Water Bodies Automatic Classification

Contains modified Copernicus Sentinel data (2016), processed by ESA, CC BY-SA 3.0 IGO

The aim of this project is to provide a tool for flood hazard analysis, by automatically extracting information on the seasonality of water bodies and scenarios of historical flooded areas.

This goal is achieved through the generation and analysis of an historical archive of water bodies, derived by means of an automatic procedure for the extraction of flooded areas from satellite data. MODIS (Moderate Resolution Imaging Spectroradiometer) Terra and Aqua optical data have been used for this purpose, since they allow: frequent monitoring due to their daily frequency, regional analysis thanks to their global coverage (250 m spatial resolution) and the generation of a 10 years historical archive.

Since the imagery can be affected by the presence of cloud coverage and cloud shadows can be wrongly classified as water bodies because of the spectral response very similar to the one of water, a temporal composite technique has been adopted. The composition algorithm is made by the following steps:

  • automated detection of water bodies from daily satellite imagery;
  • combination of the 2 daily images of MODIS Terra and Aqua satellites;
  • time compositing over a 10-days period;
  • extraction of water bodies.
Figure 1 (Credits: Flooded areas representation in occasion of the Bangladesh flood of June-August 2004. The image is referred to the situation during the last decade of July.)

The compositing algorithm allows the creation of an historical archive starting from February 2000.

Different information on flood hazard can be extracted from the archive, such as:

  • Historical flooded areas can be derived from the comparison with reference water information (figure 1). Since water bodies extension can change during the season, it is necessary to detect the variability of water bodies and consequently to create a dynamic reference water archive.
  • The recurrence of flooded areas during the same 10-days period can be considered an indicator of hazard during a particular moment of the year and it is proportional to the number of times in which an area resulted to be flooded in the 10 years in the same 10-days period, thus allowing to consider water bodies seasonality (figure 2).
Figure 2 (Credits: Number of times in which a pixel resulted to be covered by water in the 10 years during the last 10-days period of July in Bangladesh. The pixels which have never resulted to be covered by water are shown in brown.)

The procedure can be adopted for different regions of the globe. At the moment it is being validated for the countries of Bangladesh and Myanmar.