Multi-Feature Fusion for Grassland Information Extraction and Temporal Change Monitoring

Xiyuan Liu, Jie Chen, Zijin Liu , Xiaohua Zhu, Shigao Jiang, Wanli Kang

Grasslands, as an important terrestrial ecosystem, play a vital role in maintaining ecological security and promoting regional sustainable development. However, in complex mountainous environments, grassland information extraction often suffers from insufficient accuracy, and the underlying mechanisms driving its changes remain unclear. To address these issues, this study takes Chengbu Miao Autonomous County in Hunan Province as the study area. Based on the Google Earth Engine (GEE) platform, a multi-feature fusion framework was constructed by integrating optical imagery, Sentinel-1 SAR data, vegetation indices, texture features, topographic factors, and time-series characteristics. A Random Forest (RF) model was then employed to achieve high-accuracy grassland information extraction. On this basis, multi-year time-series data were utilized to analyze the spatial distribution of grasslands and their interannual variation characteristics. At the county scale, fractional vegetation cover (FVC) was introduced to quantitatively characterize grassland ecological conditions. Furthermore, a typical grassland region, Nanshan Pasture, was selected as a key study area. From both pixel and regional scales, the effects of topographic factors (DEM, slope, and aspect), human activities, and climatic factors (temperature and precipitation) on FVC were systematically analyzed. The results indicate that multi-feature fusion significantly improves grassland classification accuracy, with optical and SAR data demonstrating strong complementarity under complex surface conditions. The spatial distribution of grasslands in Chengbu County remains generally stable, although localized variations are observed. Grassland area fluctuates between 47.32 and 52.15 km², with relatively small overall changes and no persistent trend of significant expansion or contraction. Meanwhile, FVC values are mainly distributed within the range of 0.59–0.87, exhibiting stage-wise fluctuations, suggesting that grassland ecological conditions remain generally stable but show certain interannual variability. Within the Nanshan Pasture area, FVC shows a significant negative correlation with DEM (r = -0.2839) and a weak negative correlation with human activities (r = -0.1563), while slope and aspect exhibit no significant influence. Among climatic factors, land surface temperature (LST) shows a weak correlation with FVC, whereas precipitation is significantly negatively correlated with FVC (r = -0.589), with its effects primarily acting indirectly through processes such as solar radiation and soil moisture. Overall, grassland vegetation changes in the study area exhibit a pattern characterized by “natural dominance, human regulation, and indirect climatic influence.” The multi-feature fusion-based grassland extraction and multi-scale driving factor analysis framework proposed in this study can provide technical support for grassland resource monitoring, ecological assessment, and scientific management in complex terrain regions.