HASARD Single Image service specifications

This service derives a binary mask of standing water, which can be used to identify potential flood-affected regions. It takes as input the Sigma Nought single band asset of a radar calibrated dataset. The service supports input from the Sentinel-1, RCM, TerraSAR-X, TanDEM-X, and SAOCOM-1 missions.

The tutorial of the HASARD Single Image service is available in this section.

Service Description

The HASARD Single Image service derives a map of satellite detected surface waters from a single co-pol backscatter image. It supports backscatter images from multiple missions. This service relies on the “HASARD”¹ flood mapping algorithm that enables systematic, automatic and reliable SAR mapping of terrestrial water bodies. HASARD implements the Hierarchical Split-Based Approach (HSBA), described by Chini et al. 2017², and is specifically designed to support flooding-related disaster risk reduction at global scale. The underlying end-to-end image processing chain is based on a patented technology from the Luxembourg Institute of Science and Technology (LIST)³ that combines and fully integrates innovative hierarchical image splitting and open water backscatter modelling. The service was developed by LIST and WASDI.

The HASARD Single Image service derives an accurate standing water map from a single satellite image (Sigma Nought asset). In this computation the service employs the Height Above the Nearest Drainage (HAND) dataset to exclude false alarms in slope areas due to layover. HAND auxiliary data is taken from the single band asset hand included into the HAND Auxiliary Dataset, which is derived from the MERIT-Hydro dataset ⁴,⁵. HASARD Single Image generates as output a water mask overview and a single band assets both in COG format. The service also returns as output the single band backscatter asset and the HAND asset employed as input in the computation.

Workflow

The HASARD Single Image processing service applies the workflow below:

Input

The input of the HASARD Single Image service is a single band Sigma0 asset in dB, originating from a Calibrated Dataset [CD], and the single band asset from the HAND Auxiliary Dataset. Thus, to employ the service you need the following two single band assets:

1. A co-pol Sigma Nought asset in dB (e.g. s0_db_x_hh for TerraSAR-X) from a SAR Calibrated Dataset acquired during crisis,

2. the hand single band asset providing relative elevations in m included into the HAND Auxiliary Dataset.

Parameters

The HASARD Single Image service requires a specified number of mandatory and optional parameters. All service parameters are listed in the below Table 1.

Table 1 - Service parameters for the HASARD Single Image processor.

Input co-pol sigma nought single band asset

The first mandatory parameter is to specify the single band asset that shall be used as input for the water mapping computation (Co-pol Sigma Nought in dB).

Input HAND single band asset

For the second mandatory parameter, the user must specify the single band asset from the HAND Auxiliary Dataset to be used for the water mapping computation.

HSBA parameters

The following are mandatory parameters three through six:

• Starting tiling level,

• Check bimodal distribution (Ashman D),

• Minimum number of pixels per tile,

• Number of pixels for small object removal

These are the parameters of the retrieval algorithm. Default values are given in Table 1 and are the ones recommended for a generic use of the algorithm.

Starting tiling level

The HSBA consists of two main steps. First, the original image is split into separate subregions (i.e., tiles) of different sizes. Later, all the tiles showing a clear bimodal behavior with two Gaussian balanced populations are selected for threshold determination. In HSBA, the hierarchical tiling of the image is done using a quad- tree decomposition, which consists of iteratively decomposing image regions into four equally sized quadrants (i.e., tiles), the so-called subquadrants. When the level is equal to 1 the image is split into 4 tiles, equal to 2 in 16 tiles, equal to 3 in 64 tiles, etc Default value is 0.

Check bimodal distribution (Ashman D)

The 2nd of the four mandatory HSBA parameters is dedicated to check the bimodal distribution. The Ashman D (AD) coefficient quantifies how well two Gaussian distributions are separated by considering the distance between their mean values and their dispersions, i.e., standard deviations. For a mixture of two Gaussian distributions, AD > 2 is required for a clear separation of the distributions. Default value is 2.5.

Minimum number of pixels per tile

HSBA searches for tiles of variable size allowing the parametrization of the distributions of two classes. The size of the tiles is not fixed a priori, but is automatically adapted to the information content retrieved from the image. When selecting bimodal tiles at each level the algorithm discard the ones having a low number of pixels. With this parameter the user can define the minimum value of pixels per tile to be considered in the selection of bimodal tiles. Default value is 4000.

Number of pixels for small object removal

The algorithm includes a spatial filtering of the change detection binary mask to remove small isolated object units of water pixels. With this parameter the user can specify the number of pixels composing the smaller object unit. Default value is 20.

HAND threshold

In the seventh mandatory parameter the user must specify a relative elevation threshold to be considered when mapping water extent. Pixels classified as water that have values of Height Above the Nearest Drainage (HAND) above this threshold in meters (e.g. 15m) are masked out. In particular, this threshold is used to derive a binary mask of potential floodplains from the HAND dataset which is then applied for masking the water extent.

where((hand<=15),1,0);

Water backscatter threshold

In the last mandatory parameter the user must insert a threshold value in decibel (e.g. -20 db) to be used for the computation of water extent from the backscatter signal. The user shall choose an optimum value of Sigma Nought in dB for the image thresholding (e.g. the value that separates the bimodal distribution).

A good practice is to assign a value that is 3 or 4 decibels higher than the estimated backscattering value of water, taking into account the frequency, polarisation, and incidence angle of the available SAR image. Default value in dB for the water backscatter threshold is -19.

Output

The HASARD Single Image processor provides as output the following products:

1. Water extent overview,

2. Water extent single band asset,

3. Input co-pol sigma nought single band asset,

4. Input HAND single band asset co-located to the sigma nought.

HASARD Single Image products specifications can be found in the below tables.

Vectorize HASARD water mask single band asset

HASARD-SI's binary water mask can be spatially filtered and / or converted to a vector polygon using the FilterVectorize service of the ESA Charter Mapper.

To polygonize the water-mask single band asset employ the FilterVectorize service in Vectorize mode by selecting only true values DN=1 (water).

If unsatisfied with the spatial filtering offered within the HASARD Single Image service, given the Number of pixels for small object removal parameter, the FilterVectorize service can be used to further post-process the binary mask and remove small isolated clusters of water pixels. Thus, to obtain a further spatially filtered water extent vector, employ the FilterVectorize service in the Filter and Vectorize mode.