INFORMATION
- Category area_flood
Main goals
The efficient combination of physically-based numerical models with Artificial Intelligence (AI) techniques can improve the flood risk forecasts delivered by flood Early Warning Systems (EWS). The project will take as starting point the physically-based forecasts implemented in an already existing operational EWS (MERLIN) that delivers daily flood forecasts in several basins of NW Spain. Different AI techniques will be tested with the objective of improving the reliability of the rainfall, streamflow and water depth forecasts provided by MERLIN.
To achieve this goal, the following specific objectives will be addressed during the project:
- Goal 1. To improve the short-term rainfall forecasts provided by a physically-based weather model by analysing forecast errors and implementing error-correction techniques using supervised machine learning algorithms.
- Goal 2. To improve the streamflow forecasts given by physically-based hydrological models using ensemble modelling methods based on machine learning algorithms.
- Goal 3. To improve the efficiency of water depth forecasts given by high-resolution 2D hydraulic models using a surrogate modelling approach.
Methodology
The project is organised around 6 WP. The first one (WP1) is focused on a thorough evaluation of the physically-based rainfall, streamflow and inundation forecasts provided by the current implementation of the EWS MERLIN. This evaluation will allow us to quantify the main weaknesses of the current methodologies used in MERLIN, and will also serve as the reference against which the performance of the new methods proposed in this project will be evaluated.
WP2, WP3 and WP4 are focused on the development of new methods combining physically-based models and different type of AI techniques, in order to improve the current flood hazard forecasts given by the 3 modules of MERLIN. Different methods will be tested, tuned and evaluated in order to find the most suitable ones to be implemented in an operational EWS, as MERLIN.
In WP5 we will explore the added value of the streamflow forecasts provided by MERLIN to improve the predictions of coastal hydrodynamic models during flood events. We will use as pilot case the Ría de Vigo estuary, in which there is an operational hydrodynamic model implemented by Meteogalicia that gives daily predictions of coastal currents, water temperature and salinity.
WP6 is devoted to dissemination of the project results and transfer of knowledge. We will implement the most efficient methodologies developed in the project in the operational version of MERLIN.
WP 1. Performance of physically-based forecast models implemented in MERLIN
The aim of this WP is to assess the current performance of the EWS MERLIN, which will be the starting point of the project. We will evaluate the performance of the three modules included in MERLIN by comparing the numerical forecasts with the available observed data of rainfall maps in the region covered by the EWS, hydrographs at different control points, and depths in Areas of Potential Significant Flood Risk (APSFR). The performance of the rainfall forecast will be addressed first, since a poor rainfall forecast will affect the streamflow prediction, just as a poor streamflow calculation will affect the water depths computed by the hydraulic module. Efforts will focus on qualitatively identifying events with a poor fit in terms of rainfall, peak discharge or flood warnings.
- Task 1.1. Evaluation of the short-term rainfall forecasts given by the WRF model
- Task 1.2. Evaluation of the hydrograph forecasts given by HEC-HMS
- Task 1.3. Evaluation of the water depth forecasts given by Iber
WP 2. Improving short-term rainfall forecasts with error correction learning techniques
The aim of this WP is to improve the rainfall forecasts delivered by physically-based weather models that are typically implemented in EWS. We will work with the forecasts delivered by the model WRF model, which is used by Meteogalicia to provide daily forecasts of rainfall in the whole region of Galicia (NW Spain), which are then used as input in the hydrological module of MERLIN.
- Task 2.1. Diagnosis of the WRF rainfall forecast error
- Task 2.2. Correction of the WRF rainfall forecast error
WP3. Improving physically-based hydrological forecasts with ensemble modelling
WP3 is focused on applying ensemble modelling techniques to improve the streamflow forecasts given by a single physically-based hydrological model. The aim is to complement the physically-based forecast with DL-based forecasts, providing an AI-based ensemble prediction. This kind of technique has been used with success in previous hydrological modelling studies, and here we will explore its extension to hydrological forecasts within an EWS.
- Task 3.1. Streamflow forecasting with DL techniques
- Task 3.2. Streamflow forecasting with a lumped hydrological model
- Task 3.3. Multi-model ensemble streamflow forecasting
- Task 3.4. Forecasting reservoir release
WP4. Improving water depth forecasts in APSFR with a surrogate modelling approach
The aim of WP4 is to reduce the computational time needed to obtain water depth forecasts in Areas of Potential Significant Flood Risk (APSFR). The standard approach for this purpose is to use a 2D hydraulic model. In the case of MERLIN, this is done with the software Iber+, which is the High Performance Computing version of the software Iber, in 14 APSFR. Even if the numerical efficiency of Iber+ is very high, thanks to its implementation in Graphics Processing Units (GPU), the computational time is still high for real time applications (as it is the case of an EWS), limiting the spatial resolution of the model and/or the number of runs that can be performed daily in order to forecast different scenarios or to quantify the prediction uncertainty. In this WP we will use an AI-based surrogate modelling approach to produce very fast water depth forecasts in two pilot APSFR. The surrogate models will be trained using a collection of Iber+ precalculated runs, corresponding to a suitable set of representative boundary conditions.
- Task 4.1. Implementation of Iber+ models in the pilot cases
- Task 4.2. Surrogate modelling of water depths
- Task 4.3. Installation of a camera-based water level estimation technology
- Task 4.4. Processing of the camera images
WP5. Exploiting streamflow forecasts to improve operational coastal models
- Task 5.1. Extension of streamflow predictions to river mouths
- Task 5.2. Impact of streamflow forecasts on predicted coastal dynamics
- Task 5.3. Pipeline discharge forecasts to operational oceanographic models
WP6. Dissemination and transfer
- Task 6.1. Project internal reports, made available through the website of the research team
- Task 6.2. Presentation of results in scientific journals and conferences
- Task 6.3. Transfer of technology to the regional water administration (Augas de Galicia)
- Task 6.4. Educational outreach
Dissemination
The dissemination plan includes the following actions:
Internal reports and scripts
- D1. Report on the current performance of MERLIN.
- D2. Scripts for error-correction learning uploaded to a public repository.
- D3. Model MHIA (Task 3.2) uploaded to a public repository.
- D4. Scripts for ensemble hydrological modelling (Task 3.4) uploaded to a public repository.
- D5. Scripts for surrogate modelling with Iber (Task 4.2).
- D6. Report on the impact of streamflow forecasts on coastal dynamics model.
Scientific papers
- Cea, L., Sañudo, E., Montalvo, C., Farfán, J., Puertas, J., Tamagnone, P. (2025). Recent advances and future challenges in urban pluvial flood modelling. Urban Water Journal, 22(2), 149–173. https://doi.org/10.1080/1573062X.2024.2446528
- Farfán et al. (2025). Integrating Net Rainfall Calculation in Deep Learning-Based Surrogate Modeling Frameworks for 2D Flood Prediction. Under review
National and international Conferences and Seminars
- Conference 1
- Conference 2
People
- Luis Cea | luis.cea@udc.es
- Jerónimo Puertas | jeronimo.puertas@udc.es
- Luis Pena | luis.pena@udc.es
- Ignacio Fraga | ignacio.fraga@udc.es
- Esteban Sañudo | e.sanudo@udc.es
- Juan Farfán | j.farfan@udc.es
- Carlos Montalvo | carlos.montalvo@udc.es
- Martín Montenegro |