Integrating Otsu Thresholding and Random Forest for Land Use/Land Cover (LULC) Classification and Seasonal Analysis of Water and Snow/Ice
Xuexia Sun, Xiaoyao Li, Bingxiang Tan, Jian Gao, Lei Wang, Shimei Xiong
Abstract
Accurate land use/land cover (LULC) classification and the detection of seasonal dynamics are crucial for effective environmental monitoring and resource management. To improve the precision and temporal resolution of regional LULC classification products, this study combined the Otsu threshold method and Random Forest algorithm to generate a 10 m-resolution land cover classification map for Wensu County based on Sentinel-2 imagery, with a particular focus on orchard categories, and investigated the seasonal dynamics of LULC between winter and summer. The results show that the overall accuracy (OA) of the water and snow/ice models was 85.50%, with a Kappa coefficient of 0.8088; for the vegetation model, the OA was 93.77%, with a Kappa coefficient of 0.8755. Feature importance analysis indicated that terrain features were key factors in improving classification performance. Seasonal dynamics analysis showed that the snow/ice coverage area in winter increased by 6379.18 square kilometers compared to that in summer, with 5252.85 km2 of bare land and 910.66 km2 of grassland being covered by snow/ice. Meteorological data analysis revealed that land cover changes caused by winter snowfall were primarily concentrated in areas where temperatures exceeded −8 °C, while land cover changes were smaller in areas with either low or high precipitation. These findings provide valuable data support for regional resource management and agricultural development.