Risk Analysis, Vol. 34, No. 8, 2014
DOI: 10.1111/risa.12156
Flood Hazard and Flood Risk Assessment Using a Time Series of Satellite Images: A Case Study in Namibia Sergii Skakun,1,2,∗ Nataliia Kussul,1,3 Andrii Shelestov,1,2,3 and Olga Kussul3
In this article, the use of time series of satellite imagery to flood hazard mapping and flood risk assessment is presented. Flooded areas are extracted from satellite images for the flood-prone territory, and a maximum flood extent image for each flood event is produced. These maps are further fused to determine relative frequency of inundation (RFI). The study shows that RFI values and relative water depth exhibit the same probabilistic distribution, which is confirmed by Kolmogorov-Smirnov test. The produced RFI map can be used as a flood hazard map, especially in cases when flood modeling is complicated by lack of available data and high uncertainties. The derived RFI map is further used for flood risk assessment. Efficiency of the presented approach is demonstrated for the Katima Mulilo region (Namibia). A time series of Landsat-5/7 satellite images acquired from 1989 to 2012 is processed to derive RFI map using the presented approach. The following direct damage categories are considered in the study for flood risk assessment: dwelling units, roads, health facilities, and schools. The produced flood risk map shows that the risk is distributed uniformly all over the region. The cities and villages with the highest risk are identified. The proposed approach has minimum data requirements, and RFI maps can be generated rapidly to assist rescuers and decisionmakers in case of emergencies. On the other hand, limitations include: strong dependence on the available data sets, and limitations in simulations with extrapolated water depth values. KEY WORDS: Direct damage; flood hazard mapping; flood risk assessment; remote sensing
1. INTRODUCTION
In recent years, flood management has shifted from protection against floods to managing the risks of floods (European Flood Risk Directive).(2) To provide flood risk assessment, corresponding flood hazard and flood risk maps need to be elaborated. Flood risk is a function of two arguments: hazard probability and vulnerability.(2–11) In other words, risk is a mathematical expectation of vulnerability (consequences) function. Flood probabilities are determined in order to produce flood hazard maps. Hydraulic modeling of a peak flow is performed to derive critical flood characteristics such as water depth and flow velocity. These parameters are used as inputs to mortality and damage functions to estimate potential losses of life(5) and economical damages(7) due to floods. However, running hydraulic models faces many uncertainties(12) due to the lack of hydrological and other required data, their
Over last decades, we have witnessed an upward global trend in natural disaster occurrence. Hydrological and meteorological disasters are the main contributors to this pattern. In 2011, hydrological disasters, such as floods and wet mass movements, represented 52% of the overall disasters reported, causing 139.8 million victims and more than U.S. $70 billion in damages.(1) 1 Space
Research Institute NASU-SSAU, Kyiv, 03680, Ukraine. University of Life and Environmental Sciences of Ukraine, Kyiv, 03680, Ukraine. 3 National Technical University of Ukraine “Kyiv Polytechnic Institute,” Kyiv, 03056, Ukraine. ∗ Address correspondence to Sergii Skakun, Space Research Institute NASU-SSAU, Glushkov Ave., 40, build. 4/1, Kyiv, 03680, Ukraine; tel: +38-044-5262553; fax: +38-044-5264124; [email protected]. 2 National
1521
C 2013 Society for Risk Analysis 0272-4332/14/0100-1521$22.00/1
1522 incompleteness and imperfection.(2) The use of space-borne remote sensing data to flood risk mapping is a complement approach to the existing flood modeling techniques.(8,12–15) Remote sensing from space plays an important role in flood modeling and flood risk assessment.(11,13–21) Satellite images acquired in both optical and microwave range of the electro-magnetic emission are utilized for solving many problems related to flood risk management. The use of optical satellite imagery is usually limited by cloud cover, rain conditions, and nighttime. On the other hand, synthetic-aperture radar (SAR) measurements from space are independent of daytime and weather conditions. The satellite-derived flood extent(22–26) is invaluable for calibration and validation of hydraulic models.(12,27–31) Moreover, availability of flood extent maps in near real time (within 24 hours after satellite image acquisition(19,20,30,32–36) ) can greatly benefit rescuers during flooding and can also be used for rapid damage assessment. A combination of satellite-derived flood areas with a digital elevation model (DEM) allows indirect retrieval of water stages.(37–39) Water level observations derived from remote sensing imagery can be further assimilated into a hydrodynamic model to decrease forecast uncertainty.(40,41) Many studies have shown the efficiency of using satellite images for flood damage assessment.(42–47) Different types of damages caused by floods are mapped and assessed using remote sensing imagery: urban areas,(42,45) agriculture (crop) areas,(42,46) or general land cover types.(43,44,47) Satellite images can be directly used for flood hazard mapping. In Ref. 8, a novel approach for rapid flood risk mapping is proposed based on the use of radar satellite data. Flood extent maps derived from satellite imagery using different image processing techniques are fused to generate an event-specific flood hazard map. The flood risk map is further generated from the flood hazard map and vulnerabilityweighted land cover vector data. In Ref. 47, a flood hazard map is produced based on a flood frequency map retrieved from a time series of satellite images. Depending on the land cover and elevation each category is empirically assigned a weight score. The resulting flood hazard map derived from satellite imagery is compared to a flood depth map derived from a hydrodynamic model. It is reported that there is a 98% spatial agreement between two maps. In this article, the use of time series of satellite imagery to flood hazard mapping and flood risk as-
Skakun et al. sessment is presented. Flooded areas are extracted from each satellite image, and a maximum flood extent image for each flood event is generated. These maps are further integrated to determine relative frequency of inundation (RFI), which is used to provide flood risk assessment. The presented methodology is applied to the Katima Mulilo region of Namibia and has been used within the “Namibia SensorWeb Pilot Project.” This project aims at developing an operational transboundary flood management decision support system for the southern African region to provide useful flood and water-borne disease forecasting tools for local decisionmakers.(48,49) The remainder of the article is organized as follows. The method used for flood hazard mapping from a time series of satellite images and for flood risk assessment is outlined in Section 2. Results are presented in Section 3. A discussion of the obtained results and conclusions is given in Section 4.
2. METHODOLOGY AND TEST SITE DESCRIPTION 2.1. RFI Retrieval from a Time Series of Optical Satellite Images Let X = x(y, d) y∈Y,d∈Dy be a set of optical satellite images acquired over the given region. Note that Y denotes a set of years and Dy denotes days of the year y for which satellite imagery is available. The image x(y, d) is acquired on the day d of the year y. It is assumed that all satellite images are preprocessed for radiometric and geometric corrections, and clouds, shadows, and missing data are identified and masked for each image. Let u(y, d) be a flood extent image extracted from the satellite image x(y, d) using some classification algorithm. Within the classification procedure each pixel (i, j) of the image u(y, d) is assigned one of the following classes: u(y, d, i, j) = water/no water/no data. The “water” class is assigned to the pixels that are identified as flooded areas on the image x(y, d); the “no water” class is assigned to the pixels that are identified as dry areas on the image x(y, d); the “no data” class is reserved for the pixels that are identified as clouds, shadows, or missing data on the image x(y, d). Let Ey be a set of flood events during the year y, and Dey denotes days of the year y on which satellite imagery was acquired during flood event e ∈ Ey .
Flood Hazard and Flood Risk Assessment
1523
For each flood event e ∈ Ey , a maximum flood extent image u(e) is generated using the following equation: ⎧ e ⎪ ⎪ water, if d ∈ Dy exists suchas : u(y, d, i, j) = water, ⎪ ⎪ ⎨ no water, if for all d ∈ De : u(y, d, i, j) = no water, y u(e, i, j) = ⎪ no data, if for each d ∈ De : u(y, d, i, j) = ⎪ y ⎪ ⎪ ⎩ no water or no data. (1)
In other words, if a pixel is classified as “water” on at least one flood extent image u(y, d), then this pixel is assigned a “water” class on the resulting maximum flood extent image u(e) for the flood event e. If a pixel is classified as “no water” on every image acquired during the flood event e then it is assigned a “no water” class on the resulting maximum flood extent image u(e). And finally, the pixel is assigned a “no data” class if it is classified either a “no water” or “no data” class during the flood event e. This is done in such a way since it is unknown whether the “no data” pixel was actually flooded or not during the flood event in question. The derived maximum flood extent images u(e) are used to derive a RFI value: 1 1u(e,i, j)=water , (2) RFI(i, j) = |U| u(e)∈U
where U = {u(e) : y ∈ Y, e ∈ Ey and u(e, i, j) = water or no water}, and 1() is an indicator function. Therefore, for the region (i, j) only those flood extent images are taken into consideration for which this pixel is assigned either “water” or “no water.” The general diagram for determination of the RFI value is shown in Fig. 1. Let w(e) be a maximum water level recorded for the flood event e. Let w(1) < w(2) 0
Risk by damage categories, USD
7.850 1/30 Dwelling units Health facilities Schools Roads, m Dwelling units Health facilities Schools Roads
Total risk, USD
summarizes these estimates. RFI values are estimated for different water levels according to Equation (3) that represent 30-, 8.5-, 5.1-year floods.(51) Amount of damage categories with nonzero RFI values, estimated risk by categories, and a total risk
7.370 1/8.5 7,017 8 8 71,140 1,924,398 8,158 10,302 151,638 2 094,497
7.027 1/5.1 5,468 8 6 65,530 1,571,570 7,575 8,321 135,877 1 723,343
4,379 8 4 58,010 1,276,324 6,625 6,687 113,667 1 403,302
of damages (estimated using Equation (5)) are presented in Table V. Estimates for the 2009 flood (a 30-year flood) show that at least eight schools, eight health facilities, and 7,017 dwelling units were flooded in the Caprivi region. These assessment
1532
Skakun et al.
Fig. 9. Bed elevation profiles for cross-sections 1–4 derived from SRTM DEM at 90-m spatial resolution.
results correspond to the estimates reported in Ref. 51. About 71 km of roads (mainly district) are detected with RFI > 0, meaning they were flooded during 1989–2012 at least once. (Amount of roads that were flooded in the Caprivi region is not given in Ref. 51.) Within all scenarios considered, damages to dwelling units represent about 91% of total damages. Although direct damages to roads, health facilities, and schools are smaller compared to dwelling units, indirect damages should be taken into account to quantify risk more adequately. In particular, flooded and damaged roads will considerably reduce ability for timely evacuation and increase time necessary for providing assistance by rescuers. Flooded and damaged health facilities will reduce capacities for treating and providing assistance to affected people. It will therefore require people to be transported to nearby nonflooded health facilities, which coupled with damaged road networks, will reduce response
time. All these factors need to be also quantified and properly incorporated into the flood risk model (Equation (4)). The corresponding flood risk map for 2009 is generated, which allows identification of regions with the highest risk. For this, the region is divided into the grid of cells at 960 m × 960 m and a total risk is estimated for each cell according to Equation (5). The derived map is shown in Fig. 11. In general, the risk is distributed uniformly all over the region. The cities and villages with the highest risk are: Lisikili, Muzii, Kasika, Ivilivinzi, and Ngwese. Fig. 12 shows an example of flood risk map for the region of the village of Kasika. 4. DISCUSSION AND CONCLUSIONS Namibia, and the Caprivi region in particular, is prone to floods. A 2009 flood that was the largest
Flood Hazard and Flood Risk Assessment
1533
Fig. 10. Empirical CDF (panels A and B) and dependence between water depth and RFI (C and D) for cross-sections 1 (left panels) and 4 (right panels).
since 1969 and corresponded to a 30-year flood with 7.85-m water level was followed by moderate floods in 2010 and 2011. Flood risk assessment becomes an extremely important issue for the region from different perspectives: economical, social, environmental, health, etc. In this article, we take advantage of freely available archived satellite images and propose an approach for rapid flood hazard mapping and flood risk assessment in the Katima Mulilo region (Caprivi, Namibia). In particular, 97 Landsat-5 and Landsat-7 satellite images acquired from 1989 to 2012 are processed and analyzed to extract flood extent images that are further integrated to estimate relative RFI. The study shows that RFI values and water depth (normalized on maximum water level) exhibit the same probabilistic distribution, which is confirmed by the KS test. Furthermore, when analyzing river crosssections the RFI values can be considered as samples
(in general case irregular) of the water depth continues function. These sample values can be used to form a step-wise function that approximates relative water depth. It is shown that under the condition of regular samples the error of approximation will decrease as the number of samples increases. Therefore, the produced RFI map can be used as a flood hazard map, especially in cases when flood modeling is complicated by the lack of available data and high uncertainties. It should be, however, noted that only a coarse resolution DEM (90-m spatial resolution) was available for the Namibian case, and consequently the derived water depths were not very accurate. This should be taken cautiously into consideration when analyzing the relationship between RFI and relative water depth, and how accurately water depth can be restored from the RFI values. In our case
1534
Skakun et al.
Fig. 11. Flood risk for the Katima Mulilo region with values aggregated into cells of 960 m × 960 m.
Fig. 12. Flood hazard map (RFI) overlaid with dwelling units (left) and flood risk map (right) for the region of the village of Kasika (Caprivi, Namibia).
the obtained coefficient of determination of linear regression between RFI and relative water depth varied considerably (0.21–0.65), meaning there were regions with poor correspondence between these two parameters. Therefore, a more accurate DEM is nec-
essary to study more thoroughly a relationship between water depth and RFI. The derived RFI map is further used for flood risk assessment. The following direct damage categories are considered: dwelling units, roads, health
Flood Hazard and Flood Risk Assessment facilities, and schools. These categories accounted for more than 55% of total damages reported during a 2009 flood in the Caprivi region.(51) Though information on other categories of damages has not been available in this study, it provided useful information on the geographical distribution of regions with highest risk. It is estimated that 7,017 dwelling units, eight schools, eight health facilities, and 71 km of district roads in the Katima Mulilo regions were flooded in 2009, which corresponded to estimates provided in the “Post Disaster Needs Assessment 2009” report.(51) Dwelling units provided a major contribution (up to 91%) to the total estimated risk comparing to other damage categories considered. However, it should be noted that the risk associated with flooded and affected roads and health facilities should include not only direct damages but also indirect damages. These damages should account for reduced evacuation time and transportation capacities as well as reduced capacities for providing assistance and treatment to people affected. This is one of the areas where the proposed model needs to be improved in the future. Another issue that should be taken into consideration is that in our case flood damages were estimated using only historical data for the single flood event reported in Ref. 51. Detailed damage estimates on other flood events in the region were not available, to our best knowledge. This did not allow us to provide statistically consistent estimates of flood damages. (By statistically consistent estimates, we mean estimates that converge in probability to the true value.) Therefore, more historical data on flood damages (e.g., from neighbor countries) are necessary to incorporate them into the proposed flood risk model. The presented approach for flood hazard mapping using time series of satellite data has the following advantages. It has minimum data requirements, and RFI maps can be generated rapidly to assist rescuers and decisionmakers in case of emergencies. These maps can be further used for rapid flood risk assessment to identify regions with the highest risk. Furthermore, the derived flood extent maps can be used to calibrate and validate hydrological models once high-resolution topographic data become available. On the other hand, the presented approach has certain limitations. It is a data-driven methodology and, therefore, the quality of the derived RFI maps is dependent on the available satellite images. The proposed approach can be hardly applied for predicting flood inundated areas with installed weirs. While
1535 within the proposed approach the RFI values can be used to interpolate water depth, they cannot be used to extrapolate water depth. In other words, it is difficult to apply the proposed methodology to simulate scenarios with the water depth values more than the maximum water depth value available in remote sensing observations. Other case studies should be considered to assess efficiency of the presented approach for other regions and under different conditions (hydrometeorological, quality, and availability of satellite data, etc.). REFERENCES 1. Guha-Sapir D, Vos F, Below R, Ponserre S. Annual Disaster Statistical Review 2011: The Numbers and Trends. Brussels: Centre for Research on the Epidemiology of Disasters (CRED), 2012. 2. Mostert E, Junier SJ. The European flood risk directive: Challenges for research. Hydrology and Earth System Sciences Discussion, 2009; 6(4):4961–4988. 3. Jonkman SN, van Gelder PHAJM, Vrijling JK. An overview of quantitative risk measures for loss of life and economic damage. Journal of Hazardous Materials, 2003; A99:1–30. 4. Hoes O, Schuurmans W. Flood standards or risk analyses for polder management in the Netherlands. Journal of Irrigation and Drainage Engineering, 2006; 55:113–119. 5. Jonkman SN. Loss of life estimation in flood risk assessment, PhD thesis, Delft University, 2007. 6. Jonkman SN, Kok M, Vrijling JK. Flood risk assessment in the Netherlands: A case study for dike ring South Holland. Risk Analysis, 2008; 28(5):1357–1374. 7. Jonkman SN, Boˇckarjova M, Kok M, Bernardini P. Integrated hydrodynamic and economic modelling of flood damage in the Netherlands. Ecological Economics, 2008; 66:77–90. 8. Schumann G, Di Baldassarre G. The direct use of radar satellites for event-specific flood risk mapping. Remote Sensing Letters, 2010; 1(2):75–84. 9. Kussul NN, Sokolov BV, Zyelyk YI, Zelentsov VA, Skakun SV, Shelestov AYu. Disaster risk assessment based on heterogeneous geospatial information. Journal of Automation and Information Science, 2010; 42(12):32–45. 10. Kellens W, Terpstra T, De Maeyer P. Perception and communication of flood risks: A systematic review of empirical research. Risk Analysis, 2013; 33:24–49. 11. Taubenbock H, Post J, Roth A, Zosseder K, Strunz G, Dech S. A conceptual vulnerability and risk framework as outline to identify capabilities of remote sensing. Natural Hazards and Earth System Sciences, 2008; 8:409–420. 12. Horritt MS. A methodology for the validation of uncertain flood inundation models. Journal of Hydrology, 2006; 326:153– 165. 13. Bates PD, Horritt MS, Smith CN, Mason DC. Integrating remote sensing observations of flood hydrology and hydraulic modelling. Hydrological Processes, 1997; 11:1777–1795. 14. Bates PD. Invited commentary: Remote sensing and flood inundation modelling. Hydrological Processes, 2004; 18:2593– 2597. 15. Lecca G, Petitdidier M, Hluchy L, Ivanovic M, Kussul N, Ray N, Thieron V. Grid computing technology for hydrological applications. Journal of Hydrology, 2011; 403(1–2):186–199. 16. Voigt S, Kemper T, Riedlinger T, Kiefl R, Scholte K, Mehl H. Satellite image analysis for disaster and crisis management
1536
17.
18. 19. 20. 21. 22.
23.
24.
25. 26.
27. 28. 29.
30.
31.
32. 33. 34. 35.
support. IEEE Transactions on Geoscience and Remote Sensing, 2007; 45(6):1520–1528. Schumann G, Bates PD, Horritt MS, Matgen P, Pappenberger F. Progress in integration of remote sensing–derived flood extent and stage data and hydraulic models. Reviews of Geophysics, 2009; 47(4): Art. RG4001. Stryker T, Jones B. Disaster response and the international charter program. Photogrammetric Engineering & Remote Sensing, 2009; 75(12):1342–1344. Kussul N, Shelestov A, Skakun S. Grid and sensor web technologies for environmental monitoring. Earth Science Informatics, 2009; 2(1–2):37–51. Cossu R, Schoepfer E, Bally P, Fusco L. Near real-time SARbased processing to support flood monitoring. Journal of RealTime Image Processing, 2009; 4:205–218. Guo H. Understanding global natural disasters and the role of earth observation. International Journal of Digital Earth, 2010; 3(3):221–230. Kussul NN, Lupian EA, Shelestov AY, Skakun SV, Tishchenko YG, Hluchy L. Determination of inundated territories on the basis of integration of heterogeneous data. Journal of Automation and Information Science, 2007; 39(12): 42–51. Kravchenko AN, Kussul NN, Lupian EA, Savorsky VP, Hluchy L, Shelestov AY. Water resource quality monitoring using heterogeneous data and high-performance computations. Cybernetics and Systems Analysis, 2008; 44(4):616–624. Schumann G, Di Baldassarre G, Bates PD. The utility of spaceborne radar to render flood inundation maps based on multialgorithm ensembles. IEEE Transactions on Geoscience and Remote Sensing, 2009; 47(8):2801–2807. Skakun S. A neural network approach to flood mapping using satellite imagery. Computing and Informatics, 2010; 29(6):1013–1024. Pulvirenti L, Chini M, Pierdicca N, Guerriero L, Ferrazzoli P. Flood monitoring using multi-temporal COSMO-SkyMed data: Image segmentation and signature interpretation. Remote Sensing of Environment, 2011; 115(4):990–1002. Horritt MS, Bates PD. Evaluation of 1D and 2D numerical models for predicting river flood inundation. Journal of Hydrology, 2002; 268:87–99. Di Baldassarre G, Schumann GJ-P, Bates PD. Near real time satellite imagery to support and verify timely flood modelling. Hydrological Processes, 2009; 23:799–803. Di Baldassarre G, Schumann GJ-P, Bates PD. A technique for the calibration of hydraulic models using uncertain satellite observations of flood extent. Journal of Hydrology, 2009; 367:276–282. Di Baldassarre G, Schumann GJ-P, Brandimarte L, Bates PD. Timely low resolution SAR imagery to support floodplain modelling: A case study review. Surveys in Geophysics, 2011; 32:255–69. Khan SI, Yang H, Wang J, Yilmaz KK, Gourley JJ, Adler RF, Brakenridge GR, Policelli F, Habib S, Irwin D. Satellite remote sensing and hydrologic modeling for flood inundation mapping in Lake Victoria basin: Implications for hydrologic prediction in ungauged basins. IEEE Transactions on Geoscience and Remote Sensing, 2011; 49(1):85–95. Bakan GM, Kussul NN. Fuzzy ellipsoidal filtering algorithm of static object state. Problemy Upravleniya I Informatiki (Avtomatika), 1996; 5:77–92. Shelestov A, Kussul N, Skakun S. Grid technologies in monitoring systems based on satellite data. Journal of Automation and Information Science, 2006; 38(3):69–80. Shelestov A, Kussul N. Using the fuzzy-ellipsoid method for robust estimation of the state of a grid system node. Cybernetics and Systems Analysis, 2008; 44(6):847–854. Kussul N, Shelestov A, Skakun S, Li G, Kussul O. The wide area grid testbed for flood monitoring using earth observation
Skakun et al.
36.
37.
38.
39. 40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2012; 5(6):1746–1751. Kussul N, Shelestov A, Skakun S, Kravchenko O, Gripich Y, Hluchy´ L, Kopp P, Lupian E. The data fusion grid infrastructure: Project objectives and achievements. Computing and Informatics, 2010; 29(2):319–334. Schumann G, Matgen P, Pappenberger F. Conditioning water stages from satellite imagery on uncertain data points. Geoscience and Remote Sensing Letters, 2008; 5:810– 813. Matgen P, Schumann G, Henry J, Hoffmann L, Pfister L. Integration of SAR-derived inundation areas, high precision topographic data and a river flow model toward real-time flood management. International Journal of Applied Earth Observation and Geoinformation, 2007; 9(3):247–263. Mason DC, Bates PD, Dall’Amico JT. Calibration of uncertain flood inundation models using remotely sensed water levels. Journal of Hydrology, 2009; 368:224–236. Garc´ıa-Pintado J, Neal JC, Mason DC, Dance SL, Bates PD. Scheduling satellite-based SAR acquisition for sequential assimilation of water level observations into flood modelling. Journal of Hydrology, 2013; 495:252–266. Giustarini L, Matgen P, Hostache R, Montanari M, Plaza D, Pauwels VRN, De Lannoy GJM, De Keyser R, Pfister L, Hoffmann L, Savenije HHG. Assimilating SAR-derived water level data into a hydraulic model: A case study. Hydrology and Earth System Sciences, 2011; 15:2349–2365. Qi H, Altinakar MS. Simulation-based decision support system for flood damage assessment under uncertainty using remote sensing and census block information. Natural Hazards, 2011 Nov; 59(2):1125–1143. van der Sande CJ, de Jong SM, de Roo APJ. A segmentation and classification approach of IKONOS-2 imagery for land cover mapping to assist flood risk and flood damage assessment. International Journal of Applied Earth Observation and Geoinformation, 2003; 4(3):217–229. Serpico SB, Dellepiane S, Boni G, Moser G, Angiati E, Rudari R. Information extraction from remote sensing images for flood monitoring and damage evaluation. Proceedings of the IEEE, 2012; 100(10): 2946–2970. Taubenbock H, Wurm M, Netzband M, Zwenzner H, Roth A, Rahman A, Dech S. Flood risks in urbanized areas—Multisensoral approaches using remotely sensed data for risk assessment. Natural Hazards and Earth System Sciences, 2011; 11:431–444. Tapia-Silva F-O, Itzerott S, Foerster S, Kuhlmann B, Kreibich H. Estimation of flood losses to agricultural crops using remote sensing. Physics and Chemistry of the Earth, Parts A/B/C, 2011; 36(7–8):253–265. Dewan AM, Kumamoto T, Nishigaki M. Flood hazard delineation in greater Dhaka, Bangladesh using an integrated GIS and remote sensing approach. Geocarto International, 2006; 21(2):33–38. Kussul N, Mandl D, Moe K, Mund JP, Post J, Shelestov A, Skakun S, Szarzynski J, Van Langenhove G, Handy M. Interoperable infrastructure for flood monitoring: SensorWeb, Grid and Cloud. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2012; 5(6): 1740– 1745. Mandl D, Frye S, Cappelaere P, Handy M, Policelli F, Katjizeu M, Van Langenhove G, Aube G, Saulnier J, Sohlberg R, Silva JA, Kussul N, Skakun S, Ungar SG, Grossman R, Szarzynski J. Use of the earth observing one (EO-1) satellite for the Namibia SensorWeb flood early warning pilot. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2013; 6(2): 298–308. Vapnik VN. An overview of statistical learning theory. IEEE Transactions on Neural Networks, 1999; 10(5):988– 999.
Flood Hazard and Flood Risk Assessment 51. Post Disaster Needs Assessment 2009. A Report Prepared by the Government of the Republic of Namibia, with support from the International Community, 2009. Available at: http://siteresources.worldbank.org/INTAFRICA/Resources/ WorldBankNamibiaReportFAlr.pdf, Accessed July 12, 2013. 52. Van Langenhove G. Hydro-meteorological extremes—Floods and droughts in Namibia [Internet]. Climate Change Capacity Building Workshop, 2012 October 09–11, Windhoek, Namibia. Available at: http://www.sarva.org.za/sadc/ download/namibia2012 11.pdf, Accessed on July 30, 2013.
1537 53. Chander G, Markham BL, Helder DL. Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. Remote Sensing of Environment, 2009; 113:893–903. 54. Frazier PS, Page KJ. Water body detection and delineation with landsat TM data. Photogrammetric Engineering & Remote Sensing, 2000; 66(12):1461–1467. ¨ 55. Aldroubi A, Grochenig K. Nonuniform sampling and reconstruction in shift-invariant spaces. SIAM Review, 2001; 43(4):585–620.
DOI: 10.1111/risa.12156
Flood Hazard and Flood Risk Assessment Using a Time Series of Satellite Images: A Case Study in Namibia Sergii Skakun,1,2,∗ Nataliia Kussul,1,3 Andrii Shelestov,1,2,3 and Olga Kussul3
In this article, the use of time series of satellite imagery to flood hazard mapping and flood risk assessment is presented. Flooded areas are extracted from satellite images for the flood-prone territory, and a maximum flood extent image for each flood event is produced. These maps are further fused to determine relative frequency of inundation (RFI). The study shows that RFI values and relative water depth exhibit the same probabilistic distribution, which is confirmed by Kolmogorov-Smirnov test. The produced RFI map can be used as a flood hazard map, especially in cases when flood modeling is complicated by lack of available data and high uncertainties. The derived RFI map is further used for flood risk assessment. Efficiency of the presented approach is demonstrated for the Katima Mulilo region (Namibia). A time series of Landsat-5/7 satellite images acquired from 1989 to 2012 is processed to derive RFI map using the presented approach. The following direct damage categories are considered in the study for flood risk assessment: dwelling units, roads, health facilities, and schools. The produced flood risk map shows that the risk is distributed uniformly all over the region. The cities and villages with the highest risk are identified. The proposed approach has minimum data requirements, and RFI maps can be generated rapidly to assist rescuers and decisionmakers in case of emergencies. On the other hand, limitations include: strong dependence on the available data sets, and limitations in simulations with extrapolated water depth values. KEY WORDS: Direct damage; flood hazard mapping; flood risk assessment; remote sensing
1. INTRODUCTION
In recent years, flood management has shifted from protection against floods to managing the risks of floods (European Flood Risk Directive).(2) To provide flood risk assessment, corresponding flood hazard and flood risk maps need to be elaborated. Flood risk is a function of two arguments: hazard probability and vulnerability.(2–11) In other words, risk is a mathematical expectation of vulnerability (consequences) function. Flood probabilities are determined in order to produce flood hazard maps. Hydraulic modeling of a peak flow is performed to derive critical flood characteristics such as water depth and flow velocity. These parameters are used as inputs to mortality and damage functions to estimate potential losses of life(5) and economical damages(7) due to floods. However, running hydraulic models faces many uncertainties(12) due to the lack of hydrological and other required data, their
Over last decades, we have witnessed an upward global trend in natural disaster occurrence. Hydrological and meteorological disasters are the main contributors to this pattern. In 2011, hydrological disasters, such as floods and wet mass movements, represented 52% of the overall disasters reported, causing 139.8 million victims and more than U.S. $70 billion in damages.(1) 1 Space
Research Institute NASU-SSAU, Kyiv, 03680, Ukraine. University of Life and Environmental Sciences of Ukraine, Kyiv, 03680, Ukraine. 3 National Technical University of Ukraine “Kyiv Polytechnic Institute,” Kyiv, 03056, Ukraine. ∗ Address correspondence to Sergii Skakun, Space Research Institute NASU-SSAU, Glushkov Ave., 40, build. 4/1, Kyiv, 03680, Ukraine; tel: +38-044-5262553; fax: +38-044-5264124; [email protected]. 2 National
1521
C 2013 Society for Risk Analysis 0272-4332/14/0100-1521$22.00/1
1522 incompleteness and imperfection.(2) The use of space-borne remote sensing data to flood risk mapping is a complement approach to the existing flood modeling techniques.(8,12–15) Remote sensing from space plays an important role in flood modeling and flood risk assessment.(11,13–21) Satellite images acquired in both optical and microwave range of the electro-magnetic emission are utilized for solving many problems related to flood risk management. The use of optical satellite imagery is usually limited by cloud cover, rain conditions, and nighttime. On the other hand, synthetic-aperture radar (SAR) measurements from space are independent of daytime and weather conditions. The satellite-derived flood extent(22–26) is invaluable for calibration and validation of hydraulic models.(12,27–31) Moreover, availability of flood extent maps in near real time (within 24 hours after satellite image acquisition(19,20,30,32–36) ) can greatly benefit rescuers during flooding and can also be used for rapid damage assessment. A combination of satellite-derived flood areas with a digital elevation model (DEM) allows indirect retrieval of water stages.(37–39) Water level observations derived from remote sensing imagery can be further assimilated into a hydrodynamic model to decrease forecast uncertainty.(40,41) Many studies have shown the efficiency of using satellite images for flood damage assessment.(42–47) Different types of damages caused by floods are mapped and assessed using remote sensing imagery: urban areas,(42,45) agriculture (crop) areas,(42,46) or general land cover types.(43,44,47) Satellite images can be directly used for flood hazard mapping. In Ref. 8, a novel approach for rapid flood risk mapping is proposed based on the use of radar satellite data. Flood extent maps derived from satellite imagery using different image processing techniques are fused to generate an event-specific flood hazard map. The flood risk map is further generated from the flood hazard map and vulnerabilityweighted land cover vector data. In Ref. 47, a flood hazard map is produced based on a flood frequency map retrieved from a time series of satellite images. Depending on the land cover and elevation each category is empirically assigned a weight score. The resulting flood hazard map derived from satellite imagery is compared to a flood depth map derived from a hydrodynamic model. It is reported that there is a 98% spatial agreement between two maps. In this article, the use of time series of satellite imagery to flood hazard mapping and flood risk as-
Skakun et al. sessment is presented. Flooded areas are extracted from each satellite image, and a maximum flood extent image for each flood event is generated. These maps are further integrated to determine relative frequency of inundation (RFI), which is used to provide flood risk assessment. The presented methodology is applied to the Katima Mulilo region of Namibia and has been used within the “Namibia SensorWeb Pilot Project.” This project aims at developing an operational transboundary flood management decision support system for the southern African region to provide useful flood and water-borne disease forecasting tools for local decisionmakers.(48,49) The remainder of the article is organized as follows. The method used for flood hazard mapping from a time series of satellite images and for flood risk assessment is outlined in Section 2. Results are presented in Section 3. A discussion of the obtained results and conclusions is given in Section 4.
2. METHODOLOGY AND TEST SITE DESCRIPTION 2.1. RFI Retrieval from a Time Series of Optical Satellite Images Let X = x(y, d) y∈Y,d∈Dy be a set of optical satellite images acquired over the given region. Note that Y denotes a set of years and Dy denotes days of the year y for which satellite imagery is available. The image x(y, d) is acquired on the day d of the year y. It is assumed that all satellite images are preprocessed for radiometric and geometric corrections, and clouds, shadows, and missing data are identified and masked for each image. Let u(y, d) be a flood extent image extracted from the satellite image x(y, d) using some classification algorithm. Within the classification procedure each pixel (i, j) of the image u(y, d) is assigned one of the following classes: u(y, d, i, j) = water/no water/no data. The “water” class is assigned to the pixels that are identified as flooded areas on the image x(y, d); the “no water” class is assigned to the pixels that are identified as dry areas on the image x(y, d); the “no data” class is reserved for the pixels that are identified as clouds, shadows, or missing data on the image x(y, d). Let Ey be a set of flood events during the year y, and Dey denotes days of the year y on which satellite imagery was acquired during flood event e ∈ Ey .
Flood Hazard and Flood Risk Assessment
1523
For each flood event e ∈ Ey , a maximum flood extent image u(e) is generated using the following equation: ⎧ e ⎪ ⎪ water, if d ∈ Dy exists suchas : u(y, d, i, j) = water, ⎪ ⎪ ⎨ no water, if for all d ∈ De : u(y, d, i, j) = no water, y u(e, i, j) = ⎪ no data, if for each d ∈ De : u(y, d, i, j) = ⎪ y ⎪ ⎪ ⎩ no water or no data. (1)
In other words, if a pixel is classified as “water” on at least one flood extent image u(y, d), then this pixel is assigned a “water” class on the resulting maximum flood extent image u(e) for the flood event e. If a pixel is classified as “no water” on every image acquired during the flood event e then it is assigned a “no water” class on the resulting maximum flood extent image u(e). And finally, the pixel is assigned a “no data” class if it is classified either a “no water” or “no data” class during the flood event e. This is done in such a way since it is unknown whether the “no data” pixel was actually flooded or not during the flood event in question. The derived maximum flood extent images u(e) are used to derive a RFI value: 1 1u(e,i, j)=water , (2) RFI(i, j) = |U| u(e)∈U
where U = {u(e) : y ∈ Y, e ∈ Ey and u(e, i, j) = water or no water}, and 1() is an indicator function. Therefore, for the region (i, j) only those flood extent images are taken into consideration for which this pixel is assigned either “water” or “no water.” The general diagram for determination of the RFI value is shown in Fig. 1. Let w(e) be a maximum water level recorded for the flood event e. Let w(1) < w(2) 0
Risk by damage categories, USD
7.850 1/30 Dwelling units Health facilities Schools Roads, m Dwelling units Health facilities Schools Roads
Total risk, USD
summarizes these estimates. RFI values are estimated for different water levels according to Equation (3) that represent 30-, 8.5-, 5.1-year floods.(51) Amount of damage categories with nonzero RFI values, estimated risk by categories, and a total risk
7.370 1/8.5 7,017 8 8 71,140 1,924,398 8,158 10,302 151,638 2 094,497
7.027 1/5.1 5,468 8 6 65,530 1,571,570 7,575 8,321 135,877 1 723,343
4,379 8 4 58,010 1,276,324 6,625 6,687 113,667 1 403,302
of damages (estimated using Equation (5)) are presented in Table V. Estimates for the 2009 flood (a 30-year flood) show that at least eight schools, eight health facilities, and 7,017 dwelling units were flooded in the Caprivi region. These assessment
1532
Skakun et al.
Fig. 9. Bed elevation profiles for cross-sections 1–4 derived from SRTM DEM at 90-m spatial resolution.
results correspond to the estimates reported in Ref. 51. About 71 km of roads (mainly district) are detected with RFI > 0, meaning they were flooded during 1989–2012 at least once. (Amount of roads that were flooded in the Caprivi region is not given in Ref. 51.) Within all scenarios considered, damages to dwelling units represent about 91% of total damages. Although direct damages to roads, health facilities, and schools are smaller compared to dwelling units, indirect damages should be taken into account to quantify risk more adequately. In particular, flooded and damaged roads will considerably reduce ability for timely evacuation and increase time necessary for providing assistance by rescuers. Flooded and damaged health facilities will reduce capacities for treating and providing assistance to affected people. It will therefore require people to be transported to nearby nonflooded health facilities, which coupled with damaged road networks, will reduce response
time. All these factors need to be also quantified and properly incorporated into the flood risk model (Equation (4)). The corresponding flood risk map for 2009 is generated, which allows identification of regions with the highest risk. For this, the region is divided into the grid of cells at 960 m × 960 m and a total risk is estimated for each cell according to Equation (5). The derived map is shown in Fig. 11. In general, the risk is distributed uniformly all over the region. The cities and villages with the highest risk are: Lisikili, Muzii, Kasika, Ivilivinzi, and Ngwese. Fig. 12 shows an example of flood risk map for the region of the village of Kasika. 4. DISCUSSION AND CONCLUSIONS Namibia, and the Caprivi region in particular, is prone to floods. A 2009 flood that was the largest
Flood Hazard and Flood Risk Assessment
1533
Fig. 10. Empirical CDF (panels A and B) and dependence between water depth and RFI (C and D) for cross-sections 1 (left panels) and 4 (right panels).
since 1969 and corresponded to a 30-year flood with 7.85-m water level was followed by moderate floods in 2010 and 2011. Flood risk assessment becomes an extremely important issue for the region from different perspectives: economical, social, environmental, health, etc. In this article, we take advantage of freely available archived satellite images and propose an approach for rapid flood hazard mapping and flood risk assessment in the Katima Mulilo region (Caprivi, Namibia). In particular, 97 Landsat-5 and Landsat-7 satellite images acquired from 1989 to 2012 are processed and analyzed to extract flood extent images that are further integrated to estimate relative RFI. The study shows that RFI values and water depth (normalized on maximum water level) exhibit the same probabilistic distribution, which is confirmed by the KS test. Furthermore, when analyzing river crosssections the RFI values can be considered as samples
(in general case irregular) of the water depth continues function. These sample values can be used to form a step-wise function that approximates relative water depth. It is shown that under the condition of regular samples the error of approximation will decrease as the number of samples increases. Therefore, the produced RFI map can be used as a flood hazard map, especially in cases when flood modeling is complicated by the lack of available data and high uncertainties. It should be, however, noted that only a coarse resolution DEM (90-m spatial resolution) was available for the Namibian case, and consequently the derived water depths were not very accurate. This should be taken cautiously into consideration when analyzing the relationship between RFI and relative water depth, and how accurately water depth can be restored from the RFI values. In our case
1534
Skakun et al.
Fig. 11. Flood risk for the Katima Mulilo region with values aggregated into cells of 960 m × 960 m.
Fig. 12. Flood hazard map (RFI) overlaid with dwelling units (left) and flood risk map (right) for the region of the village of Kasika (Caprivi, Namibia).
the obtained coefficient of determination of linear regression between RFI and relative water depth varied considerably (0.21–0.65), meaning there were regions with poor correspondence between these two parameters. Therefore, a more accurate DEM is nec-
essary to study more thoroughly a relationship between water depth and RFI. The derived RFI map is further used for flood risk assessment. The following direct damage categories are considered: dwelling units, roads, health
Flood Hazard and Flood Risk Assessment facilities, and schools. These categories accounted for more than 55% of total damages reported during a 2009 flood in the Caprivi region.(51) Though information on other categories of damages has not been available in this study, it provided useful information on the geographical distribution of regions with highest risk. It is estimated that 7,017 dwelling units, eight schools, eight health facilities, and 71 km of district roads in the Katima Mulilo regions were flooded in 2009, which corresponded to estimates provided in the “Post Disaster Needs Assessment 2009” report.(51) Dwelling units provided a major contribution (up to 91%) to the total estimated risk comparing to other damage categories considered. However, it should be noted that the risk associated with flooded and affected roads and health facilities should include not only direct damages but also indirect damages. These damages should account for reduced evacuation time and transportation capacities as well as reduced capacities for providing assistance and treatment to people affected. This is one of the areas where the proposed model needs to be improved in the future. Another issue that should be taken into consideration is that in our case flood damages were estimated using only historical data for the single flood event reported in Ref. 51. Detailed damage estimates on other flood events in the region were not available, to our best knowledge. This did not allow us to provide statistically consistent estimates of flood damages. (By statistically consistent estimates, we mean estimates that converge in probability to the true value.) Therefore, more historical data on flood damages (e.g., from neighbor countries) are necessary to incorporate them into the proposed flood risk model. The presented approach for flood hazard mapping using time series of satellite data has the following advantages. It has minimum data requirements, and RFI maps can be generated rapidly to assist rescuers and decisionmakers in case of emergencies. These maps can be further used for rapid flood risk assessment to identify regions with the highest risk. Furthermore, the derived flood extent maps can be used to calibrate and validate hydrological models once high-resolution topographic data become available. On the other hand, the presented approach has certain limitations. It is a data-driven methodology and, therefore, the quality of the derived RFI maps is dependent on the available satellite images. The proposed approach can be hardly applied for predicting flood inundated areas with installed weirs. While
1535 within the proposed approach the RFI values can be used to interpolate water depth, they cannot be used to extrapolate water depth. In other words, it is difficult to apply the proposed methodology to simulate scenarios with the water depth values more than the maximum water depth value available in remote sensing observations. Other case studies should be considered to assess efficiency of the presented approach for other regions and under different conditions (hydrometeorological, quality, and availability of satellite data, etc.). REFERENCES 1. Guha-Sapir D, Vos F, Below R, Ponserre S. Annual Disaster Statistical Review 2011: The Numbers and Trends. Brussels: Centre for Research on the Epidemiology of Disasters (CRED), 2012. 2. Mostert E, Junier SJ. The European flood risk directive: Challenges for research. Hydrology and Earth System Sciences Discussion, 2009; 6(4):4961–4988. 3. Jonkman SN, van Gelder PHAJM, Vrijling JK. An overview of quantitative risk measures for loss of life and economic damage. Journal of Hazardous Materials, 2003; A99:1–30. 4. Hoes O, Schuurmans W. Flood standards or risk analyses for polder management in the Netherlands. Journal of Irrigation and Drainage Engineering, 2006; 55:113–119. 5. Jonkman SN. Loss of life estimation in flood risk assessment, PhD thesis, Delft University, 2007. 6. Jonkman SN, Kok M, Vrijling JK. Flood risk assessment in the Netherlands: A case study for dike ring South Holland. Risk Analysis, 2008; 28(5):1357–1374. 7. Jonkman SN, Boˇckarjova M, Kok M, Bernardini P. Integrated hydrodynamic and economic modelling of flood damage in the Netherlands. Ecological Economics, 2008; 66:77–90. 8. Schumann G, Di Baldassarre G. The direct use of radar satellites for event-specific flood risk mapping. Remote Sensing Letters, 2010; 1(2):75–84. 9. Kussul NN, Sokolov BV, Zyelyk YI, Zelentsov VA, Skakun SV, Shelestov AYu. Disaster risk assessment based on heterogeneous geospatial information. Journal of Automation and Information Science, 2010; 42(12):32–45. 10. Kellens W, Terpstra T, De Maeyer P. Perception and communication of flood risks: A systematic review of empirical research. Risk Analysis, 2013; 33:24–49. 11. Taubenbock H, Post J, Roth A, Zosseder K, Strunz G, Dech S. A conceptual vulnerability and risk framework as outline to identify capabilities of remote sensing. Natural Hazards and Earth System Sciences, 2008; 8:409–420. 12. Horritt MS. A methodology for the validation of uncertain flood inundation models. Journal of Hydrology, 2006; 326:153– 165. 13. Bates PD, Horritt MS, Smith CN, Mason DC. Integrating remote sensing observations of flood hydrology and hydraulic modelling. Hydrological Processes, 1997; 11:1777–1795. 14. Bates PD. Invited commentary: Remote sensing and flood inundation modelling. Hydrological Processes, 2004; 18:2593– 2597. 15. Lecca G, Petitdidier M, Hluchy L, Ivanovic M, Kussul N, Ray N, Thieron V. Grid computing technology for hydrological applications. Journal of Hydrology, 2011; 403(1–2):186–199. 16. Voigt S, Kemper T, Riedlinger T, Kiefl R, Scholte K, Mehl H. Satellite image analysis for disaster and crisis management
1536
17.
18. 19. 20. 21. 22.
23.
24.
25. 26.
27. 28. 29.
30.
31.
32. 33. 34. 35.
support. IEEE Transactions on Geoscience and Remote Sensing, 2007; 45(6):1520–1528. Schumann G, Bates PD, Horritt MS, Matgen P, Pappenberger F. Progress in integration of remote sensing–derived flood extent and stage data and hydraulic models. Reviews of Geophysics, 2009; 47(4): Art. RG4001. Stryker T, Jones B. Disaster response and the international charter program. Photogrammetric Engineering & Remote Sensing, 2009; 75(12):1342–1344. Kussul N, Shelestov A, Skakun S. Grid and sensor web technologies for environmental monitoring. Earth Science Informatics, 2009; 2(1–2):37–51. Cossu R, Schoepfer E, Bally P, Fusco L. Near real-time SARbased processing to support flood monitoring. Journal of RealTime Image Processing, 2009; 4:205–218. Guo H. Understanding global natural disasters and the role of earth observation. International Journal of Digital Earth, 2010; 3(3):221–230. Kussul NN, Lupian EA, Shelestov AY, Skakun SV, Tishchenko YG, Hluchy L. Determination of inundated territories on the basis of integration of heterogeneous data. Journal of Automation and Information Science, 2007; 39(12): 42–51. Kravchenko AN, Kussul NN, Lupian EA, Savorsky VP, Hluchy L, Shelestov AY. Water resource quality monitoring using heterogeneous data and high-performance computations. Cybernetics and Systems Analysis, 2008; 44(4):616–624. Schumann G, Di Baldassarre G, Bates PD. The utility of spaceborne radar to render flood inundation maps based on multialgorithm ensembles. IEEE Transactions on Geoscience and Remote Sensing, 2009; 47(8):2801–2807. Skakun S. A neural network approach to flood mapping using satellite imagery. Computing and Informatics, 2010; 29(6):1013–1024. Pulvirenti L, Chini M, Pierdicca N, Guerriero L, Ferrazzoli P. Flood monitoring using multi-temporal COSMO-SkyMed data: Image segmentation and signature interpretation. Remote Sensing of Environment, 2011; 115(4):990–1002. Horritt MS, Bates PD. Evaluation of 1D and 2D numerical models for predicting river flood inundation. Journal of Hydrology, 2002; 268:87–99. Di Baldassarre G, Schumann GJ-P, Bates PD. Near real time satellite imagery to support and verify timely flood modelling. Hydrological Processes, 2009; 23:799–803. Di Baldassarre G, Schumann GJ-P, Bates PD. A technique for the calibration of hydraulic models using uncertain satellite observations of flood extent. Journal of Hydrology, 2009; 367:276–282. Di Baldassarre G, Schumann GJ-P, Brandimarte L, Bates PD. Timely low resolution SAR imagery to support floodplain modelling: A case study review. Surveys in Geophysics, 2011; 32:255–69. Khan SI, Yang H, Wang J, Yilmaz KK, Gourley JJ, Adler RF, Brakenridge GR, Policelli F, Habib S, Irwin D. Satellite remote sensing and hydrologic modeling for flood inundation mapping in Lake Victoria basin: Implications for hydrologic prediction in ungauged basins. IEEE Transactions on Geoscience and Remote Sensing, 2011; 49(1):85–95. Bakan GM, Kussul NN. Fuzzy ellipsoidal filtering algorithm of static object state. Problemy Upravleniya I Informatiki (Avtomatika), 1996; 5:77–92. Shelestov A, Kussul N, Skakun S. Grid technologies in monitoring systems based on satellite data. Journal of Automation and Information Science, 2006; 38(3):69–80. Shelestov A, Kussul N. Using the fuzzy-ellipsoid method for robust estimation of the state of a grid system node. Cybernetics and Systems Analysis, 2008; 44(6):847–854. Kussul N, Shelestov A, Skakun S, Li G, Kussul O. The wide area grid testbed for flood monitoring using earth observation
Skakun et al.
36.
37.
38.
39. 40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2012; 5(6):1746–1751. Kussul N, Shelestov A, Skakun S, Kravchenko O, Gripich Y, Hluchy´ L, Kopp P, Lupian E. The data fusion grid infrastructure: Project objectives and achievements. Computing and Informatics, 2010; 29(2):319–334. Schumann G, Matgen P, Pappenberger F. Conditioning water stages from satellite imagery on uncertain data points. Geoscience and Remote Sensing Letters, 2008; 5:810– 813. Matgen P, Schumann G, Henry J, Hoffmann L, Pfister L. Integration of SAR-derived inundation areas, high precision topographic data and a river flow model toward real-time flood management. International Journal of Applied Earth Observation and Geoinformation, 2007; 9(3):247–263. Mason DC, Bates PD, Dall’Amico JT. Calibration of uncertain flood inundation models using remotely sensed water levels. Journal of Hydrology, 2009; 368:224–236. Garc´ıa-Pintado J, Neal JC, Mason DC, Dance SL, Bates PD. Scheduling satellite-based SAR acquisition for sequential assimilation of water level observations into flood modelling. Journal of Hydrology, 2013; 495:252–266. Giustarini L, Matgen P, Hostache R, Montanari M, Plaza D, Pauwels VRN, De Lannoy GJM, De Keyser R, Pfister L, Hoffmann L, Savenije HHG. Assimilating SAR-derived water level data into a hydraulic model: A case study. Hydrology and Earth System Sciences, 2011; 15:2349–2365. Qi H, Altinakar MS. Simulation-based decision support system for flood damage assessment under uncertainty using remote sensing and census block information. Natural Hazards, 2011 Nov; 59(2):1125–1143. van der Sande CJ, de Jong SM, de Roo APJ. A segmentation and classification approach of IKONOS-2 imagery for land cover mapping to assist flood risk and flood damage assessment. International Journal of Applied Earth Observation and Geoinformation, 2003; 4(3):217–229. Serpico SB, Dellepiane S, Boni G, Moser G, Angiati E, Rudari R. Information extraction from remote sensing images for flood monitoring and damage evaluation. Proceedings of the IEEE, 2012; 100(10): 2946–2970. Taubenbock H, Wurm M, Netzband M, Zwenzner H, Roth A, Rahman A, Dech S. Flood risks in urbanized areas—Multisensoral approaches using remotely sensed data for risk assessment. Natural Hazards and Earth System Sciences, 2011; 11:431–444. Tapia-Silva F-O, Itzerott S, Foerster S, Kuhlmann B, Kreibich H. Estimation of flood losses to agricultural crops using remote sensing. Physics and Chemistry of the Earth, Parts A/B/C, 2011; 36(7–8):253–265. Dewan AM, Kumamoto T, Nishigaki M. Flood hazard delineation in greater Dhaka, Bangladesh using an integrated GIS and remote sensing approach. Geocarto International, 2006; 21(2):33–38. Kussul N, Mandl D, Moe K, Mund JP, Post J, Shelestov A, Skakun S, Szarzynski J, Van Langenhove G, Handy M. Interoperable infrastructure for flood monitoring: SensorWeb, Grid and Cloud. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2012; 5(6): 1740– 1745. Mandl D, Frye S, Cappelaere P, Handy M, Policelli F, Katjizeu M, Van Langenhove G, Aube G, Saulnier J, Sohlberg R, Silva JA, Kussul N, Skakun S, Ungar SG, Grossman R, Szarzynski J. Use of the earth observing one (EO-1) satellite for the Namibia SensorWeb flood early warning pilot. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2013; 6(2): 298–308. Vapnik VN. An overview of statistical learning theory. IEEE Transactions on Neural Networks, 1999; 10(5):988– 999.
Flood Hazard and Flood Risk Assessment 51. Post Disaster Needs Assessment 2009. A Report Prepared by the Government of the Republic of Namibia, with support from the International Community, 2009. Available at: http://siteresources.worldbank.org/INTAFRICA/Resources/ WorldBankNamibiaReportFAlr.pdf, Accessed July 12, 2013. 52. Van Langenhove G. Hydro-meteorological extremes—Floods and droughts in Namibia [Internet]. Climate Change Capacity Building Workshop, 2012 October 09–11, Windhoek, Namibia. Available at: http://www.sarva.org.za/sadc/ download/namibia2012 11.pdf, Accessed on July 30, 2013.
1537 53. Chander G, Markham BL, Helder DL. Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. Remote Sensing of Environment, 2009; 113:893–903. 54. Frazier PS, Page KJ. Water body detection and delineation with landsat TM data. Photogrammetric Engineering & Remote Sensing, 2000; 66(12):1461–1467. ¨ 55. Aldroubi A, Grochenig K. Nonuniform sampling and reconstruction in shift-invariant spaces. SIAM Review, 2001; 43(4):585–620.
