New method improves extraction accuracy of lake water bodies in Central Asia

Lakes play an important role in terrestrial ecosystems and have a significant impact on human production and life. Remote sensing techniques have been widely used to monitor changes in lakes. Several studies have been conducted pertaining to extraction of lake bodies using optical remote sensing images; however, owing to the influence of ice and snow in the lake and other parameters such as noise, the accuracy and stability of lake extraction still need to be improved. This study designed a new method by combing K-means clustering and the flood fill method (KCFFM) to improve the accuracy and stability of extraction results from complex backgrounds. Using Landsat 8 Operational Land Imager (OLI) data for five lakes in Central Asia, the accuracy and stability of KCFFM in summer and winter were evaluated and compared with those of five other methods. KCFFM showed the highest accuracy and stability among all the methods, especially during the lake ice period. For KCFFM, the Kappa coefficient was greater than 0.97 and 0.92 and the overall accuracy was greater than 99% and 98% in summer and winter, respectively. The area error rate of KCFFM was less than 3%, except for Lake Alakol (less than 10%). In addition, KCFFM significantly decreased the area error rate compared to other methods (1% to 15% in summer and 10% to 50% in winter). The proposed method optimized the time continuity, accuracy, and stability of lake body extraction. Thus, KCFFM can provide valuable basic data for monitoring lake water bodies.

Automated summary

This study introduced a new method, KCFFM, combining K-means clustering and the flood fill method, to improve the accuracy and stability of lake extraction from complex backgrounds. The results showed that KCFFM had the highest accuracy and stability among all methods, especially during the lake ice period in Central Asia.