Remote sensing inversion of reed height and spatial distribution characteristics in the Bosten Lake Wetland

As the dominant species in Bosten Lake Wetland of Xinjiang, Phragmites australis plays a vital role in the wetland dynamic evolution and ecological functional maintenance, and its growth status directly reflects the health level of the wetland ecosystem. Satellite remote sensing technology has become an efficient and reliable means to monitor the dynamic changes of large-scale reed vegetation, which can overcome the limitations of traditional field surveys such as small monitoring range and high time cost. To systematically explore the height characteristics and spatial distribution patterns of reeds in the Bosten Lake Wetland, this study adopted multi-source data including Sentinel-1A/2A satellite remote sensing images acquired in September 2023, unmanned aerial vehicle (UAV) imagery, and field measured data. Taking artificial visual interpretation and field sampling surveys as the discrimination basis, multiple machine learning models were constructed to realize the remote sensing inversion of reed height and the extraction of reed spatial distribution characteristics. The results showed that the total reed coverage area of Bosten Lake Wetland was 422.55 km². Reeds were mainly distributed in multiple typical wetland areas including the Huangshui Ditch, the west bank of the main lake, the Luoxia Bay, the Kushui Ditch, and the small lake wetland on the southern bank of Bosten Lake. The overall spatial distribution presented a continuous and narrow zonal pattern around the lake and along the rivers, showing a tight adjacency relationship with the water body. Except for the fragmented reed distribution in the west bank of the main lake and the surrounding areas of the Kaidu River, large-scale and densely distributed reed communities were formed in other wetland regions. In this study, several optimal reed height inversion models under different feature combinations were constructed based on multiple machine learning algorithms. Among all the established models, the random forest regression model exhibited the best fitting performance. The precision evaluation results indicated that the coefficient of determination (R²) of the optimal model reached 0.87, while the root mean square error (RMSE) and mean absolute error (MAE) were 0.50 m and 0.29 m, respectively, demonstrating high stability and prediction accuracy of the model. The statistical results of the reed area showed that the total reed coverage area of Bosten Lake Wetland was 422.55 km². According to the difference in plant height, reeds were classified into four growth types, namely low-type, medium-low type, medium-high type, and high-type, with corresponding areas of 4.47 km², 73.41 km², 160.04 km², and 184.63 km², respectively. In addition, the overall growth condition of reeds in the natural wetland area was superior to that in the artificially managed reed area. This study clarifies the spatial distribution law and height differentiation characteristics of reeds in the Bosten Lake Wetland, and constructs a high-precision reed height inversion model, which can provide a reliable theoretical reference and technical support for the ecological protection, vegetation monitoring, and scientific management of reed resources in the Bosten Lake Wetland.