A Novel Water Change Tracking Algorithm for Dynamic Mapping of Inland Water Using Time-Series Remote Sensing Imagery

Fine-scale mapping of inland water is important for regional or global ecosystem monitoring; however, due to the spectral heterogeneity of inland water, large-scale dynamic mapping of these water bodies is usually challenging. In this article, we propose a water change tracking (WCT) algorithm, that uses only optical data, for accurate inland water mapping. Simultaneously, time-series of Landsat imagery are used as an example for illustrating the effectiveness of the method. First, time-series of Minimum Normalized Water Score (MNWS) images and corresponding `bad-pixel' (cloud, ice and snow) masks were developed using time-series of Landsat imagery. Secondly, based on the MNWS time-series stack and `bad-pixel' masks, time-series trajectories were formed to track changes in water bodies. A 30-m Dynamic Inland Water Body Map (DIWBM) of China was then produced to test the robustness of the WCT algorithm. Eight test sites were selected, and the water detection algorithms used in the Global 1 arc-second Water Body Map (G1WBM) and the function of mask (FMASK) were applied to our data to provide a comparison with the DIWBM results. The results indicated that the DIWBM gave a better performance of 96.20% against 85.91% for the G1WBM results and against 76.36% for the FMASK results. The DIWBM achieved user's accuracies of 98.36% and 92.53%, and producer's accuracies of 97.45% and 95.39% for permanent and temporary water, respectively. The proposed method was also better at identifying small water bodies. It is concluded, therefore, that the WCT algorithm is a promising method for large-scale dynamic inland water mapping.