基于岭回归和人工神经网络估测森林可燃物负荷量

Estimation of Forest Fuel Load Based with Ridge Regression and Artificial Neural Networks

选取东北林业大学帽儿山实验林场为研究区域,以少量野外定位调查数据及与其对应的遥感和GIS信息为基础,利用岭回归和人工神经网络分析方法,对森林可燃物负荷量估测模型及其影响因子进行系统研究。结果表明:对于TM3、TM(4×3)/7、TM4/3、海拔等10个影响可燃物负荷量估测的主要因子,利用岭回归方法可以克服变量间由于存在复共线性关系对求解待定参数所造成的不利影响。建立以像元为单位的岭回归和岭回归与神经网络组合估测模型,模型平均绝对百分比误差分别为17.6%和11.7%,2种方法可用于实现特定林场尺度森林可燃物负荷量的定量估测,其中组合模型效果较好。

Based on data of a field positioning survey and the corresponding remote sensing and GIS, the forest fuel load and the influence factors were researched by using ridge trace analysis and artificial neural networks in Maoershan experimental forest station of Northeast Forestry University. Ridge regression method can overcome the negative impact imposed the undetermined parameters there exist in the multicollinearity relationship solution between variables which include ten main influence factors, i.e., TM3, TM（4×3） /7, TM4/3 and altitude. A model was established for estimating forest fuel load with the unit of pixel, and Ridge Regression and Artificial Neural Networks MAPE. The deviation of estimation by the two models was 17.6% and 11.7%. The result indicated that the quantitative estimation of forest fuel load for regional forests could be achieved.