Giustarini, L. et al. Probabilistic Flood Mapping Using Synthetic Aperture Radar Data. Ieee T Geosci Remote 54, 6958–6969 (2016).

Matgen et al. [29] introduced a SAR-based flood mapping technique that combines thresholding, region growing, and change detection, benefitting from the respective strengths of the specific processes, while avoiding their individual shortcomings. The technique in [29] offered comparable performance with benchmark methods despite not requiring any ground truth data for optimizing the parameters of the different processes

In a follow-up study, [6] extended the method to make it fully automated and demonstrated its capability to efficiently retrieve flood extent from images with a broad range of spatial resolutions. They evaluated the method’s performance in the specific setting of a built-up environment. A remaining weakness of this algorithm is its reliance on an observable bimodality in the histogram of backscatter values in the image [30].

  1. Giustarini, L. et al. A Change Detection Approach to Flood Mapping in Urban Areas Using TerraSAR-X. Ieee T Geosci Remote 51, 2417–2430 (2013).
  1. Pulvirenti, L., Marzano, F. S., Pierdicca, N., Mori, S. & Chini, M. Discrimination of Water Surfaces, Heavy Rainfall, and Wet Snow Using COSMO-SkyMed Observations of Severe Weather Events. Ieee T Geosci Remote 52, 858–869 (2014).

In a first attempt to overcome this shortcoming, [31] proposed an alternative procedure for core water body identification.

  1. Lu, J. et al. Automated flood detection with improved robustness and efficiency using multi-temporal SAR data. Remote Sens Lett 5, 240–248 (2014).

More recently, building on the work in [6] and [29], [32] has introduced a hierarchical split-based approach (HSBA), in order to improve the robustness and accuracy of the flood mapping algorithm. In a given SAR image, HSBA first searches for tiles, of potentially different sizes, whose corresponding histograms of backscatter values show an observable bimodality. It then uses a flood mapping approach named HSBA-Flood to retrieve flood extent over the entire given SAR image.

  1. Chini, M. et al. An automatic SAR-based flood mapping algorithm combining hierarchical tiling and change detection

32はどんだけ探してもネットでみつからないので、多分こっち

  1. Chini, M., Hostache, R., Giustarini, L. & Matgen, P. A Hierarchical Split-Based Approach for Parametric Thresholding of SAR Images: Flood Inundation as a Test Case. Ieee T Geosci Remote 55, 6975–6988 (2017).

In addition to enabling the mapping of flood extent in near real time, the EO-derived information shows high potential for applications in hydraulic modeling, particularly for model calibration (e.g., [33]–[37]) and evaluation (e.g., [38] and [39]).

For evaluation of hydraulic models

  1. Paz, A. R. da, Collischonn, W., Tucci, C. E. M. & Padovani, C. R. Large‐scale modelling of channel flow and floodplain inundation dynamics and its application to the Pantanal (Brazil). Hydrol Process 25, 1498–1516 (2011).
  1. Yamazaki, D., Kanae, S., Kim, H. & Oki, T. A physically based description of floodplain inundation dynamics in a global river routing model. Water Resour Res 47, (2011).

For calibration of hydraulic models

  1. Wood, M. et al. Calibration of channel depth and friction parameters in the LISFLOOD-FP hydraulic model using medium- resolution SAR data and identifiability techniques. Hydrol Earth Syst Sc 20, 4983–4997 (2016).
  1. Domeneghetti, A. et al. The use of remote sensing-derived water surface data for hydraulic model calibration. Remote Sens Environ 149, 130–141 (2014).
  1. Tarpanelli, A., Brocca, L., Melone, F. & Moramarco, T. Hydraulic modelling calibration in small rivers by using coarse resolution synthetic aperture radar imagery. Hydrol Process 27, 1321–1330 (2013).
  1. Pappenberger, F., Frodsham, K., Beven, K., Romanowicz, R. & Matgen, P. Fuzzy set approach to calibrating distributed flood inundation models using remote sensing observations. Hydrol Earth Syst Sc 11, 739–752 (2007).
  1. Baldassarre, G. D., Schumann, G. & Bates, P. D. A technique for the calibration of hydraulic models using uncertain satellite observations of flood extent. J Hydrol 367, 276–282 (2009).

In the last decade, several studies used SAR-derived water levels for calibrating uncertain hydraulic model parameters (e.g., [3] and [12]).

  1. Hostache, R. et al. Water Level Estimation and Reduction of Hydraulic Model Calibration Uncertainties Using Satellite SAR Images of Floods. Ieee T Geosci Remote 47, 431–441 (2009).
  1. Schumann, G. et al. High-Resolution 3-D Flood Information from Radar Imagery for Flood Hazard Management. Ieee T Geosci Remote 45, 1715–1725 (2007).

Moreover, sequential assimilation of remote-sensing-derived water levels into flood forecasting systems has attracted interest in recent years, with several proof-of-concept studies demonstrating the ability of these data sets to improve model predictions [22], [40]–[44]. The aforementioned methods all make use of water levels, which can be estimated indirectly along the flood edges derived from SAR images by intersecting the shorelines with floodplain topography. Extracting water levels represents an intermediate step before assimilation of data derived from EO can be performed: this stage adds the digital terrain model (DTM) uncertainty to the observation un-certainty. In this case, the extraction of water levels represents an additional step in the processing chain leading toward the assimilation of EO data into numerical models. It goes without saying that this processing step adds the uncertainty of the DTM and that of the water level extraction procedure to the uncertainty associated with the SAR data and the flood extent processing algorithm.

  1. Mason, D. C., Schumann, G. J.-P., Neal, J. C., Garcia-Pintado, J. & Bates, P. D. Automatic near real-time selection of flood water levels from high resolution Synthetic Aperture Radar images for assimilation into hydraulic models: A case study. Remote Sens Environ 124, 705–716 (2012).
  1. Matgen, P. et al. Towards the sequential assimilation of SAR-derived water stages into hydraulic models using the Particle Filter: proof of concept. Hydrol Earth Syst Sc 14, 1773–1785 (2010).
  1. Hostache, R., Lai, X., Monnier, J. & Puech, C. Assimilation of spatially distributed water levels into a shallow-water flood model. Part II: Use of a remote sensing image of Mosel River. J Hydrol 390, 257–268 (2010).
  1. Giustarini, L. et al. Assimilating SAR-derived water level data into a hydraulic model: a case study. Hydrol Earth Syst Sc 15, 2349–2365 (2011).
  1. García-Pintado, J., Neal, J. C., Mason, D. C., Dance, S. L. & Bates, P. D. Scheduling satellite-based SAR acquisition for sequential assimilation of water level observations into flood modelling. J Hydrol 495, 252–266 (2013).
  1. García-Pintado, J. et al. Satellite-supported flood forecasting in river networks: A real case study. J Hydrol 523, 706–724 (2015).

The direct assimilation of flood extent maps into hydrodynamic models [45] would thus be beneficial because it would eliminate the intermediate step of having to extract water levels first. In a framework of data assimilation, uncertainties associated with observations have to be quantified in order to ensure an optimal use of the assimilation filters. However, most SAR image processing methods provide flood extent estimates in the form of binary maps. While binary maps are already useful to insurance agencies, decision makers, and humanitarian relief organizations, they do not offer any indication on the uncertainty associated with the pixel classification. As such, each pixel is classified as either flooded or non-flooded, without any uncertainty to characterize its state. A more informative alternative is offered by probabilistic flood maps, where the estimated status of any given pixel is represented by a probability value in the continuous range [0, 1]. Applications of such probabilistic flood maps are not restricted to data assimilation. Indeed, a map displaying the uncertainty in the predicted state of each pixel represents a valuable source of information in its own right, as it reflects the level of confidence that can be attributed to the mapped flood extent.

  1. Matgen, G. Corato, M. Chini, R. Hostache, and L. Giustarini, “Improved flood forecasting through the assimilation of SAR-derived flood probability maps into 2D hydrodynamic models,” presented at the Proceedings IGARSS, Milan, Italy, Jul. 26–31, 2015.

With the exception of the aforementioned study in [28], previous studies on the characterization of uncertainties in flood extent maps have been generally based on random realizations of potential sources of uncertainty (e.g., [3], [11], [34], and [46]–[48]).

  1. Hostache, R. et al. Water Level Estimation and Reduction of Hydraulic Model Calibration Uncertainties Using Satellite SAR Images of Floods. Ieee T Geosci Remote 47, 431–441 (2009).
  1. Baldassarre, G. D., Schumann, G. & Bates, P. D. A technique for the calibration of hydraulic models using uncertain satellite observations of flood extent. J Hydrol 367, 276–282 (2009).
  1. Schumann, G., Matgen, P. & Pappenberger, F. Conditioning Water Stages from Satellite Imagery on Uncertain Data Points. Ieee Geosci Remote S 5, 810–813 (2008).
  1. Refice, A. et al. SAR AND INSAR FOR FLOOD MONITORING: EXAMPLES WITH COSMO/SKYMED DATA. 2013 Ieee Int Geoscience Remote Sens Symposium - Igarss 703–706 (2013) doi:10.1109/igarss.2013.6721254.
  1. Giustarini, L. et al. Accounting for image uncertainty in SAR-based flood mapping. Int J Appl Earth Obs 34, 70–77 (2015).

Unfortunately, the number of realizations and the procedure to characterize uncertainties tend to be rather subjective. For example,[46] investigated uncertainty in SAR-derived water stages, for a single SAR image and a single flood mapping procedure, and identified two main sources of uncertainty: 1) the parameter value applied to classify a pixel as flooded (i.e., flooded/ non-flooded classification threshold) and 2) geocoding of the image itself.


They tested four different threshold values and 50 image geocodings to obtain an ensemble of binary flood maps and corresponding SAR-derived water levels. In a similar study, [34] considered uncertainty stemming from the available SAR image and the applied classification procedure. They computed ten binary flood maps by combining two available SAR images that were acquired at nearly the same time but having different resolutions, with five different flood mapping procedures. These case studies show that it is important, as well as far from trivial, to correctly and objectively quantify uncertainty in flood mapping.


More recently, [47] has implicitly introduced a semiautomated approach that allows integrating ancillary information to derive a posteriori probabilistic maps of flood inundation, accounting for different scattering responses to the presence of water. Giustarini et al. [48] proposed the use of a nonparametric bootstrap method to address speckle uncertainty. While focusing only on the specific component of the total uncertainty that derives from speckle, they proposed a methodology to objectively determine the minimum number of realizations capable of taking into account speckle uncertainty. However, for a reliable assessment of flood mapping uncertainty and a subsequent successful data assimilation, a probabilistic map would need to take into account all uncertainty components.

The aim of this paper is to introduce a statistical framework for estimating uncertainty associated with SAR-derived flood extent maps. The method is based on the combination of Bayes’ inference mixture modeling and the algorithm in [32]. The approach in [32] is used to decompose the image pixel histogram of backscatter values into two populations: one corresponding to flooded (open water) pixels and the other corresponding to non-flooded pixels. Then, instead of computing a binary flood map by applying thresholding, region growing, and change detection, the algorithm here proposed produces a probabilistic flood map using the estimated backscatter distributions of the two pixel populations. The approach takes into account image uncertainty by using the image histogram to estimate the probability of each pixel to be flooded or non-flooded, based on its backscatter value.

  1. Matgen, P. et al. Towards an automated SAR-based flood monitoring system: Lessons learned from two case studies. Phys Chem Earth Parts B C 36, 241–252 (2011).

The major difference between HSBA-Flood and the algorithm in [6] is that HSBA-Flood optimizes its flood mapping parameters (thresholding, region growing, and change detection) only using pixels in the tiles selected as showing an identifiable bimodality and, later, applies these parameters to the entire SAR image, in order to derive a binary flood map. The work in [32] thus represents an extension of the algorithm in [6] and allows for a robust flood mapping approach, regardless of image size, sensor characteristics, and overall histogram bimodality.

In the case of log-transformed SAR intensity data, this population can be approximated by a Gaussian distribution [49], [50].

  1. Xie, H., Pierce, L. E. & Ulaby, F. T. Statistical Properties of Logarithmically Transformed Speckle. Ieee T Geosci Remote 40, 721–727 (2002).
  1. F. Ulaby, R. Moore, and A. Fung,Microwave Remote Sensing, Active andPassive, Volume Scattering and Emission Theory, Advances System andApplications,