Use of radar rainfall estimates and forecasts to prevent flash flood in real time by using a road inundation warning system
Highlights
► Road inundation warning system represents a promising tool to flood risk management. ► Quantitative precipitation estimates and forecasts can improve road inundations detection. ► QPE and QPF are particularly useful to survey small ungauged watersheds. ► Observed road inundations could be used to improve hydrological modelling.
Introduction
Mediterranean regions are subject to violent flash floods, resulting in heavy economic damages, estimated at a billion Euros in France over the last two decades Gaume et al., 2004 and, in some cases, human casualties, as illustrated by the recent events in Nîmes (1988), Vaison-la-Romaine (1992), Tarragona (1994), Biescas (1996), Corbière (1999), Alger (2001), Gard (2002) and Var (2010). Flash floods are identified as the consequence of an intense rain event producing several hundreds of mm in few hours (Creutin and Borga, 2003, Collier, 2007, Younis et al., 2008). During this type of event, spatial and temporal variability of rainfall appears to be the main factor controlling the hydrological response (Chancibault et al., 2007, Le Lay and Saulnier, 2007) and this evolution is very difficult to predict. Flash floods typically occur in quick response watersheds for two main reasons: (i) a short concentration time due to the size generally under few hundreds km2, (ii) flood flows that are essentially composed of surface runoff water or at least fast responding runoff processes (Creutin et al., 2009). That makes very difficult for emergency management services to anticipate and deliver flash flood warnings in real time.
This is particularly true concerning the road network that could be strongly affected during flash floods. In a situation of risk, the state of the road network has appeared as a major concern within these affected regions for two main reasons. First, many flash flood victims are in fact motor vehicle passengers trapped in inundated roads (Staes et al., 1994, Bourque et al., 2007). Second, emergency services require a clear overview of possible road conditions in order to efficiently plan interventions and identify safe access or evacuation routes.
Based on these considerations, a Road Inundation Warning System (RIWS) for flash flood prone areas has been recently developed and tested on the North part of the Gard region (France) frequently affected by flash floods (Versini et al., 2010a). Coupling a susceptibility analysis of river road intersections (representing one part of the vulnerability to flooding) based on geographical information (Versini et al., 2010a, Versini et al., 2010b) distributed hydrological model, the RIWS has provided promising results. Tested on real cases, it was able to correctly assess the inundation risk with an acceptable level of accuracy. Nevertheless, this previous work has opened many ways of investigation before being applied in a decision support system. First, operational services interested by the RIWS has advised to study its possible application on a territory where it has not been calibrated to test the transferability of the whole prototype. Secondly, as the spatio-temporal distribution of rainfall has appeared to have a major influence on the state of the road network, the hydrological model had to be adapted to take into account distributed rainfall products, especially those based on weather radar. Indeed, one important feature of road submersion is the significant number of targets (that could be located on very small watersheds) regarding the limited coverage of rain and stream gauges, making this framework close to ungauged conditions. For example, the Gard region (580 km2) is covered by 38 stream gauges for 2480 crossing structures.
Accurate Quantitative Precipitation Estimates (QPE) are also crucial for operational flash flood forecasting. Ground-based operational weather radars currently appear as the only instrument able to provide valuable information with a high spatial (1 km2) and temporal (tens of minutes) resolution. The density of automated rain gauges network is generally too low and not adapted to flash flood short time and space resolutions. In this case rainfall estimation uncertainties are still a major factor limiting the accuracy of rainfall-runoff modelling (Moulin et al., 2009). Moreover, rainfall estimated using satellite remote sensing is still under development and not sufficiently advanced to be used in an operational mode. Consequently, radar QPE is accepted as one of the most reliable data that can be used for hydrological applications (Corral et al., 2000, Borga et al., 2006, Cole and Moore, 2008).
This is also the case concerning Quantitative Precipitation Forecasts (QPF). Although few works have focused on using QPF based on weather radar data, results show significant improvements in the quality of forecasted hydrographs (Corral et al., 2000, Dolciné, 2001, Berenguer et al., 2005, Borga et al., 2006, Boudevillain et al., 2006, Van Horne et al., 2006, Vivoni et al., 2006, Cole and Moore, 2008). These radar-based QPF are usually limited to forecasting time ranging from 10 to 120 min. Tested on rather large basins (from hundreds to thousands km2), the anticipation of flow peak could be estimated, with enough quality, with a lead-time for up to few hours. It represents a notable improvement for fast response basins such those in Mediterranean regions. It is also recognized that the nature of the event has an important effect on the quality of the forecasted flow estimates. In Collier (2007) a review is made to study how flash floods are forecasted considering the limitations and uncertainties involved in both meteorological and hydrological models of the forecasting system. The author concludes the possibility to deliver valuable information from a flash flood risk management point of view limited to a lead-time of 2 h.
This paper deals with a practical application of this statement. The main objective of the present work is to test the use of radar-based QPE and QPF on a specific hydrological application devoted to the road network. The spatio-temporal variability information provided by the radar based precipitation estimates and forecasts will be tested using the RIWS. The warning system will be firstly transported and adapted to a new basin located in the South part of the Gard region. It will be then applied to reproduce the specific storm of 29–30 September 2007 during which 19 roads were submerged. Predicted road inundations will be compared to what actually occurred. This will allow us to assess both the transfer of the RIWS on an area where it has not been calibrated, and the use of the available QFE and QPF products for flood forecasting in a framework reproducing operational conditions.
This paper is organised as follows: the next section presents the scope of study in more detail, including a description of the area of study, the rainfall products, and the RIWS. Section 3 describes the methodology applied to: (i) transfer and test the RIWS to the new domain, and (ii) test the information provided by the QPE/QPF, and their benefits in the detection and prediction of inundated roads. The results obtained during the 29–30 September 2007 storm are presented in Section 4. Finally, Section 5 will conclude on both topics presented in Section 3.
Section snippets
The Gard region
The Gard region (South of France) was used to develop and test the RIWS because it is frequently affected by severe flash floods (Delrieu et al., 2005, Gaume et al., 2009). This region has a typical Mediterranean climate characterized by frequent and very heavy storm events occurring especially in autumn. The 1 in 10 year daily precipitation exceeds 100 mm on the plateaus (eastern part) and 150 mm in the mountainous western part of the area (CNRS/INPG, 1997). Single storm events can produce
The case of the 29–30 September 2007
The rainfall event happened during the night between the 29 and 30 September 2007 and was one of the most important that occurred in the Vistre and Vidourle watersheds over the last years. A stationary storm with a “V” shape moved from the west to the east and intensive precipitations fell on the central part of the watersheds between 22:00 and 00:00. Between 80 and 120 mm were measured on both watersheds and locally more than 200 mm. At 2:30 the Vidourle river overflowed in Sommières. Rainfall
Radar-based QPE and QPF direct analysis
Rainfall accumulations during the entire storm have been computed in a window including both Vistre and Vidourle watersheds for every type of data (OHM-CV kriging, QPE and QPF from MF and SPC). They are presented in Fig. 2. Scatter plots comparing radar-based QPE with kriged data are illustrated on Fig. 3. Hyetograms representing mean areal intensity (with 15 min time step) at the watershed scale for both Vidourle and Vistre watersheds are also presented for every rainfall data in Fig. 4.
Conclusion
The road inundation warning system developed for flash flood prone areas, and recently calibrated on the North part of the Gard region, has been applied on the South part of the Gard. Working in a framework simulating operational conditions, the RIWS has been tested to predict the inundated roads during the 29–30 September 2007 event. These results obtained for this specific storm event are very similar to those obtained in the calibration area (Versini et al., 2010b). They are promising and
Acknowledgments
The study described in this paper has been carried out with the help of Meteo France and the Direction Départementale de l’Equipement du Gard which provided radar QPE and QPF data. Special thanks are due to the INPG of Grenoble and the OHM-CV (Cevennes-Vivarais Hydro-Meteorological Observatory) and especially Guy Delrieu, Laurent Bonnifait and Brice Boudevillain for providing historical meteorological data on the Gard region. The author thanks especially Eric Gaume from the Laboratoire Central
References (41)
- et al.
Hydrological modelling using raingauge and radar-based estimators of areal rainfall
J. Hydrol.
(2008) - et al.
A semi-distributed hydrological model using rainfall estimates by radar. Application to Mediterranean basins
Phys Chem Earth Part B
(2000) - et al.
Hydrologic analysis of the Aude, France, flash flood 12 and 13 November 1999
J. Hydrol.
(2004) - et al.
A compilation of data on European flash floods
J. Hydrol.
(2009) - et al.
The sensitivity of hydrological models to spatial rainfall patterns: an evaluation using observed data
J. Hydrol.
(1994) - et al.
Human exposure to flash floods – relation between parameters and human vulnerability during a storm of September 2002 in Southern France
J. Hydrol.
(2008) - et al.
Les crues meurtrières, du Rousillon aux Cévennes
Ann. Géogr.
(2001) - Ayral, P.-A., 2005. Contribution à la spatialisation du modèle opérationnel de prévision des crues éclair ALTHAÏR....
- et al.
Hydrological validation of a radar-based nowcasting technique
J. Hydrometeorol.
(2005) - Borga, M., 2008. Realtime Guidance for Flash Flood Risk...