Determination of surface water area using multitemporal SAR imagery

Abstract

Inland water and freshwater constitute a valuable natural resource in economic, cultural, scientific and educational terms. Their conservation and management are critical to the interests of all humans, nations and governments. In many regions these precious heritages are in crisis. The main focus of this research is to investigate the capability of time variable ENVISAT ASAR imagery to extract water surface and assess the water surface area variations of lake Poyang in the basin of Yangtze river, the largest freshwater lake in China. Nevertheless, the lake has been in a critical situation in recent years due to a decrease of surface water caused by climate change and human activities.

In order to classify water and land areas and to achieve the temporal changes of water surface area from ASAR images during the period 2006-2011, the image segmentation technique was implemented. For this purpose, a thorough analysis of the SAR system and its properties is first discussed. Indeed, some impairments can affect the SAR imaging signals. These impairments such as different types of scattering, surface roughness, dielectric property of water, speckle and geometric distortions can reduce SAR image quality. To avoid these distortions or to reduce their impact, it is therefore important to pre-process SAR images effectively and accurately. All the images were pre-processed using NEST software provided by ESA.

To calculate the water surface area, each image was tiled into 9 parts and then it is segmented using two different methods. Firstly histogram for each tile is observed. Using a local adaptive thresholding technique, two local maxima were determined on the histogram and then in between these local maxima, a local minimum is determined which can be considered as the threshold. In the second technique a Gaussian curve was fitted using Levenberg-Marquardt method (1944 and 1963) to obtain a threshold. These thresholds are used to segment the image into homogeneous land and water regions. Later, the time series for both methods is derived from the estimated water surface areas.

The results indicate an intense decreasing trend in Poyang Lake surface area during the period 2006-2011. Especially between 2010 and 2011, the lake significantly lost its surface area as compared to the year 2006. Finally, the results are presented for both locally adaptive thresholding and Levenberg-Marquardt methods. These results illustrate the effectiveness of the locally adaptive thresholding method to detect water surface change. A continuous monitoring of water surface change would lead to a long term time series, which is definitely beneficial for water management purposes.