Earth observation and case-based systems for flood risk management J.-B. HENRY, K. FELLAH, S. CLANDILLON, B. ALLENBACH, P. DE FRAIPONT Service Régional de Traitement d’Image et de Télédétection (SERTIT), Pôle API – Bd. S. Brandt, F-67400 ILLKIRCH [email protected] Tel: +33(0)390-244-644 – Fax: +33(0)390-244-646 Abstract – The case-based approach is presented through two of SERTIT’s major involvements in flood risk management. The first is operational crisis management action, contracted by ESA on the River Saône March 2001 flood. Archived and scheduled remote sensing data are used, leading to the delivery of rapid flood mapping products to the Civil Protection authorities. The second consists of research and development actions on well documented historical events. The case definition is based on the analysis of imagery products and environmental data, to both describe flood hazard and vulnerability. This approach stresses the synergy between landscape, its evolution and flood impact. Both optical and radar data are used in these two applications, with specified pre-processing levels. Index terms – Case-based approach, flood management, optical and radar remote sensing, decision making, rapid mapping systems.

INTRODUCTION Flooding often appears as the most devastating and frequent natural hazard. Flood events have a huge impact on human beings, settlements and ways of life. Even though their consequences always dramatically affect populations, their origins can vary from one region to another, and from season to season. Several space agencies (ESA, CSA, CNES, ISRO, NOAA)1 are currently working on crisis management support through imagery services and communication networks. Within this framework, several projects have been performed recently and many partnerships were established between these agencies, industrial companies, value-adder firms and National Civil Protection Services. For example, the International Charter on Space and Major Disasters has permitted many interventions during various events (http://www.disasterscharter.org). Through these actions, space agencies aim to prove the efficiency of remotely sensed information for crisis management purposes. This paper treats two case studies of Earth Observation (EO) data integration into geo-information systems (GIS) as an aid in the different phases of risk management. The first is 1 European Space Agency, Canadian Space Agency, Centre National d’Etudes Spatiales, Indian Space Research Organisation, National Oceanic and Atmospheric Administration.

SERTIT’s involvement in crisis management over the River Saône, for the March 2001 flood. Then, the case-based approach is explained, using the results from the crisis management phase. EARTH OBSERVATION DATA AND RISK M ANAGEMENT Over many years, researchers have concentrated efforts in using remote sensing data in the different phases of risk management. Historically, the difficulties of scheduling satellite acquisitions over determined areas, the delay between acquisition and delivery to value-adder firms, the processing delay after reception limited remote sensing techniques to prevention and modeling purposes. Nowadays, the increasing number of EO missions, progress in telecommunications, image processing, improved technical and scientific know-how, and finally, better user awareness permit their use in different phases of risk management. Prevention, prevision, mitigation EO datasets have been widely exploited in prevention studies as a cartographic tool, essentially for land use description. Many examples are given in the literature covering regional to local scales studies, depending on sensor resolution and swath. Earth observation systems provide adequate spatial and spectral resolutions, even if revisit frequency is lower [1]. On the other hand, prevision also needs information over large areas to improve its predictions and requires much auxiliary information, that can issue from field measurement for the parameterization, calibration and validation of empirical models. Studies have developed simulation models that can account for various remotely sensed parameters such as terrain morphology descriptors [2], soil nature and related hydraulic properties [3], and land use to define boundary conditions for models [4] by mapping vegetated areas which contribute to the evapo-transpiration process. Soil moisture is difficult to quantify and monitor, because of its spatial and temporal variability, even if much progress has been made in microwave remote sensing [5], [6], [7], and through campaigns such as AGRISCAT, NOPEX, Washita, HAPEXSahel.

Crisis management

Cartographic outputs

Studies have stressed the usefulness of different types of remote sensing and cartographic data [8], [9]. Generally, information extraction from satellite datasets in a crisis situation employs change detection techniques, requiring an image of the event and an archive image. Crisis management requires high temporal frequency data, to manage the event’s dynamics, at a good spatial resolution. Current EO systems resolutions still appear to be too coarse for an accurate operational use, even if they give useful results [10]. The main difference with the preceding phases is that speed is of the essence. All retrieved data and mapping products are reusable for mitigation and prevention purposes [11].

One of the major goals of the International Charter on Space and Major Disasters is to furnish EO derived cartographic products to operational services. Therefore, time delay is the key factor of this kind of action. These rapid flood mapping products are delivered within 6 hours. Thus, final products are e-mailed to operational users to speed up, once again, the information availability. Cartographic web servers would be the next enhancement to provide dynamic broadcasting and customizable map products.

CRISIS M ANAGEMENT A PPLICATION In March 2001 the River Saône which runs south across a wide plain, towards the river Rhône, burst its banks. A heavy rainfall event occurred in mid-March, resulting in significant widespread flooding, starting on the 19th , from Mâcon to Villefranche-sur-Saône, North of Lyon France’s second biggest city. Data transmission and pre-processing The French Civil Protection Authorities activated the International Charter on Space and Major Disasters on March the 27th 2001, coordinated by ESA. Archive ERS and SPOT images are sent to SERTIT, while other satellite acquisitions are programmed. During the delay between satellite activation and crisis data reception, the archive images are prepared in order to facilitate the flood plain mapping. All raw data transfers are realized using a high bandwidth FTP protocol to retrieve PRI ERS-2 images (approx. 10 min per image). Geometric and radiometric corrections are then performed in order to produce interpretable mosaics. All data are referenced to the national cartographic system. Thus, geometric coherence is obtained and GIS integration for map production is facilitated. Information extraction Data dependant algorithms are used to extract flood extents. Difference or ratio algebra between an archive image and the flood image, combined with an appropriate threshold can be applied both on optical and radar datasets. The global process appears to be more efficient when both images are acquired under similar conditions, i.e. resolution, incidence angle, season, imaging mode…

CASE -BASED A PPROACH The case-based approach aims to propose a synthetic vision of a flood event. Combining flood extent maps, meteorological and gauge data, a case is defined by the synergistic analysis of each component and its contribution to the event’s intensity. Thus, the database consists of a GIS fed with geographical and environmental data layers. Geographical dataset This layer is constituted of two main information layers. Firstly, it is important to obtain a good description of the landscape. This is possible by the numerical and visual interpretation of optical data, creating land use maps. These maps are elaborated with a precise nomenclature, which is highly detailed in urban areas. A feature data fusion performed on actualized and historical classification products allows a study of land use dynamics, and especially urban dynamics. This process can emphasize urbanization trends, in or close to a frequently inundated area. Secondly, analysis performed on images acquired during flood events permit an instantaneous flood extent cartography, using the same extraction methods as in a crisis situation. When successive images are available, their intersection leads to a flood dynamic cartography, permitting flood duration and frequency mapping [8]. These two geographical components can be easily combined to precisely assess the impact of an event on urban areas. It is thus possible to evaluate local land planning policies and how they account for a known hazard. Environmental dataset Contrary to geographical dataset, this second layer is comprised of variables measured at point sources, i.e. precipitation, water levels and regional water balance. Time series of rainfall and hydrometric measurements are available for each event. A quick analysis shows that a standard integration period, of 10 days in our study, enables the efficient description of an event, its intensity, and thus the elaboration of a synthetic overview. The water balance’s

utility is to provide a global appreciation of the hydrologic state of the catchment, assuming that the same rainfall would not have the same consequences on hydrometry depending on soil moisture. In practice, the composition of the environmental dataset involves the building of a geo-information attributes table locating measurement stations. For each variable (precipitation, river flow…), the table is elaborated with raw measurements and time-integrated values.

A CKNOWLEDGMENTS This work is presently being pursued within the PACTES program (Prévention et Anticipation des Crues au moyen des Techniques Spatiales) which is financed by the French Ministry of Research and coordinated by the CNES (Centre National d’Etudes Spatiales). REFERENCES [1]

Case description The case description is the final step of the database’s development. It uses its different layers and puts in light, as far as possible, the correlation between a given rainfall, the flood extent and observed river flow. It is founded on the simple assumption of cause and effect. Hence with equivalent conditions, the same causes will produce the same effects. Even if reductive, this hypothesis allows the analysis of an occurring event by a quick comparison with historical cases. Finally, the main idea is to give operational users elements with which to qualify a majority of the events they have to face. In this way, the knowledge of each component of observed events is important and useful. This constitutes the loop back, from the mitigation to the prevention process and is susceptible to be used by many user communities: hydrologists, environment analysts as well as local administrations and insurance companies. DISCUSSION As the intensity and frequency of flood events cannot be fully controlled, solutions must be found to limit their impacts on human beings. The development of crisis action frameworks using advanced space technologies is a first step, which needs to be completed by further studies to improve the knowledge of the natural processes involved. Thus, the case-based approach aims to synthesize all information acquired during an event. Through this simple assumption of cause and effect, a quick quantification of any flood situation might be enabled. Moreover, the use of EO data appears to be taking hold in flood risk management phases. The production of EO derived cartographic data on natural hazards, over large areas, is widely appreciated by scientists, crisis management services, land planners and insurance companies. Future developments aim to improve the synergistic use of remotely sensed information with other geographical ancillary data. This would trend towards EO systems being completely integrated into flood hazard management chains.

M.F. Baumgartner and G.M. Apfl, Remote sensing and geographic information systems, Hydrol. Sc. J., Vol. 41 (4), pp. 593-607, 1996. [2] K.J. Beven and M.J. Kirkby, A physically -based variable contributing area model of basin hydrology, Hydrol. Sc. Bull., Vol. 24, pp. 27-53, 1979. [3] B. Müschen, W.-A. Flügel, V. Hochschild, K. Steinnocher, P. De Fraipont, K. Fellah, G. Rodolfi, A. Colpaert, K. Krzywinski and F. Quiel, Parameterization of Hydrological, Erosion and Solute Transport Models by Remote Sensing Techniques and GIS Analyses in European Catchments, Proc. of the 28th Int. Symp. on Remote Sensing of Environment, Capetown 27-31 March, 4p., 2000 [4] P. D. Bates, M. S. Horritt, C. N. Smith and D. Mason, Integrating remote sensing observations of flood hydrology and hydraulic modeling, Hydrol. Proc., Vol. 11 (14), pp. 1777-1795, 1997. [5] P.C. Dubois, J. Van Zyl and T Engman, Measuring Soil moisture with Imaging Radars, IEEE Trans. Geosci. Rem ote Sensing, Vol. 33 (4), pp. 915-926, 1995. [6] E.T. Engman and N. Chauhan, Status of Microwave Soil Moisture Measurements with Remote Sensing, Remote Sensing Envi., Vol. 51, pp. 189-198, 1995. [7] K. Fellah, P. Bally, Y. Besnus, C. Meyer, M. Rast and P. de Fraipont, Impact of SAR radiometric accuracy in hydrological and agroenvironmental applications, Proc. of Retrieval of bio- and geophysical parameters from SAR data for land applications, Toulouse 17-20 October, pp. 337-346, 1995. [8] N. Tholey, S. Clandillon and P. De Fraipont, The contribution of spaceborne SAR and optical data in monitoring flood events: examples in northern and southern France, Hydrol. Proc., Vol. 11 (14), pp. 1409-1413, 1997. [9] G. Profeti and H. Macintosh, Flood management through LANDSAT TM and ERS SAR data: a case study, Hydrol. Proc., Vol. 11 (10), pp. 1397-1408, 1997. [10] R. Oberstadler, H. Hönsch and D. Huth, Assessment of the mapping capabilities of ERS-1 SAR data for flood mapping: a case study in Germany, Hydrol. Proc., Vol. 11 (10), pp. 1415-1425, 1997. [11] H. Yésou, P. Chastanet, P. de Fraipont, P. Dossmann, N. Stock, J. Béquignon, Mapping floods in France, Backscatter, Vol. 12 (3), pp. 23-26, 2001.