ecop-shoreline

Shoreline delineation module developed in the framework of ECOPOTENTIAL H2020 project

Shoreline detection module

The shoreline detection algorithm has been implemented by Starlab in python v2.7 as a separate module of the whole chain of processing. This allows us to provide such module for its integration in the ECOPOTENTIAL virtual laboratory.

Dependencies

For an easy installation of the necessary libraries:

pip install -r requirements.txt

Input

The input data to the shoreline detection algorithm are Sentinel-1 Interferometric Wide (IW) GRDH dual polarization (VH and VV) images, which have ALREADY undergone the typical SAR pre-processing phase, including calibration and range doppler terrain correction. At Starlab we implemented our pre-processing chain but we will not provide it, since other ECOPOTENTIAL applications using Sentinel-1 data should also use such algorithms that therefore should be implemented separately and made available to all partners (and users) to run their modules.

Shoreline module functions definitions

The shoreline detection module includes two callable functions.

Each Sentinel-1 band is firstly processed separately by using the function:

def waterbody(band, logarithmic = True, clipping = [5,98],  filter_kernel_size = 25, opening_kernel_size = 7, closing_kernel_size = 7, fill_holes = True, sand_max_gray_level = 5)

where:

band: Sentinel-1 VV or VH polarization band, as float32 numpy array

The users can set up the parameters:

logarithmic (example used for Curonian Lagoon: true): if true a logarithmic intensity rescaling is applied to the image; otherwise a linear rescaling is applied.
clipping (example used for Curonian Lagoon: [5, 98]): lower and upper limits used by the percentile function used in the intensity rescaling step.
filter_kernel_size (example used for Curonian Lagoon: 21): size of the median filter applied for noise reduction.
opening_kernel_size (example used for Curonian Lagoon: 5): size of the kernel used in the opening morphological operation.
closing_kernel_size (example used for Curonian Lagoon: 5): size of the kernel used in the closing morphological operation.
fill_holes (example used for Curonian Lagoon: true): if true the algorithm fills all gaps in land, regardless their size; otherwise only small gaps are closed.
sand_max_gray_level (example used for Curonian Lagoon: 0): threshold used to label as land those sand banks pixels having a grey level below such value.

waterbody output

The output of the waterbody function are, for each polarization:

water mask (uint8 numpy array)
water edges image (uint8 numpy array)
ancillary data

ancillary = {"scaled": img, "denoised": blur, "rawland": raw_land_mask}

where:

Scaled: image after applying the intensity rescaling (uint8 numpy array)
Denoised: image after applying the denoising filter (uint8 numpy array)
Rawland: raw land mask before applying morphological operation (uint8 numpy array).

The Rawland images (one for each polarization) correspond to band 1 and band 2 in the second function:

def waterbodydp(band1, band2, opening_kernel_size = 7,  closing_kernel_size = 7, fill_holes = True)

The waterbodydp function combines the results obtained by processing each band in the previous step.

In order to run this function, the parameter "dual_pol" must be set to "true" in the configuration file.

The editable parameters are the same used in the first step. We keep the same values also for the second function.

waterbodydp output

The output of the waterbodydp function are:

water mask (uint8 numpy array)
water edges image (uint8 numpy array)

Note

Please note that the input and output of both waterbody and waterbodydp functions are numpy arrays which do not provide geographic information. A dedicated function attaching the geographic information should be made available in the virtual laboratory, and called at the end of the waterbody(dp) module processing.

starlab-bcn/ecop-shoreline