基于BP神经网络的森林可燃物负荷量估测

Forest fuel loading estimates based on a back propagation neutral network

森林可燃物负荷量是决定林火行为的一个重要因子,因此,森林可燃物负荷量估测对于森林防火管理具有重要意义。该文利用BP神经网络方法和多元回归方法对大兴安岭地区落叶松林32块森林样地数据构建森林可燃物负荷量预测模型,用以研究利用林龄、郁闭度、平均高、胸径等林分因子估测该地区森林可燃物负荷量的方法。通过MATLAB软件实现BP神经网络森林可燃物负荷量估测模型;通过SPSS软件建立多元回归森林可燃物负荷量估测模型。BP神经网络森林可燃物负荷量估测模型拟合精度为99.9%、外推精度为65.51%;多元回归可燃物负荷量估测模型拟合精度为68.29%、外推精度为62.1%。通过比较分析,得出结论:利用BP神经网络方法估测森林可燃物负荷量是可行的;BP神经网络模型精度高于多元回归模型;由于训练样本太少,2种模型外推精度低于70%。

Forest fuel loading is an import factor related to the fire behavior;therefore,forest fuel loading estimates are very important for forest fire management.This study used data collected in 32 Larix Gmelinii forest sample locations in the Daxing＇anling region to build a multivariable regression equation and a back propagation neural network model to estimate forest fuel loadings from stand factors,such as tree age,crown cover,average tree height,and diameter at breast height.The back propagation neutral network training and simulation used for models in MATLAB with the multivariable regression equation developed using SPSS.The fitting precision of the back propagation neutral network model was 99.9% with an extrapolation precision of 65.51%.The fitting precision of the multivariable regression equation was 68.29% with an extrapolation precision of 62.1%.Thus,the back propagation neutral network model for estimating forest fuel loading from the stand factors is more accurate than the multivariable regression equation.The extrapolation accuracy of both the BP neutral network model and the multivariable regression model are less than 70% due to the small number of samples.