基于主成分、聚类与SVR组合算法的森林生物量估算方法研究

Forest Biomass Evaluation Method Based on Principle Component Analysis,Clustering and SVR

以孟家岗林场二类清查数据为基础,对1371个小班的11项指标进行主成分分析,并采用系统聚类法对小班进行分类,进而利用支持向量回归算法分别进行生物量模型训练。结果表明:7个主成分指标可反映87.995%的生物量信息;1371个小班可分为5类,各类训练模型的预测精度均在89%以上,且均以v-SVR模型为最优。在得到的5类生物量训练模型基础上估算林场森林乔木层生物量,无需分起源、树种、立地类型,能够在保证生物量估算精度的同时,大大减少工作量,可为区域生物量的估算提供一种新的方法。

Based on the forest resources inventory data of Mengjiagang forest farm, principle component analysis was used to ex- tract the principle components of 11 indexes for 1371 subeompartments, and system clustering method was used to classify these sub- compartments, and support vector regression algorithm was used to train the biomass estimation model based on the classified subeomp- artments. Experiment results showed that 7 principle components can reflect 87. 995% information of biomass; 1371 subcompartments can be divided into 5 types, and the prediction accuracy are all above 89% and v-SVR model was proved to be the optimal one among the training models. Therefore, it is possible to estimate the tree layer biomass based on the 5 types of biomass models without the in- formation of origin, species, and site. The method can guarantee the biomass estimation accuracy while greatly reducing the work- load, which provides a new regional biomass estimation method.