As a globally significant floodplain wetland

Poyang Lake features pronounced seasonal variations in its water bodies throughout the year. Accurate identification of the dynamic changes in its water bodies and landscape patterns is crucial to maintaining the stability of its aquatic ecosystem. This paper presents a water body dynamic identification model that we have constructed combining high spatial resolution with high spatiotemporal information. Based on the global surface water dataset JRC-GSW

we first develop a rule-based automatic sampling strategy using multi-source remote sensing data to obtain a high-confidence water body training dataset. Then

a classification model is trained using the Random Forest algorithm

integrating multi-source time-series data to construct a seasonal dynamic feature recognition model for water bodies. In a case study of the Poyang Lake Nanji Wetland National Nature Reserve

this recognition model is applied to examine the 2023 flood season using the data at 10-day intervals and 10-meter resolution. Results demonstrate that it has an enhanced robustness and recognition capability in handling complicated wetland environments

and improves significantly the JRC-GSW products that suffer from water body under-detection and misclassification

enabling intuitive visualization of water body dynamic processes. We reveal the overall water body landscape varies in a pattern characterized by main lake retreat-depression convergence under the condition of an earlier onset of the dry season in this year. This means that the Nanji Wetland's hydrological process is sensitive to seasonal water fluctuations and extreme drought events

a useful factor to the conservation planning and effectiveness evaluation of important wetlands.