Flood Damage Assessment Through Multitemporal COSMO-SkyMed Data and Hydrodynamic Models: The Albania 2010 Case Study

Share Embed

Descrição do Produto



Flood Damage Assessment Through Multitemporal COSMO-SkyMed Data and Hydrodynamic Models: The Albania 2010 Case Study Luca Pulvirenti, Member, IEEE, Nazzareno Pierdicca, Senior Member, IEEE, Giorgio Boni, Mattia Fiorini, and Roberto Rudari

Abstract—Flood damage assessment needs not only the estimation of the flood extent but also the information on the drainage of the floodplain and the dynamics of variables as water depth and velocity. These data might be gathered by exploiting numerical models of water propagation in floodplains, which enable to build flood scenarios in real time if reliable digital elevation models are available. However, a strong limitation for the application of numerical models could be the lack of information regarding the actual flood extent and the dynamics of flooding and receding phases as well as the locations, where water overflowed and the related flood volumes. Inundation extent can be estimated through synthetic aperture radar (SAR) data and, by exploiting the short revisit time of the images provided by the COSMO-SkyMed (CSK) constellation of four satellites, it is possible to monitor also the dynamics of the flood extent. Hence, it comes out the need of a combined use of multitemporal SAR data and numerical models for the purpose of a reliable flood damage assessment. This paper presents the major outcomes of a combined use of a multitemporal series of CSK observations and a hydrodynamic model aiming at the evaluation of damage scenarios for the flood that hit Albania in January 2010. It is shown that by adjusting the outputs of the model to match the flood extent observed by SAR, the hydrodynamic inconsistencies in CSK estimates can be corrected and a reliable assessment of water depth and water velocities can be accomplished. Index Terms—COSMO-SkyMed (CSK), floods, hydrodynamic model, SAR.



LOODS can be considered as the major natural disasters in terms of social and economic costs. Their frequency is

Manuscript received December 31, 2013; revised March 13, 2014; accepted May 20, 2014. Date of publication July 07, 2014; date of current version August 21, 2014. This work was supported by the Italian Space Agency (ASI) under contract no. I/048/07/0. L. Pulvirenti is with CIMA Research Foundation, 17100 Savona, Italy, and also with the Department of Information Engineering, Electronics, and Telecommunications, Sapienza University of Rome, 00184 Rome, Italy (e-mail: [email protected]). N. Pierdicca is with the Department of Information Engineering, Electronics, and Telecommunications, Sapienza University of Rome, 00184 Rome, Italy (e-mail: [email protected]). G. Boni is with CIMA Research Foundation, 17100 Savona, Italy, and also with the Dipartimento di Informatica, Bioingegneria, Robotica e Ingegneria dei Sistemi, University of Genoa, 16145 Genoa, Italy (e-mail: [email protected] cimafoundation.org). M. Fiorini and R. Rudari are with CIMA Research Foundation, 17100 Savona, Italy (e-mail: mattia.fi[email protected]; [email protected] cimafoundation.org). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSTARS.2014.2328012

presently increasing because of the occurrence of several extreme weather events all over the world and the damage bill is considerably growing. A prompt evaluation of the damages caused by floods during the response and early recovery phase is fundamental for relief efforts and rescue organization. Flood damage assessment requires primarily the availability of maps of the flood extent and then, possibly, other information such estimates of water depth and velocity. Information on the flood extent and the water depth are difficult to be gathered in the emergency and recovery phases through field surveys during the flood and/or a posteriori estimates of the inundation extent based on the traces of the water flow. In fact, floodwater can be fast moving as in flash floods; moreover, carrying out extensive field surveys could be unfeasible or even unsafe [1]. Aerial photography might also be used as a source of data on inundation extent, but airborne campaigns are generally very expensive and often unaffordable. In principle, a complete set of data useful for flood damage assessment, such as physically based maps of flood extent, water elevation and velocity, and flow directions can be provided by two-dimensional (2-D) numerical models of flood propagation. Although in the recent years significant improvements were attained in flood inundation modeling [2], model simulations can be affected by errors, such as those generated by the poor representation of the floodplain provided by the digital elevation models (DEMs) publicly available. The shuttle radar topography mission (SRTM) and the advanced spaceborne thermal emission and reflection radiometer (ASTER) DEMs are widely used in large scale applications. Although they can be, especially in data scarce environments, the only source of information available during emergencies, they often turn out to be inconsistent from the hydraulic point of view for local scale applications [3] (for instance because of a too low resolution). This lack of consistency may cause inaccuracies in the predicted water depth that on their turn can give rise to errors in the estimated inundation extent. Information on the flood extent can be provided by satellite remote sensing data (e.g., [4]); inundation maps are presently produced in near-real time (e.g., [5], [6]) and represent fundamental products for international or national agencies for disaster monitoring and relief efforts in the early phases of disaster risk reduction (DRR) [6]. In the last years, new tools and portals, which take advantage of Internet/World Wide Web technologies (e.g., [7], [8]), have been set up to quickly deliver remote sensingderived flood maps to end users and decision makers [9].

1939-1404 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.


Presently, the most effective approach for remotely sensing flood extent from space is the use of synthetic aperture radar (SAR) images. Working in the microwave range of the electromagnetic spectrum (1–10 GHz), SAR is sensitive to water and is also capable to collect data independent of time of day and to observe the Earth surface even in cloudy and rainy conditions (except for very intense precipitations, which can affect higher frequency sensors [10], [11]). Moreover, SAR offers a high spatial resolution, which, with the new generation of X-band instruments such as TerraSAR-X and COSMO-SkyMed (CSK), can reach 1 m. Even the problem of the revisit time of SAR data (critical for their operational use for flood monitoring) is now overcome thanks to the present (CSK) and forthcoming (Sentinel-1) constellations of satellites. It was demonstrated that through multitemporal CSK data, it is possible to monitor the dynamics of the flood evolution [12]. However, mapping flooded vegetation and urban areas can be still considered as a critical issue for operational inundation mapping using SAR data (see Section III-A), although new insights on these topics were recently provided [6], [12]–[15]. From the previous discussion, it emerges that both the use of SAR data and the use of physical models for flood damage assessment have strengths and weaknesses. The latter can be, at least partially, mitigated through a joint use of the two approaches. Identifying the best fit between flood extent predictions of hydrodynamic models and (multitemporal) SARderived inundation maps makes it possible to verify whether the near real-time description of the inundation dynamics (e.g., recession of the water) obtained through SAR is physically consistent. On the other hand, physical models allow for the prediction of the temporal evolution of a flood event, i.e., for the construction of a risk scenario [3] useful for emergency managers to take their decisions; multitemporal SAR observations of the area of interest can be programmed in order to verify the reliability of the scenario and, in case, to modify it. In this context, it is worth underlining that the CSK constellation has a response time (i.e., the time between the user request and the product availability) that is less than 72 h, whereas the revisit time (period between two successive acquisitions over the same target) is less than 12 h [16]. Note that the aforementioned values are referred to the nominal operative mode (routine mode) and that they could be even shortened in the crisis and very urgent CSK operative modes (response time less than 36 or 18 h, respectively [16]). The cross-fertilization between (multitemporal) SAR data and physical models can consist of constraining the model outputs to match radar observations in order to obtain reliable estimates of the model parameters, thus coping with the aforementioned problem of the accuracy of the representation of the floodplain and with the lack of dense ground based observations (unavailable in most of the flood risk prone areas of the world). Moreover, model-derived flood extent maps can in principle fill possible gaps that might be present in the SAR-derived maps because of the limited area imaged by SAR (for CSK, the swath extension in the standard Stripmap image mode is in the order of ), or because radar observation of floodwater is hampered by layover or shadowing, or by the presence of very dense vegetation.


Despite of the strict connection that can be established between SAR remote sensing of floods and inundation modeling, the advances achieved in the recent years in both fields were integrated only to some degree [17]. In general, the use of SAR images of floods and physical models was limited to one or at the most two [18] radar images and to water level estimation, often performed by exploiting the availability of a high spatial resolution DEM (in the order of 2 m, as in [18] and [19]). This availability cannot be generally ensured in most of the flood prone geographical regions, where only low-resolution DEMs (e.g., SRTM) can be used, as previously underlined. Recently, the Italian Space Agency (ASI) funded a project in which the skills of experts in both remote sensing and hydrology have been combined. This project named OPERA (OPerational Earth observation based RAinfall-runoff forecast [20]: www. opoeraproject.it) aimed at assessing the added value of remote sensing data on a complete operational flood management system at national scale, which includes the different phases of the DRR cycle (mitigation, preparedness, response, recovery). Throughout the period 2009–2011, the OPERA team, formed by several research institutes in Italy coordinated by CIMA Research Foundation, was activated by the end user, i.e., the Italian Department of Civil Protection (DPC), owing to the occurrence of inundations not only in Italy but also in other countries such as Pakistan and Albania (see [21]), where DPC was involved in the relief efforts. OPERA paid specific attention to the data provided by the CSK constellation, because of its aforementioned capability to offer short response and revisit times that makes it particularly useful for operational flood monitoring. In this paper, the major outcomes of an activity, carried out within the framework of the OPERA project, aiming at interpreting the dynamics of the flood that hit the region of Shkodër (Albania) in January 2010 are discussed; the activity was carried out as part of the preoperational demonstration phase of the project. Its results were used during the emergency phase of the event, by DPC, in the framework of the Italian–Albanian international cooperation, for assessing the dynamics of the drying of the water in order to coordinate the response during the emergency phase. This case study concerns specifically the near real-time use of satellite images in the DRR cycle (reponse phase), which is commonly undertaken by operational services like the International Charter “Space and Major Disasters,” or the Copernicus Emergency Management Service. It consists of monitoring the early post-event emergency phase, following the acquisition of satellite images, which occurs in almost all cases when the event is already in its paroxysmal phase. We focus on the utility of a dense temporal series of SAR data, such as those available through CSK for this case study, used in combination with a hydrodynamic model to monitor over a long time (in the order of 3 weeks) the natural drainage of the floodplain. The methodology set out in this paper can also be extended to the flood rising phase without substantial conceptual differences, if SAR observations are programmed in advance owing to a flood risk forecast. The paper is organized as follows. Section II is dedicated to a short description of the Shkodër flood and the available CSK data. In the first part of Section III, a brief review of the effects of








RA, right ascending; RD, right descending.

Fig. 1. Area hit by the Shkodër flood. Left panel: CSK image acquired on January 10, 2010 overlaid to a Google Earth image. Right panel: SRTM DEM of the area. Lower panel: zoom on the area highlighted by the yellow ellipse. The scale bars represent the backscattering coefficient [left panel, (dB)] and the height [right panel, (m)].

the presence of floodwater on SAR images is provided; then, the key features of the algorithm developed within the framework of the OPERA project to map floods using radar data are summarized. In the last part of Section III, the characteristics of the hydrodynamic model and its calibration are synthetically described. Section IV discusses the results and Section V draws some concluding remarks. II. ALBANIA 2010 CASE STUDY AND THE AVAILABLE DATASET The Albania flood, which began on January 4, 2010, was basically caused by the persistent rainfall in the basin of the Drin River between late December 2009 and early January 2010. In addition, high temperatures in mountainous regions gave rise to snow melting during December 2009. The subsequent high flows in the Drin River worried managers of the three dams built on the river; they considered the conditions of the three related reservoirs (Fierzë, Koman, and Vau i Dejës) as critical and decided to partially drain these reservoirs in order to guard against further precipitation. This caused the increase in the discharge downstream the three artificial lakes up to approximately , where the maximum river capacity does not exceed [3]. It resulted in the overflow of the Buna River, which flows in the floodplain downstream the Shkodër Lake and receives the waters of the Drin (see the left panel of Fig. 1). To get the order of magnitude of the exceptional nature of the event, it has to be considered that, during December 2009, the average recorded rainfall depth in the Drin catchment was approximately 200 mm/month, more than 200% of the average accumulation for the same period. According to international flood relief organizations, the extension of the inundation in the floodplain downstream of the Shkodër Lake extended for about 10 500 ha. Approximately 2500 houses were flooded, and approximately 6000 people were evacuated, starting on January 5, 2010; some areas remained flooded until the end of January 2010 [3].

Within the framework of the OPERA project, being the Italian DPC involved in the relief efforts, ASI provided the OPERA team with a set of CSK observations of the flood (Single-look Complex Slant products: SLC), performed in Stripmap mode and horizontal polarization; their characteristics are summarized in Table I. In addition, an archived image of the same area acquired on July 7, 2009, in ScanSAR mode, horizontal polarization, right descending orbit, and at an observation angle of 22 was made available by ASI to be used as benchmark. The SLC Stripmap images were multi-looked (two looks in both range and azimuth directions) calibrated and geocoded by means of the ENVI/ SARSCAPE software to derive the backscattering coefficient ; for this purpose, the SRTM DEM was used. The pixel size of the geocoded images was set to ; the ScanSAR image was oversampled to the same pixel size. Fig. 1 shows the area hit by the Shkodër flood; the left panel shows the CSK image gathered on January 10, 2010 overlaid to a Google Earth image, whereas the SRTM DEM of the area is displayed in the right panel. The large dark region clearly visible in the left panel basically corresponds to the inundated area; looking at the right panel, it can be seen that the flooded area is substantially flat. In fact, within the aforementioned dark region, the mean and the standard deviation of the DEM’s height are in the order of 3 and 10 m, respectively. The reason for the presence of the ellipses and of the white arrows (right panel) in Fig. 1 (as well as for the zoom on the area highlighted by the yellow ellipse) will be explained in Section IV. III. FLOOD MAPPING ALGORITHM AND THE HYDRODYNAMIC MODEL The algorithm and the models used to derive a range of products useful to build an event scenario for the Shkodër flood and to monitor the temporal dynamic of the drainage of the floodplain are described hereafter. Before summarizing the key features of the flood mapping algorithm, it is worthwhile to briefly resume the main physical bases of the SAR remote sensing of inundations. A. Radar Signatures of Floodwater The main electromagnetic mechanism producing a high contrast in SAR images between flooded and nonflooded areas is the specular reflection of the radar signal (e.g., [21]–[23]). In general,


floodwater covering the terrain is much smoother than the surrounding nonflooded land and reflects most of the impinging radar signal toward the specular direction, so that the backscatter is very low and flooded areas appear dark in tone. Inundation mapping may be complicated by wind roughening and by vegetation emerging from water, both producing high radar returns [23] that decrease the contrast between flooded and nonflooded areas in SAR images. However, it could happen that inundated vegetation appears brighter than the surrounding soil. In fact, a typical radar signature of flooded vegetation is determined by the increase in that may occur because of the double bounce mechanism that involves floodwater and vertical structures such as stems or trunks that act as corner reflectors. The increase in due to the double bounce is difficult to predict because it depends on the structure, geometry, and density of vegetation, as well as on sensor parameters (incidence angle and polarization) and water height [13]. It is generally larger at low incidence angle, because of the higher penetration through the vegetation canopy, and at horizontal polarization that presents a greater surface reflection coefficient than vertical one. To detect the possible increase in , it is in any case required the availability of a preflood image to be used as benchmark within a change detection approach. In flooded urban areas, both specular and corner reflections may occur, but layover and shadowing effects strongly complicate inundation mapping [6], [14]. A thorough review of the radar microwave signatures of floods has been recently provided by Pierdicca et al. [21]. B. Flood Mapping Algorithm The flood mapping procedure designed in the framework of the OPERA project is basically a classification algorithm founded on the fuzzy logic in which the integration of different decision rules is performed. These rules are based not only on SAR data but also on ancillary data such as a thematic land cover map and a DEM. Hereafter, we briefly summarize the basis of the algorithm, whose schematic flow diagram is shown in Fig. 2, while we refer the reader to [22] and [24] for a more detailed description. While in classical set theory, an element either belongs or does not belong to a set, elements of a fuzzy set have degrees of membership (MDs) that are defined through membership functions (MFs) whose values are real numbers between 0 (no membership) and 1 (maximum membership). The parameters of the MFs are a pair of thresholds (Th1 and Th2) pertaining to the value of the backscatter. For instance, if regions of low backscatter are searched for (see Section III-A), Th1 is such that if the MD to set of flooded areas is 1, whereas if the degree is 0. values within the interval between Th1 and Th2 have an MD to the set of flooded areas that depends on its position in the interval itself. If a preflood image is available, the fuzzy algorithm searches also for areas in which floodwater causes an increase in the backscatter due to the double bounce; this is done only for vegetated agricultural areas and forested regions, so that a land cover map distinguishing at least these two cover classes is required in this case. By denoting as the difference between the backscattering coefficients measured in the flood and in the preflood observations, if the MD to set of flooded


Fig. 2. Schematic flow diagram of the flood mapping algorithm. Yellow indicates input data; orange indicates output data.

the degree is 1. Note that areas is 0, whereas if is usually expressed in dB units (i.e., the log-ratio) to account for the multiplicative nature of speckle noise. The values of Th1 and Th2 are provided to the algorithm, for standard land cover classes (bare soil, agricultural land, forests), in the form of look-up tables (LUTs); the LUTs are determined according to the expected values of , in both flooded and nonflooded conditions, predicted by some well-established electromagnetic models [25]–[27] able to simulate the radar measurements in different environmental conditions and for different SAR configurations (frequency, polarization, incidence angle). Note that, if a preflood image is available, the change detection approach is preferably applied to images collected with the same looking geometry, as interferometric pairs; otherwise the difference between the incidence angles must be taken into account as done for instance in [13], where the variation of with the incidence angle was approximated by , in agreement with [28]. Interferometric pairs can also be useful to deal with possible inaccuracies of the land cover maps. For instance, increases in can be searched in areas that are labeled as vegetated by the land cover map and in which the complex coherence is small; the combination of these two conditions can ensure that the areas are actually covered by vegetation. In any case, the algorithm requires that the thematic map is able to distinguish very broad classes (those listed above), so that it can be expected that these classes are discriminated with fairly good accuracy by commonly used land cover maps (e.g., CORINE). The fuzzy rules based on SAR data (and land cover) are complemented by other rules that consider also the information derived from a DEM, such as elevation, slope (flood prone areas are generally flat), and local incidence angle (to avoid false alarms due to shadowing, as done in [11]). Then, the map of flooded areas is finally produced by applying the so-called “defuzzification” process (see Fig. 2) that assigns each element of a fuzzy set to a class (flooded or nonflooded pixels, in this case), according to the final MD resulting from the previous combination of rules; we simply accomplish a threshold defuzzification, i.e., a pixel is labeled as flooded if the final MD is greater than a threshold often set to the intermediate value of 0.5.



Although the procedure described above was initially developed within a pixel-based approach [22], [24], it was successively applied within a segment-based approach [12] in order to increase the spatial homogeneity of the derived flood maps. For this purpose, a preliminary segmentation of the SAR image through techniques such as mathematical morphology (as done in [13]) ore region growing (as done in [11]) is preliminary applied in order to divide the image into separate objects that are homogeneous with respect to properties such as pixel intensity and texture, and are strictly related to the real targets present in the scene. For each object, the average backscatter is computed and used in the successive processing. C. Hydrodynamic Model and Its Calibration The hydrodynamic model, already proposed in [29], is based on the numerical solution of a simplified subset of equations derived from the Shallow Water formulation of Navier–Stokes equations [29]–[31]. The numerical solution was developed for the specific needs of real-time flooding simulations, i.e., for the purpose of obtaining a high computational efficiency through the use of a simplified set of initial conditions (ICs) and boundary conditions (BCs). The model, basically representing the wave propagation (it can be considered a diffusion wave model [32]), is expressed by [29]


and ( ) are the horizontal velocity components, is the water depth, is the ground elevation, is the Chézy friction coefficient ( ), and are the horizontal coordinates, and indicates time. For the empirical expression of Gauckler–Strickler can be used [29], where is the Strickler roughness coefficient ( ) that depends mostly on vegetation cover and surface type and can be derived from land cover maps. Under the assumptions of complete turbulence and infinitely large flow section, the horizontal velocity components can be computed as

∇ ∇ . where Equations (2) are solved numerically using a storage-cell based method (see Fig. 3 that shows the flow scheme) whose grid is derived from the DEM that provides the elevation information . Since the scope of the model is, in this case, the generation of real-time civil protection scenarios, it was reformulated to use a simplified set of ICs and BCs and to be constrained by SAR observations (as described at the end of this section). In particular, the BCs are represented by the flooding sources (overflow points), i.e., by the location of the pixels where water enters in the model domain, by the volumes of

Fig. 3. Scheme used in the numerical solution of the hydrodynamic model for cell , .

water entering the system and the position of sinks in the flood domain if any. ICs are represented by the water level in the floodplain and in the flooding sources (representing conceptually the level over the levees). For instance, the ICs for the initial phase of the flood are described by non-null water level in the flooding sources only. From the flooding points, the water volume is propagated in the neighboring cells of the model domain by computing the discharge ( ) through cell borders. The latter can be evaluated as

where and define the cell size (see Fig. 3). Discharges are also used to compute new water levels at each cell for the next time step ( )

where and denote the vertical and horizontal coordinates of a cell, respectively (see Fig. 3). Since, as discussed in the introduction, the exact location of the overflow points is generally unknown and the same applies for the related volumes, the model was conceived to derive reliable BCs and ICs from SAR observations. To this aim, a set of model runs corresponding to a set of ICs and BCs was carried out; the run producing the output that best matched, in terms of flood extent, the SAR observation was then selected. This processing required an amount of supervision by a flood model expert. IV. RESULTS AND DISCUSSION The Albania flood case study was selected to test the integration of remote-sensing-derived maps of flood extent with physical models aiming at performing a flood damage assessment


Fig. 4. Left panels: color coded map of water depths predicted by the hydrodynamic model. Right panels: comparison between the simulations of hydrodynamic model and the flood extent detected by CSK (coded in magenta); the black dashed lines represent the limit of the CSK swath. Upper panels: January 10, 2010; lower panels: January 16, 2010.

basically consisting of producing physically consistent multitemporal maps of the flooded areas and of water depths and velocities. The algorithm described in Section III-B was applied to all the images of the series of CSK data (Table I) in order to monitor the dynamics of the flood, and in particular the water receding; the CORINE land cover map was used as ancillary data. As the first SAR-derived inundation map was available (that corresponding to the CSK acquisition performed on January 10, 2010), the hydrodynamic model was run for the purpose of accomplishing a flood damage assessment and designing a future scenario. The model used the SRTM DEM, so that its spatial resolution was set to 90 m (i.e., the DEM resolution) and its domain basically corresponds to that shown in the right panel of Fig. 1. The definition of the consistent set of BCs and ICs based on SAR data was carried out as described in the previous section; for the sake of concision, the simulation results obtained for the period between January 10 and January 16, 2010 are discussed hereafter because the most significant results in terms of model outputs needed for damage assessment (e.g., maximum water depths) were obtained in that period. As an example of the model results, the left panels of Fig. 4 show the water depths obtained for January 10 (the day in which CSK observed the maximum inundation extent) and January 16, 2010. It can be seen that the water depth presents a large variability, but also a clear spatial pattern. The maximum water depths are placed near the coast, while the upstream floodplain has significantly lower depths. In the right panels of Fig. 4, the maps of flooded areas as derived by CSK data (in magenta) are superimposed to those of water depths. From a qualitative visual comparison between right and left panels of Fig. 4, it can be stated that the agreement is generally good and that the major difference is due to the limited swath of the SAR image (black dashed lines in the right panels). This highlights that a model can complement the satellite–derived


map by filling its gaps due to the limited swath of the radar, as discussed in the introduction. To provide also a quantitative evaluation of the agreement, the SAR-derived maps were first smoothed and then undersampled to the model resolution, while the water depth maps were transformed into flood maps by labeling as flooded the pixels where the water depth exceeded 0.05 m. Taking as reference the SAR derived maps and considering only the overlap between the SAR swath and the model domain, the overall accuracy (OA), defined as the sum of the flood/flood and nonflood/nonflood agreements divided by the total number of pixels was computed for both January 10 and January 16, 2010. OA turned out to be very high for the former day ( ) and quite high even for January 16 ( ). It must be underlined that, while the good agreement between the flood maps obtained for January 10 could be expected since the model was initialized and outputs were validated with the SAR observation, the agreement obtained for January 16 means that on one hand the scenario drawn by the model is reliable and, on the other hand, that the CSK-derived map is physically consistent. Besides the limited SAR swath, other causes of the discrepancies between SAR and model are the fact that the model respects the hydraulic consistency (and is therefore more reliable than SAR) and the lack of detail of the DEM (SAR is more reliable in this case). From the latter point of view, it must be underlined that without constraining the model with SAR observations, the model-derived flood extent map would have included also the area highlighted by the red ellipses in the upper panels of Fig. 4. To explain this behavior, the same region is highlighted by red ellipses in Fig. 1; it can be seen that the area is flat and its ground elevation is low, so that, based only on the information provided by the low-resolution SRTM DEM, the model predicts that floodwater propagates in this area, as well. However, the lower panel of Fig. 1 (a snapshot taken from Google Earth) shows that the region that appears dark in the CSK image acquired on January 10, 2010 (basically corresponding to the flooded area, see Section III-B) has a well-defined boundary (yellow ellipse) that corresponds to a road whose presence is missed by the DEM. This demonstrates how the low detail in the representation of the floodplain can cause errors in the model predictions and the importance of SAR in mitigating them. The left panel of Fig. 5 shows the resulting maximum water depths in each pixel of model domain obtained during the period of the simulation, whereas the right panel shows the modulus of the maximum water speeds reached during the same period according to the model. Fig. 6 shows the maps obtained for the last CSK observations of the area of interest, i.e., January 25 and 31, 2010. The areas where floodwater was still present on January 25–31 according to CSK basically correspond to those where the largest water elevations were predicted by the hydrodynamic model. This indicates that the scenario that can be predicted based on the model outputs is in substantial agreement with the SAR observations performed on the successive days and that the water receding observed by SAR is consistent from the physical point of view. By running the hydrodynamic model, it is also possible to get other kinds of information, useful for emergency management. During the emergency, the most urgent demands of civil protection operators were to have indications about the timing of the



Fig. 5. Left panel: maximum water depths reached during the Shkodër flood as predicted by the hydrodynamic model. Right panel: maximum water speeds reached during the Shkodër flood as predicted by the hydrodynamic model.

output was corrected with the SAR-derived inundation maps by setting the ICs and BCs in order to produce model outputs in agreement with SAR observations. A general consistency between SAR-derived flood extents and the water depths predicted by the model was found. In particular, the maximum water depths were obtained in the areas where floodwater was present for the longest period of time, according to SAR. The role of the model to fill gaps present in the SARderived flood maps because of the radar limited swath was demonstrated as well as the utility of SAR to cope with the model errors due to the possible inaccuracies of the DEMs. Some additional products useful for emergency management, as water velocities and flow directions, were also generated by running the model. Ground truth information was not available, but that the good agreement between SAR-derived and model-derived maps can be considered as an indication of the reliability of the products that were generated to perform the flood damage assessment. It must be finally pointed out that the selection of the model run matching the SAR observation was basically carried out by a skilled flood modeler able not only to assess the agreement between the flood maps but also their consistency from the hydraulic point of view. Future work will concern the increase in the degree of automation of this process. ACKNOWLEDGMENT

Fig. 6. Flood extent maps as detected by CSK. Left panel: January 25, right panel: January 31. Cyan: flooded; blue: permanent water body. The black dashed lines represent the limit of the CSK swath.

recession phase of the flood. The combination of information from the hydrodynamic model and the CSK observations between January 10 and 16 made it possible to provide reliable information about the evolution of the scenario. In particular, the model provided also the horizontal flow speed components and and thus the flow directions. The corresponding vectors are shown in the right panel of Fig. 1 (white arrows). This information, validated by CSK multitemporal observations, can give important elements for the decision maker. It indicates, for instance, which areas will be first accessible and which roads will be restored in a short time. Furthermore, the areas in which the model indicates stagnation of water should be those where the receding has to be artificially accelerated through pumping interventions. V. CONCLUSION A joint use of SAR data and a hydrodynamic model aiming at monitoring the temporal dynamics of floods was presented. Both the flood mapping algorithm and the model were introduced and their application to the flood that hit Albania in January 2010 was described. The short revisit time of CSK images was exploited, so that the acquisition of a dense temporal series of radar data allowed us to perform a satellite monitoring of the drainage of the floodplain with high temporal resolution. To assess the usefulness of these multitemporal SAR observations and to complement their information content for the purpose of a complete flood damage assessment, the hydrodynamic model

The authors would like to thank the ASI manager of the OPERA project, Dr. L. Candela, for her support in gathering COSMO-SkyMed data. REFERENCES [1] M. Chini, L. Pulvirenti, and N. Pierdicca, “Analysis and interpretation of the COSMO-SkyMed observations of the 2011 Japan tsunami,” IEEE Geosci. Remote Sens. Lett., vol. 9, no. 3, pp. 467–471, May 2012. [2] G. J.-P. Schumann et al., “A first large scale flood inundation forecasting model,” Water Resour. Res., vol. 49, pp. 6248–6257, 2013. [3] S. Serpico et al., “Information extraction from remote sensing images for flood monitoring and risk assessment,” Proc. IEEE, vol. 100, no. 10, pp. 2946–2970, Oct. 2012. [4] N. Longbotham et al., “Multimodal change detection, application to the detection of flooded areas: Outcome of the 2009–2010 data fusion contest,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 5, no. 1, pp. 331–342, Feb. 2012. [5] S. Martinis, A. Twele, and S. Voigt, “Towards operational near realtime flood detection using a split-based automatic thresholding procedure on high resolution TerraSAR-X data,” Nat. Hazards Earth Syst. Sci., vol. 9, pp. 303–314, 2009. [6] D. C. Mason, I. J. Davenport, J. C. Neal, G. J.-P. Schumann, and P. D. Bates, “Near real-time flood detection in urban and rural areas using high resolution synthetic aperture radar images,” IEEE Trans. Geosci. Remote Sens., vol. 50, no. 8, pp. 3041–3052, Aug. 2012. [7] P. Pagliara et al., “Dewetra, coping with emergencies,” in Proc. 8th Int. ISCRAM Conf., Lisbon, Portugal, 2011 [Online]. Available: http://www. iscramlive.org/ISCRAM2011/proceedings/papers/190.pdf [8] D. Mandl et al., “Use of the earth observing one (EO-1) satellite for the Namibia SensorWeb flood early warning pilot,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 6, no. 2, pp. 298–308, Apr. 2013. [9] L. C. Smith, “Satellite remote sensing of river inundation area, stage and discharge: A review,” Hydrol. Process., vol. 11, pp. 1427–1439, 1997. [10] F. S. Marzano et al., “Potential of high-resolution detection and retrieval of precipitation fields from X-band spaceborne synthetic aperture radar over land,” Hydrol. Earth Syst. Sci., vol. 15, pp. 859–875, 2011. [11] L. Pulvirenti, F. S. Marzano, N. Pierdicca, S. Mori, and M. Chini, “Discrimination of water surfaces, heavy rainfall and wet snow using COSMOSkyMed observations of severe weather events,” IEEE Trans. Geosci. Remote Sens., vol. 52, no. 2, pp. 858–869, Feb. 2014.


[12] L. Pulvirenti, N. Pierdicca, M. Chini, and L. Guerriero, “Monitoring flood evolution in agricultural areas using COSMO-SkyMed data: The Tuscany 2009 case study,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 6, no. 4, pp. 1807–1816, Aug. 2013. [13] L. Pulvirenti, M. Chini, N. Pierdicca, L. Guerriero, and P. Ferrazzoli, “Flood monitoring using multi-temporal COSMO-SkyMed data: Image segmentation and signature interpretation,” Remote Sens. Environ., vol. 115, pp. 990–1002, 2011. [14] L. Giustarini et al., “A change detection approach to flood mapping in urban areas using TerraSAR-X,” IEEE Trans. Geosci. Remote Sens., vol. 51, no. 4, pp. 2417–2430, Apr. 2013. [15] K. Voormansik, J. Praks, O. Antropov, J. Jagomägi, and K. Zalite, “Flood mapping with TerraSAR-X in forested regions in Estonia,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 7, no. 2, pp. 562–577, Feb. 2014. [16] F. Covello et al., “COSMO-SkyMed an existing opportunity for observing the Earth,” J. Geodyn., vol. 49, no. 3–4, pp. 171–180, Apr. 2010. [17] G. Schumann, P. D. Bates, M. S. Horritt, P. Matgen, and F. Pappenberger, “Progress in integration of remote sensing—Derived flood extent and stage data and hydraulic models,” Rev. Geophys., vol. 47, RG 4001, 2009. [18] L. Giustarini et al., “Assimilating SAR-derived water level data into a hydraulic model: A case study,” Hydrol. Earth Syst. Sci., vol. 15, pp. 2349–2365, 2011. [19] R. Hostache et al., “Water level estimation and reduction of hydraulic model calibration uncertainties using satellite SAR images of floods,” IEEE Trans. Geosci. Remote Sens., vol. 47, no. 2, pp. 431–441, Feb. 2009. [20] G. Boni et al., “The OPERA project: EO-based flood risk management in Italy,” in Proc. IEEE Int. Geosci. Remote Sens. Symp. (IGARSS), 2009, vol. 2, pp. II-929–II-932. [21] N. Pierdicca, L. Pulvirenti, M. Chini, L. Guerriero, and L. Candela, “Observing floods from space: Experience gained from COSMO-SkyMed observations,” Acta Astronaut., vol. 84, pp. 122–133, Mar./Apr. 2013. [22] N. Pierdicca, M. Chini, L. Pulvirenti, and F. Macina, “Integrating physical and topographic information into a fuzzy scheme to map flooded area by SAR,” Sensors, vol. 8, pp. 4151–4164, 2008. [23] M. S. Horritt, D. C. Mason, and A. J. Luckman, “Flood boundary delineation from synthetic aperture radar imagery using a statistical active contour model,” Int. J. Remote Sens., vol. 22, pp. 2489–2507, 2001. [24] 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,” Nat. Hazards Earth Syst. Sci., vol. 11, pp. 529–540, 2011. [25] M. Bracaglia, P. Ferrazzoli, and L. Guerriero, “A fully polarimetric multiple scattering model for crops,” Remote Sens. Environ., vol. 54, pp. 170–179, 1995. [26] P. Ferrazzoli and L. Guerriero, “Radar sensitivity to tree geometry and woody volume: A model analysis,” IEEE Trans. Geosci. Remote Sens., vol. 33, no. 2, pp. 360–371, Mar. 1995. [27] Y. Oh, “Quantitative retrieval of soil moisture content and surface roughness from multipolarized radar observations of bare soil surfaces,” IEEE Trans. Geosci. Remote Sens., vol. 42, no. 3, pp. 596–601, Mar. 2004. [28] F. T. Ulaby and C. Dobson, Handbook of Radar Scattering Statistics for Terrain. Norwood, MA, USA: Artech House, 1989. [29] P. Prestininzi and A. Fiori, “A two-dimensional parabolic model for flood assessment,” Italian J. Eng. Geol. Environ., vol. 1, pp. 5–18, 2006. [30] B. Cappelaere, “Accurate diffusive wave routing,” J. Hydraul. Eng., vol. 123, no. 3, pp. 174–181, 1997. [31] A. Leopardi, E. Oliveri, and M. Greco, “Two-dimensional modeling of floods to map risk prone areas,” J. Water Res. Plann. Manage., vol. 128, pp. 168–178, 2002. [32] V. M. Ponce and D. B. Simons, “Shallow wave propagation in open channel flow,” J. Hydraul. Div., vol. 103, no. 12, pp. 1461–1476, 1977. Luca Pulvirenti (M’07) received the Laurea degree in electronic engineering and the Ph.D. degree in electromagnetism both from Sapienza University of Rome, Rome, Italy, in 1999 and 2004, respectively. He was a Postdoctoral Researcher with the Department of Electronic Engineering, Sapienza University of Rome, from 2004 to 2009, and a Fixed-Term Researcher with the Department of Information Engineering, Electronics, and Telecommunications, Sapienza University of Rome, from 2010 to 2013. Since October 2013, he is a Researcher with CIMA Research Foundation. His research interests include microwave remote sensing (active and passive) of the atmosphere and the Earth’s surface, with particular emphasis on hydrological applications of remote sensing, and electromagnetic scattering models of rough surfaces. Dr. Pulvirenti is an Associate Editor of the IEEE GEOSCIENCE AND REMOTE SENSING LETTERS. He is a member of the Italian Society of Electromagnetism and the Italian Association of Remote Sensing.


Nazzareno Pierdicca (M’04–SM’13) received the Laurea (Doctor’s) degree (cum laude) in electronic engineering from Sapienza University of Rome, Rome, Italy, in 1981. From 1978 to 1982, he worked for the Italian Agency for Alternative Energy (ENEA). From 1982 to 1990, he worked with the Remote Sensing Division of Telespazio, Rome, Italy. In November 1990, he joined the Department of Information, Electronic, and Telecommunication Engineering, Sapienza University of Rome. He is currently a Full Professor of Remote Sensing with the Faculty of Engineering, Sapienza University of Rome. His research interests include electromagnetic scattering end emission models for sea and bare soil surfaces, microwave radiometry of the atmosphere, SAR land applications, inversion of electromagnetic models for land parameter retrieval, calibration of the spaceborne radar altimeter, detection of urban changes from SAR interferometry, and GNSS reflectometry for land applications. Prof. Pierdicca is a member of the IEEE Geoscience and Remote Sensing Society (GRSS) and a past Chairman of the GRSS Central Italy Chapter.

Giorgio Boni received the Laurea degree in hydraulic engineering from the University of Genoa, Genoa, Italy, and the Ph.D. degree in hydrodynamics from the University of Padua, Padua, Italy, in 1992 and 1997, respectively. From 1998 to 1999, he was a Visiting Scientist with MIT—R.M. Parsons Lab. working on satellite data assimilation for hydrological applications. Since 2008, he has been a Scientific Secretary of the Hydro-Meteorological Hazards section and since 2013, he has been a Vice President of the Natural Hazards Division of the European Geosciences Union. From 2008 to 2012, he was the Scientific Director of CIMA Research Foundation formed by the Italian Department of the Prime Minister Office for Civil Protection and the University of Genoa. He is currently an Assistant Professor of Environmental Engineering with the University of Genoa (since 2000) and the Vice President and Coordinator of the international activities of the Master Course in Environmental and Energy Engineering. His research interests include hydro-meteorology and eco-hydrology with special focus on remote sensing applications to hydrometeorological risk management, satellite data assimilation in hydrological models for evaporation and evapotranspiration estimation at regional scale, flash-flood forecasting, and statistical analysis of the hydrometeorological extremes. He is the author and coauthor of more than 130 papers (29 on international peer-reviewed journals).

Mattia Fiorini received the Laurea degree (cum laude) in environmental engineering on natural and industrial risk management from the University of Genoa, Genoa, Italy, in 2010, where he is currently pursuing the Ph.D. degree in monitoring of systems and environmental risk management. His research interests include both prevention (by simulations of various scenarios) and early disaster management (by real-time maps of water levels and velocity, structures damages, and pollutants and pathogenic microorganism concentration over the entire involved area).

Roberto Rudari received the Laurea degree (M.Sc. equivalent; summa cum laude) in hydraulic civil engineering from the University of Genoa, Genoa, Italy, in 1998, and the Ph.D. degree in hydraulic engineering from the University of Padua, Padua, Italy, in 2002. From 2000 to 2001, he was a Visiting Ph.D. at Parsons Lab, Massachusetts Institute of Technology (MIT), Cambridge, MA, USA. He was a Postdoctoral Researcher with the University of Genoa (2003–2004) and a Researcher with the National Research Council, Perugia (2004–2006). He had a Research position with the University of Genoa (2006–2008). Since 2008, he has been a Senior Researcher and a Project Leader with CIMA Research Foundation. He is the Programmer of the MIKEDriFt rainfall-runoff model commercialized by DHI. He was a Consultant for the WMO Associated Program on Flood Management and Consultant for the UNISDR Global Assessment Report for the Global Flood Model Development. He is currently coordinating the RASOR FP7 Project. He is an author of more than 100 papers (26 on international refereed journals). His research interests include the statistical characterization of land effects and vulnerability estimation to flood events, and the improvement of combined hydrometeorological forecast systems based on probabilistic concepts.

Lihat lebih banyak...


Copyright © 2017 DADOSPDF Inc.