Search

Estimating spatially distributed hydrological parameters in ungauged catchments poses a challenging regionalization problem and requires imposing spatial constraints given the sparsity of discharge data. A possible approach is to search for a transfer function that quantitatively relates physical descriptors to conceptual model parameters. This paper introduces a Hybrid Data Assimilation and Parameter Regionalization (HDA‐PR) approach incorporating learnable regionalization mappings, based on either multi‐linear regression or artificial neural networks (ANNs), into a differentiable hydrological model. This approach demonstrates how two differentiable codes can be linked and their gradients chained, enabling the exploitation of heterogeneous data sets across extensive spatio‐temporal computational domains within a high‐dimensional regionalization context, using accurate adjoint‐based gradients. The inverse problem is tackled with a multi‐gauge calibration cost function accounting for information from multiple observation sites. HDA‐PR was tested on high‐resolution, hourly and kilometric regional modeling of 126 flash‐flood‐prone catchments in the French Mediterranean region. The results highlight a strong regionalization performance of HDA‐PR especially in the most challenging upstream‐to‐downstream extrapolation scenario with ANN, achieving median Nash‐Sutcliffe efficiency (NSE) scores from 0.6 to 0.71 for spatial, temporal, spatio‐temporal validations, and improving NSE by up to 30% on average compared to the baseline model calibrated with lumped parameters. Multiple evaluation metrics based on flood‐oriented hydrological signatures also indicate that the use of an ANN leads to better performances than a multi‐linear regression in a validation context. ANN enables to learn a non‐linear descriptors‐to‐parameters mapping which provides better model controllability than a linear mapping for complex calibration cases. Key Points: Novel approach for regional calibration of a distributed hydrologic model using learnable and non‐linear descriptors‐to‐parameters mappingsOriginal combination of numerical adjoint model and neural network Jacobian: accurate gradients enable high‐dimensional optimizationExtensive case study in flash‐flood‐prone Mediterranean region shows effective regionalization of high‐resolution model with neural network [ABSTRACT FROM AUTHOR]

The French national service in charge of flood forecasting (SCHAPI) has implemented, in collaboration with Irstea and Météo-France, a national flash flood warning system, Vigicrues Flash, to provide timely warnings for small-to-medium ungauged basins. Operational since March 2017 for about 10,000 municipalities, this system complements flood warnings produced by the Vigicrues procedure for French monitored rivers. Vigicrues Flash is based on a discharge-threshold flood warning method called AIGA. In the current operational version, a simplified hourly distributed rainfall-runoff model ingests radar-gauge Quantitative Precipitation Estimate (QPE) grids from Météo-France at a 1-km² resolution to produce real-time peak discharge estimates on any river cell. This hourly event-based distributed model is coupled to a continuous daily rainfall-runoff model, which provides baseflow and a soil moisture index (for each 1-km² pixel) at the beginning of the hourly simulation. Every 15 minutes, the discharges are compared to regionalized flood frequency estimates, which were derived from long-term streamflow simulations. The automated warning system determines rivers exceeding the high flood and very high flood thresholds (associated to years of return periods), as well as the associated municipalities that might be impacted. Flood hazard maps are published on a web platform and warning messages are automatically sent to registered users to help them better mitigate flood risk impacts.To better anticipate flash flood events and extend the coverage of this automated service, the warning system is being enhanced to include a single fully distributed hydrologic model, run at sub-hourly time step. Several calibration and regionalization methods are being tested to better account for basins spatial heterogeneities while maintaining consistency across spatial scales. Evaluation is carried out for about 2000 French basins on the 2008-2018 period to show improvements in terms of flash flood event detection and effective warning lead time. Other enhancements include integrating high-resolution precipitation nowcasts available on a 6-hour forecast horizon, accounting for and reducing hydrometeorological uncertainties via ensemble forecasting and data assimilation, and incorporating a vulnerability assessment component to provide risk-based decision-relevant warnings. [ABSTRACT FROM AUTHOR]

Summary: Reliable and skillful precipitation ensemble forecasts are necessary to produce reliable and skilful hydrologic ensemble forecasts. It is well known that, in general, raw precipitation ensemble forecasts from the numerical weather prediction (NWP) models are not very reliable and that, for short-range prediction, human forecasters add significant skill to the NWP-generated single-valued quantitative precipitation forecasts (QPF). In this paper, we describe and evaluate a statistical procedure for producing precipitation ensemble forecasts from single-valued QPFs. The procedure is based on the bivariate probability distribution between the observed precipitation and the single-valued QPF. The distribution is modeled as a mixed-type in which the relationship between the positive observed precipitation and positive forecast precipitation is assumed to be bivariate meta-Gaussian. We also describe and comparatively evaluate a generalized meta-Gaussian model in which the model parameter is optimized by minimizing the mean Continuous Ranked Probability Score. The performance of these procedures is assessed through dependent and cross validation using data for selected river basins in the service areas of the Arkansas-Red Basin, California-Nevada and Middle-Atlantic River Forecast Centers of the National Weather Service. The validation results show that, overall, the precipitation ensembles generated by the proposed procedures are reliable and capture the skill in the conditioning single-valued forecasts very well. [ABSTRACT FROM AUTHOR]

Highlights: [•] The HMOS produces ensemble streamflow forecasts from single-valued forecasts. [•] Here, the single-valued forecasts are 6-h operational forecasts out to 5days. [•] The HMOS uses the single-valued forecast, QPF and the recent observed streamflow. [•] The HMOS ensemble forecast provides an estimate of the total predictive uncertainty. [•] In general, the HMOS ensemble forecasts are reliable and skillful. [Copyright &y& Elsevier]