Methodology for Fractional vegetation cover extraction in sparse planted plantations across complex terrain

文献类型： 外文期刊

第一作者： Ma, Yanpeng

作者： Ma, Yanpeng;Fan, Yiguang;Chen, Riqiang;Bian, Mingbo;Fan, Jiejie;Liu, Yang;Zhang, Zitai;Guo, Lixiao;Yang, Guijun;Feng, Haikuan;Ma, Yanpeng;Chen, Riqiang;Dong, Liguo;Feng, Haikuan

作者机构：

关键词： Unmanned aerial vehicle; fractional vegetation cover; machine learning; linear mixed models; color indices; topographic factors

期刊名称：INTERNATIONAL JOURNAL OF REMOTE SENSING （ 影响因子：2.6； 五年影响因子：2.9 ）

ISSN： 0143-1161

年卷期： 2025 年 46 卷 14 期

页码：

收录情况： SCI

摘要： Fractional Vegetation Coverage (FVC) serves as a critical metric for assessing forest carbon sequestration. Traditional FVC measurement methods in mountainous sparsely planted plantations with complex topography are labour-intensive and time-consuming, while (Unmanned Aerial Vehicle) UAV remote sensing has emerged as a pivotal tool for plantation FVC monitoring. However, existing FVC monitoring approaches relying on manual threshold selection exhibit significant subjective bias and limited regional generalizability. This study first extracted FVC for plantations in the Liupan Mountain region using a pixel dichotomy model based on UAV multispectral imagery, evaluating ground vegetation coverage patterns. Subsequently, colour indices (CIs) and topographic factors (elevation, slope, aspect) were acquired via UAV RGB sensors. Machine learning regression algorithms and linear mixed models (LMM) were employed to investigate the performance of FVC estimation using CI alone versus CI combined with topographic factors. The results of the study show that, (1) Spatial Heterogeneity of FVC,FVC displays distinct regional distribution patterns and pronounced spatial heterogeneity. North-eastern areas exhibit higher FVC values (mean >0.65), whereas north-western and south-eastern regions are characterized by low-to-medium coverage (mean <0.40); (2) CI performance limitations among the CI, GLI, r+b, and EXG exhibit the strongest correlations with FVC (significant at the 0.01 level). However, machine learning regression models that rely solely on CIs underperform on validation datasets (R-2 <0.8); (3) FVC estimation models integrating CI with topographic factors demonstrate superior accuracy (R-2 >0.8) when constructed using machine learning regression algorithms and Linear Mixed Models (LMM). Among these, the LMM-based approach - treating CI as fixed effects and topographic factors as random effects - achieves optimal fitting performance with robust stability, yielding validation metrics of R-2 = 0.87, RMSE = 0.05, and NRMSE = 10.28%. This study provides a scientific foundation for large-scale plantation FVC monitoring and the development of universal FVC assessment systems.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序