Estimation of Vegetation Coverage in Semi-arid Sandy Land Based on Multivariate Statistical Modeling Using Remote Sensing Data
Estimation of Vegetation Coverage in Semi-arid Sandy Land Based on Multivariate Statistical Modeling Using Remote Sensing Data
复制标题
DOI：
10.1007/s10666-013-9359-1
复制
发表时间：
2013-10-01
影响因子：
2.4
通讯作者：
Cao, Chunxiang
中科院分区：
文献类型：
--
作者：
Chen, Wei;Sakai, Tetsuro;Cao, Chunxiang
植被覆盖度的估算在干旱和半干旱桑迪的监测和管理中至关重要。但如何在全球和区域尺度上估算植被覆盖度和监测环境变化仍有待进一步研究。在此，结合野外植被调查，利用多光谱遥感数据，在理论统计模型的基础上估算植被覆盖度。首先对遥感数据进行处理，选择/提出并计算几组光谱变量，然后将其与实测植被覆盖度进行统计相关。建立了基于单变量和多变量的模型并进行了分析。在所有基于单变量的模型中，基于归一化植被指数的模型表现出最高的R（0.900）和R（2）（0.810）以及最低的标准估计误差（0.128024）。由于使用多元逐步回归分析的基于多变量的模型表现得更好，因此将其确定为局部覆盖率估计的最佳模型。最后，基于最优模型进行估计，并对结果进行交叉验证。用于验证的决定系数为0.867，均方根误差（RMSE）为0.101。基于遥感数据的大尺度植被覆盖度统计模型的建立，有助于干旱半干旱地区荒漠化的监测和防治。它可以为区域生态管理服务，具有重要意义。
The estimation of vegetation coverage is essential in the monitoring and management of arid and semi-arid sandy lands. But how to estimate vegetation coverage and monitor the environmental change at global and regional scales still remains to be further studied. Here, combined with field vegetation survey, multispectral remote sensing data were used to estimate coverage based on theoretical statistical modeling. First, the remote sensing data were processed and several groups of spectral variables were selected/proposed and calculated, and then statistically correlated to measured vegetation coverage. Both the single- and multiple-variable-based models were established and further analyzed. Among all single-variable-based models, that is based on Normalized Difference Vegetation Index showed the highest R (0.900) and R (2) (0.810) as well as lowest standard estimate error (0.128024). Since the multiple-variable-based model using multiple stepwise regression analysis behaved much better, it was determined as the optimal model for local coverage estimation. Finally, the estimation was conducted based on the optimal model and the result was cross-validated. The coefficient of determination used for validation was 0.867 with a root-mean-squared error (RMSE) of 0.101. The large-scale estimation of vegetation coverage using statistical modeling based on remote sensing data can be helpful for the monitoring and controlling of desertification in arid and semi-arid regions. It could serve for regional ecological management which is of great significance.