Optimum Method for Water Body Extraction from Multispectral Satellite Images

Aims :

This research aimed to introduce an optimal method for water body extraction from multispectral satellite images.

Methodology : 

The applied research method is practical and based on the case study. In order to the extraction of water from Landsat 8 satellite images, the modified normalized difference water index along with three thresholding methods including the average method, Otsu’s algorithm, and zero-limit thresholding were applied. For evaluation of the results, the Sentinel 2 image was classified using the support vector machine method then the class of water was used as the reference to calculate the
confusion matrix and correlation coefficients.

Findings:

 The results showed that applying the zero-limit thresholding on modified normalized difference water index for water extraction from Landsat 8 satellite images is not suitable as it cannot identify shallow areas near the coastline. In contrast, Otsu’s thresholding method was able to accurately identify water and had good spatial matching with the reference data.

Conclusion :

Modified normalized difference water index method can successfully identify water areas on multispectral images but choosing the right threshold to separate water from land has a great impact on output accuracy. Most of the past research, use zero-limit thresholding, while the low efficiency of this thresholding method was proven in the current research. Otsu’s thresholding method, in addition to increasing the accuracy, does not require user intervention and uses image statistical information, so it is ideal for the automation of the process of water body extraction from satellite images