Flash Flood Forecasting Using Integrated Meteorological– Hydrological–Hydraulic Modeling: Application in a Mediterranean River †

: This study aims at assessing meteorological, hydrological, and hydraulic modeling to develop a ﬂash ﬂood forecasting tool. The ﬂash ﬂood that occurred in the Evrotas River Basin (ERB) on 26 January 2023 is used as a case study. Precipitation over 150 mm and water depths exceeding 2.5 m were recorded. The meteorological model initialized one day before ﬂooding and simulated precipitation; the hydrological model, using meteorological input data, simulated discharge; and the hydraulic model, using discharge, estimated water depth at a bridge. The results indicate that the system can provide skillful and timely ﬂash ﬂood forecasts, thereby facilitating ﬂood warnings.


Introduction
Flash floods cause human loss and infrastructure damage every year worldwide and are among the natural hazards that are strongly influenced by climate change [1]. Accurate forecasting of flash floods is a challenging task that poses the need for a holistic, multimodel approach. In this framework, the Institute of Marine Biological Resources and Inland Waters (IMBRIW) of the Hellenic Centre for Marine Research (HCMR) has been operating and frequently upgrading a high-resolution forecasting system consisting of a meteorological and a hydrological model since September 2015. The forecasts are available online from the website: https://meteo.hcmr.gr (accessed on 22 August 2023). To work towards the development of a more sophisticated forecasting tool tailored for river flash floods, this study aims at bridging numerical weather predictions with river routing and inundation modeling, combining all in an integrated system consisting of three different models: meteorological, hydrological, and hydraulic-hydrodynamic.
To assess the capabilities of the modeling system, the severe flash flood event that occurred in the Evrotas River Basin (ERB) on 26 January 2023 and caused damage to roads, cultivations, buildings, etc., is used as a case study. The flood caused severe damage mainly in agricultural areas as the vast majority of the ERB is covered by natural and agricultural areas, while urban areas account for 1%. Therefore, the flood had crucial socioeconomical implications. The collected non-conventional flood data (i.e., photographs, videos, and mass media reports) showed that the maximum water depth at the Skala's bridge was approximately 6 m, almost equal to the bridge height. Hence, vehicles passing through the Skala's bridge were prohibited because the water reached the bridge, increasing the risk of accidents.
In this study, the meteorological model (WRF-ARW) was properly configured in forecasting mode (i.e., boundary conditions based on global model forecasts rather than analyses) to predict, approximately 1 day before the flood, the atmospheric conditions that triggered the persistent thunderstorm and heavy rainfall over the ERB. The model was set up on four nested domains covering the Mediterranean area with a horizontal resolution of 36 km, the eastern Mediterranean with 12 km, Greece with 4 km, and the ERB with 1 km. The simulation was initialized on 25 January at 12:00 UTC using the available Global Forecasting System (GFS) operational analysis of the National Centers for Environmental Prediction (NCEP) with a horizontal resolution of 0.25 • . The boundary conditions were based on the GFS operational forecasts with a time step of 3 h, while the initial sea surface temperature (SST) field was based on the real-time global (RTG) SST (resolution 0.083 • ) for 24 January (1 day before initialization, similar to operational systems, e.g., [6]), also obtained from the NCEP.
Afterwards, the hydrological model (WRF-Hydro), using hourly input data provided by the meteorological simulation such as precipitation rate (find more information in [3]), simulated the flood discharge (i.e., flood hydrograph) in the streams of the ERB. With this aim in mind, WRF-hydro was configured on the ERB with a high horizontal resolution of 100 m. For the construction of the high-resolution terrain data required in routing grids (i.e., topography, flow direction, channel grid, and stream order), the void- Finally, the hydraulic-hydrodynamic model (HEC-RAS) used the discharges provided by the hydrological simulation to estimate the water-level increase at the Skala's bridge located at the lower part of the river. The water almost overran the Skala's bridge and, thus, it was selected to be simulated and analyzed. A location slightly upstream of the Skala's bridge was used as hydrograph input for the HEC-RAS simulation. For the HEC-RAS model setup, very fine river geometry data (unmanned-aerial-vehicle-derived DEM) were used to increase the topographical accuracy ( Figure 1c). The remaining parameters were set up according to the user manual. The geometry of the Skala's bridge was retrieved from the topographical survey for the eastern Peloponnese flood management risk plans. Hydraulic model calibration was accomplished by comparing simulated water depth with non-conventional water depth data (i.e., estimated observed water depth based on photographs and videos of the flood event). The setups of the three models are similar to those presented in a previous publication [6].
In order to assess the results of simulations, various measurements during the flash flood were

Results
The flash flood event in the ERB that occurred on 26 January 2023 inundated several areas as derived from the calculation of the modified normalized difference water index (MNDWI [11]) (Figure 2a) based on the Sentinel-2A satellite image of 28 January 2023, approximately 48 h after the flood peak. The normalized difference water index (NDWI [12]) is widely used to separate water from the background based on a threshold value (c-h) Spatial distribution of 1-h precipitation (mm) from the meteorological forecast and discharge (m 3 s −1 ) from the hydrological forecast for 26 January from 07:00 to 12:00 LT, respectively.

Results
The flash flood event in the ERB that occurred on 26 January 2023 inundated several areas as derived from the calculation of the modified normalized difference water index (MNDWI [11]) (Figure 2a) based on the Sentinel-2A satellite image of 28 January 2023, approximately 48 h after the flood peak. The normalized difference water index (NDWI [12]) is widely used to separate water from the background based on a threshold value which is scene-specific according to the study area and the satellite image used. Additionally, MNDWI is enhanced and preferred in cases where built-up areas are correlated with open water in NDWI [11]. The inundation was caused by the main Evrotas river, especially at the downstream parts, as well as by several small independent streams (not shown) that connect the surrounding hills with the lowlands and which caused increased surface runoff without ending up in the sea.
The WRF-ARW model simulated the storm event and the precipitation pattern over the ERB that caused the flash flood realistically. Despite some small spatial and quantitative forecast misses, the daily precipitation forecast for 26 January agreed with the measurements (Figure 2b). Figure 2c-h demonstrate the spatial distribution of precipitation and the increase in discharge in Evrotas streams from 07:00 to 12:00 (local time; LT) of 26 January. The storm slowly moved from the southwestern to the northeastern parts of the ERB from 07:00 to 10:00 LT (Figure 2c-f). A peak of 1-h accumulated precipitation slightly exceeding 50 mm was estimated at 09:00 at the northeastern parts of the ERB (Figure 2e). It is important to note that the meteorological simulation produced precipitation before 07:00 LT and after 12:00 LT over the ERB, but only the critical phase of the storm causing the flash flood is presented here. Regarding the hydrological response, the WRF-Hydro model captured the flood features well, estimating maximum discharges ranging from 775 m 3 s −1 to 769 m 3 s −1 and from 10:00 to 12:00 LT, respectively (Figure 2f-h).
The HEC-RAS model simulated the maximum water depth at the Skala's bridge (Figure 3a). The simulated maximum water depth was estimated at about 6 m, which is very close to the observed one, according to the collected non-conventional flood data. Finally, similar results are observed in Figure 3b, where the water depth time series slightly upstream of the validation location (bridge) near the Skala town are presented.

Discussion and Conclusions
The findings of this study indicate that the integrated, three-model, flash flood forecasting system can provide skillful forecasts. The simulation results were evaluated using precipitation measurements and estimated observed water depth based on photographs and videos of the flood event, presenting very good agreement between the two. Thus, this study highlights the benefits of the combined use of numerical weather prediction and river modeling to forecast flash flood events in an accurate and timely manner. This multi-model methodological approach should be applied in more case studies, also including larger river basins, to better set up the modeling system. It is noteworthy that the network of hydrological stations of IMBRIW-HCMR is continuously being expanded to include measurements and estimates of discharge. These data will be soon available for model calibration and validation of simulation results.

Discussion and Conclusions
The findings of this study indicate that the integrated, three-model, flash flood forecasting system can provide skillful forecasts. The simulation results were evaluated using precipitation measurements and estimated observed water depth based on photographs and videos of the flood event, presenting very good agreement between the two. Thus, this study highlights the benefits of the combined use of numerical weather prediction and river modeling to forecast flash flood events in an accurate and timely manner. This multi-model methodological approach should be applied in more case studies, also including larger river basins, to better set up the modeling system. It is noteworthy that the network of hydrological stations of IMBRIW-HCMR is continuously being expanded to include measurements and estimates of discharge. These data will be soon available for model calibration and validation of simulation results.
In conclusion, the proposed modeling framework can be applied as a flood warning tool for emergency responses. The modeling system can facilitate not only operational forecasting activities but also research efforts in an interdisciplinary context of flood simulation studies. The physically based modeling approach is able to increase the forecast lead time of storms and flash floods and, therefore, it could be exploited towards a more complete management and mitigation design for floods.