摘要: |
为选出精准评估森林生物量的计量方法, 文章以广东省河源市森林为研究对象, 基于 Landsat
8 OLI 影像数据和 2017 年广东省第九次国家森林资源连续清查数据, 分析对比不同特征变量选择方法,
采用多元逐步回归模型和随机森林回归模型建立河源市森林乔木层地上生物量 (Above ground biomass,
AGB) 反演模型。 研究结果表明, 利用 Pearson 相关系数法筛选变量的多元逐步回归模型 (R
2 为 0. 505,
Ermse 为 36. 862 t·hm
-2
, Emae 为 30. 555 t·hm
-2
, Aea 为 58. 4%) 表现最佳; 相比随机森林的特征重要性
来筛选自变量, 利用 Pearson 相关系数法进行自变量选择的筛选方法更佳; 随机森林算法估算精度均优于
多元逐步回归模型。 采用随机森林算法进行森林地上生物量反演应用前景良好。 |
关键词: 森林生物量 森林资源连续清查 Landsat 8 OLI 遥感影像 随机森林 |
DOI: |
分类号: |
基金项目:广东省林业科技创新重点项目 (2021KJCX009)。 |
|
Discussion on the Remote Sensing Inversion of Forest Biomass in Heyuan Based on the Landsat 8 OLI Image |
Xiong Zhuang,Hu Zhongyue,Cao Cong,Liu Ping,Xu Zhengchun
|
1.South China Agricultural University;2.Central South Survey Planning Institute, State Administration of Forestry and Grassland
|
Abstract: |
To select the appropriate biomass estimation method to accurately evaluate forest biomass, the
forest in Heyuan city, Guangdong province was taken as the research object. Based on Landsat 8 OLI image data and the data from the Ninth National Continuous Forest Inventory of Guangdong Province in 2017, the selection methods of different characteristic variables were analyzed and compared. The multiple stepwise regression
models and random forest regression models were used to establish the estimation model of above ground biomass
(AGB) in the forest tree layer of Heyuan city. The results showed that the multiple stepwise regression model
with variable selection using the Pearson correlation coefficient method (with a coefficient of determination R
2
of
0. 505, Ermse of 36. 862 t·hm
-2
, Emae of 30. 555 t·hm
-2
, and Aea of 58. 4%) performed the best. Compared
to selecting independent variables using feature importance from the random forest, the Pearson correlation coefficient method yielded better results for independent variable selection. Nevertheless, the estimation accuracy of
the random forest algorithm was superior to that of the multiple stepwise regression model. The random forest al-gorithm has promising application prospects for AGB estimation in forests. |
Key words: forest biomass national forest inventory (NFI) Landsat 8 OLI image random forest |