floodOBIA

An object-based clustering approach for SAR-based flood mapping.

This repository contains the source codes used in the following publication:

Landuyt et al., (in review). "Flood mapping in vegetated areas using an unsupervised clustering approach on Sentinel-1 and -2 imagery". Remote Sensing.

The code itself can be referenced by the following DOI:

Prerequisites

This repository is mainly built using the following packages:

python 3.8.2
rasterio 1.1.7
geopandas 0.8.1
pygeos 0.8
rasterstats 0.15
scikit-image 0.17.1

Run the code

The main file is OBIAflood.py. This script requires the path to a .csv file specifying the settings as an argument. An example settings file, Settings.csv, is included.

Input

A description of all parameters is given in the third column of Settings.csv

The main input files are:

• A stack of 2 SAR images, both in VV and VH polarization. The following band order is assumed: reference_VH, reference_VV, flood_VH, flood_VV.
• A stack of optical bands (cloudfree image or composite). The following band order is assumed: B2, B3, B4, B5, B6, B7, B8, B8A, B11, B12.
• DEM of the ROI. In the paper, the hydrologically conditioned SRTM DEM was used.
• LC of the ROI. The code assumes Copernicus GLS land cover codes.

Output

The resulting classification is saved both as a .shp and a .tif file. In the .shp, the attribute "model" contains the final classification. In the final classification, the following classes (with label) are present:

• 0 = dry land (DL)
• 1 = permanent water (PW)
• 2 = open flooding (OF)
• 3 = flooded vegetation (FV)
• 4 = low-lying areas considered as OF
• 5 = forests possibly hiding flood