpected) outcomes. The approach supports impact assessment and risk management of extraordinary rainfall and flood events. The outcomes show the relevance for crisis management and spatial policies, and confirms the need for in depth-analysis to assess concrete adaptation options.@en

Flash flood early warning requires accurate rainfall forecasts with a high spatial and temporal resolution. As the first few hours ahead are already not sufficiently well captured by the rainfall forecasts of numerical weather prediction (NWP) models, radar rainfall nowcasting can provide an alternative. Because this observation-based method quickly loses skill after the first 2 hr of the forecast, it needs to be combined with NWP forecasts to extend the skillful lead time of short-term rainfall forecasts, which should increase decision-making times. We implemented an adaptive scale-dependent ensemble blending method in the open-source pysteps library, based on the Short-Term Ensemble Prediction System scheme. In this implementation, the extrapolation (ensemble) nowcast, (ensemble) NWP, and noise components are combined with skill-dependent weights that vary per spatial scale level. To constrain the (dis)appearance of rain in the ensemble members to regions around the rainy areas, we have developed a Lagrangian blended probability matching scheme and incremental masking strategy. We describe the implementation details and evaluate the method using three heavy and extreme (July 2021) rainfall events in four Belgian and Dutch catchments. We benchmark the results of the 48-member blended forecasts against the Belgian NWP forecast, a 48-member nowcast, and a simple 48-member linear blending approach. Both on the radar domain and catchment scale, the introduced blending approach predominantly performs similarly or better than only nowcasting (in terms of event-averaged continuous ranked probability score and critical success index values) and adds value compared with NWP for the first hours of the forecast, although the difference, particularly with the linear blending method, reduces when we focus on catchment-average cumulative rainfall sums instead of instantaneous rainfall rates. By properly combining observations and NWP forecasts, blending methods such as these are a crucial component of seamless prediction systems.

@en

This study focuses on exploring the potential of using Long Short-Term Memory networks (LSTMs) for low-flow forecasting for the Rhine River at Lobith on a daily scale with lead times up to 46 days ahead. A novel LSTM-based model architecture is designed to leverage both historical observation and forecasted meteorological data to carry out multi-step discharge time series forecasting. The feature and target selection for this deep learning (DL) model involves evaluating the use of different spatial resolutions for meteorological forcing (basin-averaged or subbasin-averaged), the impact of incorporating past discharge observations, and the use of different target variables (discharge Q or time-differenced discharge dQ). Then, the model is trained using the ERA5 dataset as meteorological forcing, and employed for operational forecast with ECMWF seasonal forecast (SEAS5) data. The forecast results are compared to a benchmark process-based model, wflow_sbm. This study also explores the flexibility of the DL model by fine-tuning the pretrained model with limited SEAS5 dataset. Key findings from feature and target selection include: (1) opting for subbasin-averaged meteorological variables significantly improves model performance compared to a basin-averaged approach. (2) Utilizing dQ as the target variable greatly boosts short-term forecast accuracy compared to using Q, with a mean absolute error (MAE) of 25 m³ s⁻¹ and mean absolute percentage error (MAPE) of 0.02 for the first lead time, ensuring reliability and accuracy at the onset of the forecast horizon. (3) While incorporating historical discharge improves the forecasting of Q, its impact on predicting dQ is less pronounced for short lead times. In the operational forecast with SEAS5, compared to the wflow_sbm model, the DL model exhibits skill in forecasting low flows as evidenced by Continuous Ranked Probability Skill Score (CRPSS) median values of all lead times above zero, and better accuracy in forecasting drought events within short lead times. The wflow_sbm model shows higher accuracy for longer lead times. In the exploration of fine-tuning approach, the fine-tuned model generates marginal short-term enhancements in forecasting low-flow events over a non-fine-tuned model. Overall, this study contributes to advancing the field of low-flow forecasting using deep learning approach.

@en

To assess the potential of radar rainfall nowcasting for early warning, nowcasts for 659 events were used to construct discharge forecasts for 12 Dutch catchments. Four open-source nowcasting algorithms were tested: Rainymotion Sparse (RM-S), Rainymotion DenseRotation (RM-DR), Pysteps deterministic (PS-D), and probabilistic (PS-P) with 20 ensemble members. As benchmark, Eulerian Persistence (EP) and zero precipitation input (ZP) were used. For every 5-min step in the available nowcasts, a discharge forecast with a 12-hr forecast horizon was constructed. Simulations using the observed radar rainfall were used as reference. Rainfall and discharge forecast errors were found to increase with both increasing rainfall intensity and spatial variability. For the discharge forecasts, this relationship depends on the initial conditions, as the forecast error increases more quickly with rainfall intensity when the groundwater table is shallow. Overall, discharge forecasts using RM-DR, PS-D, and PS-P outperform the other methods. Threshold exceedance forecasts were assessed by using the maximum event discharge as threshold. Compared to benchmark ZP, an exceedance is, on average, forecast 223 (EP), 196 (RM-S), 213 (RM-DR), 119 (PS-D), and 143 min (PS-P) in advance. The EP results are counterbalanced by both a high false alarm ratio (FAR) and inconsistent forecasts. Contrarily, PS-D and PS-P produce lower FAR and inconsistency index values than all other methods. All methods advance short-term discharge forecasting compared to no rainfall forecasts at all, though all have shortcomings. As forecast rainfall volumes are a crucial factor in discharge forecasts, a future focus on improving this aspect in nowcasting is recommended.

@en

The presence of significant biases in real-time radar quantitative precipitation estimations (QPEs) limits its use in hydrometeorological forecasting systems. Here, we introduce CARROTS (Climatology-based Adjustments for Radar Rainfall in an OperaTional Setting), a set of fixed bias reduction factors, which vary per grid cell and day of the year. The factors are based on a historical set of 10 years of 5ĝ€¯min radar and reference rainfall data for the Netherlands. CARROTS is both operationally available and independent of real-time rain gauge availability and can thereby provide an alternative to current QPE adjustment practice. In addition, it can be used as benchmark for QPE algorithm development. We tested this method on the resulting rainfall estimates and discharge simulations for 12 Dutch catchments and polders. We validated the results against the operational mean field bias (MFB)-adjusted rainfall estimates and a reference dataset. This reference consists of the radar QPE, that combines an hourly MFB adjustment and a daily spatial adjustment using observations from 32 automatic and 319 manual rain gauges. Only the automatic gauges of this network are available in real time for the MFB adjustment. The resulting climatological correction factors show clear spatial and temporal patterns. Factors are higher away from the radars and higher from December through March than in other seasons, which is likely a result of sampling above the melting layer during the winter months. The MFB-adjusted QPE outperforms the CARROTS-corrected QPE when the country-average rainfall estimates are compared to the reference. However, annual rainfall sums from CARROTS are comparable to the reference and outperform the MFB-adjusted rainfall estimates for catchments away from the radars, where the MFB-adjusted QPE generally underestimates the rainfall amounts. This difference is absent for catchments closer to the radars. QPE underestimations are amplified when used in the hydrological model simulations. Discharge simulations using the QPE from CARROTS outperform those with the MFB-adjusted product for all but one basin. Moreover, the proposed factor derivation method is robust. It is hardly sensitive to leaving individual years out of the historical set and to the moving window length, given window sizes of more than a week.

@en