Use of Temporal Series of COSMO-SkyMed Data for Flood Monitoring
Pierdicca, Nazzareno1; Pulvirenti, Luca2; Chini, Marco3; Boni, Giorgio2; Candela, Laura4
1Sapienza University of Rome, ITALY; 2CIMA Research Foundation, ITALY; 3CRP-Gabriel Lippmann, LUXEMBOURG; 4Italian Space Agency, ITALY
The potential of spaceborne Synthetic Aperture Radar (SAR) for flood mapping was demonstrated by several past investigations. The synoptic view, the capability to operate in almost all-weather conditions and during both day time and night time and the sensitivity of the microwave band to water are the key features that make SAR data useful for monitoring inundation events. Moreover, their high spatial resolution allows emergency managers to use very detailed flood maps. Among SAR instruments, COSMO-SkyMed (CSK) gives the possibility of performing frequent observations of regions hit by floods, thanks to the four-satellite constellation that offers a very short revisit time, thus allowing end users to monitor the various stages of the temporal evolution of a flood. This work analyses the potential of multitemporal series of CSK data for flood monitoring, presenting different case studies based on data acquired within the framework of the OPERA project, funded by the Italian Space Agency. This project, which started in 2007 and lasted four years, aimed at evaluating the improvements that may be obtained through the use of Earth Observation satellite data in the hydro-meteorological and hydraulic prediction chains. A pre-operational system has been implemented, which covers all the various phases of the flood risk management, from the planning and preliminary vulnerability assessment to the prediction, the emergency management and the post-event damage assessment. Using an ad-hoc data and models, a distributed infrastructure and a variety of satellites and sensors, it was demonstrated the usefulness of feeding the system with a large number of Earth Observation (EO) products from a central EO Competence Centre in order to manage central and peripheral Functional Centres involved in civil protection tasks. In this framework, a fast delivery of accurate flood maps was demonstrated to be a fundamental step and one of the most successful contribution of EO to this application.
The availability of a multitemporal series of SAR images does not always enable to follow the time evolution of the disastrous event, but also contributes to improve the quality of the flood maps by resorting to proper multitemporal data interpretation tools. In fact, the detection of the flooded areas is affected by many sources of error, both missed detections and false alarms. Most of the automatic flood detection algorithms available in the literature search for regions of low backscatter. This approach is based on the low radar return from smooth open water bodies that behave as specular reflectors. However, flooded areas do not always appear dark, so that flood extension can be underestimated. Namely, soils do not always act as smooth open water bodies, mainly because of the presence of wind or of vegetation emerging above the water surface. As for vegetation, the double-bounce mechanism involving soil and stems or trunks is generally enhanced by the floodwater, so that flooded vegetation may appear very bright in a SAR image. Even overestimation (i.e., false alarms) may occur because of complex topography producing radar shadow, wet snow being a strong absorber, as well as of heavy precipitating clouds, which may attenuate the radar echoes at higher frequencies (including X band).
CSK is particularly suitable for dealing with these ambiguous situations, which can cause inaccuracies in flood boundary delineation. In fact, the reliability of flood maps can be potentially improved by using multitemporal datasets, such as those available through CSK. In this paper, the above discussed problems and an overview of the semi-automatic image interpretation approaches developed to properly solve them are presented making reference to different case studies. In particular, the CSK images of the floods occurred in Tuscany (Central Italy) on December 2009 and in Northern Italy, close to the city of Alessandria, on April, 2009 turned out to be particularly useful to assess the added value of temporal series of data for flood mapping purposes, especially to correctly detect the water under vegetation. The dataset collected in Liguria (in Northern Italy) during a storm, in November 2011, made it necessary to properly account for the effect of very thick clouds and wet snow. The CSK acquisitions after the Japanese Tsunami in March 2011 pointed out the need to properly interpret the changes in the images that are not always related to the presence of water, including floating of debris. Our methodology to exploit a multitemporal set of radar data, basically consists of three steps: 1) performing the segmentation of the multitemporal series of CSK observations; 2) computing the average backscattering coefficient of each segment; 3) applying to the segmented images the fuzzy-logic-based approach originally developed in . Appropriate fuzzy rules to interpret the multitemporal signatures have been introduced to improve the quality of the flood maps. A certain degree of operator supervision to produce an accurate flood map is considered in any case very important.
 L. Pulvirenti, N. Pierdicca, M. Chini, and L. Guerriero, ''An algorithm for operational flood mapping from Synthetic Aperture Radar (SAR) data based on the fuzzy logic''. Natural Hazards and Earth System Sciences, vol. 11, pp. 529-540, 2011.