Evolution of water pattern in Chagan Lake Basin from 2017 to 2025 using PlanetScope images and a newly developed MS-FCN8s algorithm

Abstract

As a typical lake group in Northeast China, the Chagan Lake Basin's water dynamics are crucial for regional water security and ecological protection. High-resolution images have limited spectral information, water indices perform poorly in complex backgrounds, and deep learning models often blur boundaries or miss detections. To solve these problems, this study aims to build a new MS-FCN8s model using PlanetScope images and reveals the basin's spatiotemporal water pattern evolution. Results indicate that MS-FCN8s algorithm can integrate multi-band reflectance and water indices, showing stable and accurate extraction, with overall accuracy (OA) about 2.8% higher than traditional FCN. In addition, it outperforms U-Net, DeepLabV3+, and TransUNet in detecting small water bodies and preserving boundaries. Moreover, water body changes in the Chagan Lake Basin (2017–2025) showed a total area increased by 17.14%, indicating sustained hydrological expansion. Spatially, the pattern shifted decisively from fragmented, isolated patches to a centralized, hydrologically connected system anchored by Chagan Lake. Area growth was overwhelmingly driven by large water bodies (>1 km 2 ), whereas small water bodies (<0.1 km 2 ), though numerically dominant, contributed less than 3.5% to the total areal gain.