Water pollution evaluation through fuzzy c-means clustering and neural networks using ALOS AVNIR-2 data and water depth of Lake Hosenko, Japan

• Published in: Ecological informatics Vol. 70; p. 101761
• Main Authors: Matsui, Kai, Kageyama, Yoichi
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2022
• Subjects: ALOS AVNIR-2; Neural network algorithm; Remote sensing; Unsupervised learning; Water area; Water depth; NN; AVNIR-2; IR; DN; Water area; BN; Water depth; Remote sensing; Unsupervised learning; ALOS; FCM; ALOS AVNIR-2; GCP; RMS; Neural network algorithm; ReLU
• ISSN: 1574-9541
• DOI: 10.1016/j.ecoinf.2022.101761

Water is a critical resource for human life and an essential component of ecosystems and the environment. In recent years, water pollution prevention strategies have become increasingly important. Remote sensing techniques that periodically capture wide-area data are useful for surveying the conditions of water bodies. Using a general water quality estimation method to obtain maps consistent with the experimental data, remote sensing data and water quality values should be obtained simultaneously. However, the requirement for simultaneous observation impedes the creation of water quality estimation maps. Methods must be developed to evaluate water pollution conditions without water quality data. In this study, we propose a neural-network-based method to evaluate water pollution conditions in Lake Hosenko, Japan, using remote sensing data and the water depth of the lake. The water depth was calculated based on the water surface and lake bottom elevations. Fuzzy c-means (FCM) clustering was employed to create evaluated water pollution condition maps using only remote sensing data. The proposed method utilizes the FCM results and water depth data during the learning process. We verified the usefulness of the evaluation results obtained through comparisons with actual water pollution conditions in the lake. The results suggest that the difference in water surface elevation affects the mechanism of pollution in Lake Hosenko, and the proposed method contributes to understanding this mechanism. Therefore, the proposed method can identify the pollution condition without water quality data and support the implementation of remedial measures. •Proposed method uses remote sensing data and water depth data.•NN employed to learn relationship between results by fuzzy c-means and water depth.•Helps grasping the relation between water pollution condition and its mechanism.•Large datasets and water quality data not required for creating an evaluation map.•Estimation maps with higher frequency can be created.