Wetland information extraction and dynamic change monitoring with joint multifeature optimization and time-series imaging

Wetlands exhibit seasonal characteristics, and time-series remote sensing data can more accurately identify their spatial and temporal patterns of change. This paper takes the Hulun Lake National Nature Reserve as a research object and systematically analyses its seasonal dynamic changes. Traditional monitoring methods rely on simple time series stacking, which makes it difficult to effectively extract key features. Therefore, selecting the best features to improve classification accuracy while reducing computational cost becomes a key issue. In this paper, a 7-group feature combination extraction scheme was constructed based on time series data and feature optimization techniques, and wetland distribution maps for each quarter from 2019 to 2023 were generated using the random forest algorithm. The results of the study show that：(1) The order of importance of different features for wetland information extraction is: vegetation, water, and snow index features > red edge index features > texture features > SAR polarization index features > SAR polarization features；(2) The feature selection based on the Random Forest algorithm performed as expected, achieving an overall accuracy of 98.01% and a Kappa coefficient of 0.97. The results demonstrate that the method markedly reduces data dimensionality and enhances computational efficiency while maintaining high classification accuracy. Additionally, it provides a novel approach to feature selection for wetland information extraction and classification while also validating the potential and utility of random forests and feature selection in the analysis of long time series data.