基于非参数方法的黑河地区森林地上生物量估算及时空格局

Estimation and Spatial-temporal Pattern of Forest above Ground Biomass in Heihe Region Based on Nonparametric Method

以黑龙江省黑河地区2005年和2010年两期森林资源连续清查固定样地数据和Landsat TM5遥感影像为数据源,首先以2005年和2010年影像为研究对象,分别提取纹理、地形和气象因子共73个特征变量,依据均方差增量百分数%IncMSE(Increase Mean Squared Error%)指标筛选出前六个重要性变量;之后,分别以RF模型和传统的多元逐步回归模型估算森林AGB,并进行对比与分析;最后,对精度较高的模型分别估算两期黑河地区森林AGB并进行时空动态分析。研究结果表明:引入纹理因子RF模型可以一定程度提高模型精度,两期Radj^2分别为0.40和0.39,RMSE分别为27.41(t/hm2)和32.01(t/hm2),优于传统的多元逐步回归模型;黑河地区森林AGB整体呈现增长趋势,森林AGB大于80(t/hm2)的高生物量区域有较明显的增加,主要集中在东部和部分北部山地及中海拔区域。

Taken Heihe region in Heilongjiang Province as study area, the fixed sample plots data of continuous forest inventory and Landsat TM5 data in 2005 and 2010 were taken as data resources. First, based on the 2005 and 2010 images as a reasarch object, a total of 73 feature variables were extracted for texture, terrain, and meteorological factors, and the first six importance variables were selected according to the IncMSE(Increase Mean Squared Error%) index. Next, the estimation forest AGB was compared and analyzed between RF model and traditional multiple stepwise regression model. Finally, the high-accuracy models were used to estimate the forest AGB and space-time dynamic analysis of Heihe region in two periods. The results were showed that the introduction of the texture factor based on RF model can improve the accuracy to a certain extent. The in the two periods is 0.40 and 0.39, RMSE is 27.41(t/hm^2) and 32.01(t/hm^2), respectively, which is better than traditional stepwise multiple linear regression results. The growth trend has a significant increase in high-level areas with AGB greater than 80(t/hm^2), mainly concentrated in the eastern and some northern mountainous areas and medium-altitude areas.