In the Great Khingan State Ecosysterm Research Station in Inner Mongolia , we chose a more suitable method to esti-mate forest aboveground carbon storage with the plots data from 2012, 2013 and the synchronously acquired airborne Li-DAR data of 2012 as data sources in the study area , by comparing the model estimated accuracy of multiple linear stepwise regression and random forest regression algorithms to realize the remote sensing estimation of forest aboveground carbon storage of study area .The random forest regression algorithm was training higher accuracy ( model training accuracy R2=0.861, RMSE=11.133 t/ha and rRMSE=0.279;testing accuracy R2=0.826, RMSE=17.956 t/ha, rRMSE=0.342, the esti-mate accuracy range is in 40.898%-95.129%and its average estimate accuracy is 76.385%) than the multiple linear step-wise regression algorithm (model training accuracy R2=0.676, RMSE=11.846 t/ha and rRMSE=0.351;testing accuracy R2=0.727, RMSE=22.703 t/ha, rRMSE=0.636, the estimate accuracy range is in 45.824%-94.752%and the average estimate accuracy is 69.859%) .The percentile height and density variables of LiDAR data had significant correlation with the forest aboveground carbon storage , percentile height variable correlation is more significant .Therefore, the estimate results of to-tal forest carbon storage on regional scale using random forest regression algorithm was closer to its true distribution with ideal effects.%以内蒙古大兴安岭生态站为研究对象，以2012、2013年的66块样地数据和2012年同步获取的机载LiDAR遥感数据为数据源，分别采用多元线性回归和随机森林回归算法，通过对比不同算法间的估测精度差异，选择更适于研究区的估测方法，实现研究区森林地上碳储量的遥感估测。结果表明：随机森林回归算法的估测精度最优，模型训练精度（R2为0．861，RMSE为11．133 t／hm2，rRMSE为0．279）和预测精度（RMSE为17．956 t／hm2，rRMSE为0．342，估测精度范围40．898％～95．129％，平均估测精度76．385％）均优于多元线性回归的模型训练结果（ R2为0．676，RMSE为11．846 t／ha，rRMSE为0．351）和模型预测结果（RMSE为22．703 t／hm2，rRMSE为0．636，估测精度范围45．824％～94．752％，平均估测精度69．859％）。机载LiDAR数据的高度变量和密度变量与森林地上碳储量均具有显著相关性，高度变量相关性更为显著。随机森林回归算法对区域森林地上碳储量的估测结果趋于真实分布情况，效果比较理想。
展开▼
