A Hybrid Mangrove Identification Method by Combining the Time-Frequency Threshold of the Mangrove Index With a Random Forest Binary Classifier

Mangroves are in coastal zones where mass-energy exchange is most active. Their functions in high productivity, strong carbon sequestration capacity, and rich ecosystem services are crucial for achieving the sustainable development goals. Although various classification methods have been extensively applied in large-scale mangrove observations, they necessitate considerable sample collection and postprocessing, which hampers the long-term identification of mangroves. This study proposes a hybrid identification method that combines the time-frequency threshold of the mangrove index with a random forest binary classifier. It efficiently identifies mangroves over large-scale areas with low postprocessing and high accuracy. The hybrid method was applied to Landsat 8 OLI data to derive the mangrove distribution in China using Google Earth Engine. It demonstrated an overall accuracy rate of 92.86% and an F1 score of 0.92, a significant improvement over using either method alone. Specifically, the method utilized a small sample size to successfully obtain pure and accurate mangrove classification. The feature selection indicates that the short-wave infrared band and its associated remote sensing indices play a pivotal role in differentiating mangroves from other confusable land classes. Additionally, this method was successfully applied to Landsat 5 data, achieving an overall accuracy of 93.49%, with substantial agreement between the classification results of the two sensors. This indicates that the method has the potential for continuous monitoring of mangroves over large areas and extended periods.