基于随机森林的不同径级树木死亡影响因子研究

Influence factors of tree mortality in different size classes based on random forest

树木死亡是多种因素共同作用的结果,然而不同径级的林木,导致其死亡的原因是否相同尚不明确。以坝上地区小叶杨为对象,于2016—2017年调查了100块20m×20m样地。以林木胸径、林木相对断面积、林龄、林分密度、不同土层土壤含水量和土壤容重等11个因子为输入变量,运用随机森林算法分别不同径级建立树木死亡模型,分析各因子对不同径级树木死亡的影响大小,并得到关键影响因子的阈值。结果表明:依据不同径级树木死亡与影响因子的相关性不同,可将林木径级划分为<10cm、10—25cm、>25cm;胸径<10cm时,林木相对断面积对树木死亡的影响最大,当0.23<林木相对断面积<0.43时,树木死亡率最低;胸径为10—25cm时,密度对树木死亡的影响最大,当密度>600株/hm2时,树木死亡率随着密度的增加明显上升;胸径>25cm时,20—60cm土壤水分和林分断面积对树木死亡的影响最大且相近,当20—60cm土壤质量含水量>5%时,树木死亡率明显下降;基于随机森林算法建立的树木死亡模型可以解释树木死亡70%—90%的变异,具有一定的预测优势和应用价值,可为坝上地区树木死亡预测和防护林的经营管护提供参考。

Tree mortality correlates with many biotic and abiotic variables.However,less is known about how the relationship between influence factors and tree mortality changes with tree size.To understand size dependent relationship between drivers and tree mortality,we investigated tree mortality of Populus simonii Carr over a wide range of tree sizes in 100 plots in Bashang Area.We divided trees into three size classes based on the magnitude of the effects of influence factors on tree mortality.The random forest(RF)was used to find out the key factors for each size class.The results showed that the trees were categorized into small trees(DBH<10cm),medium trees(10≤DBH<25cm),and large trees(DBH≥25cm).The relative basal area(RBA)was the most important predictor for mortality of small sized trees.The tree mortality rate was the lowest with 0.23<RBA<0.43.Density was the most significant influence factor for mortality of medium sized trees.When stand density was higher than 600 stems hm-2,the tree mortality rate increased dramatically with the growth of stand density.The soil moisture content of 20—60 cm and stand basal area(SBA)were the most important two predictors for mortality of large trees.The tree mortality rate declined sharply when soil moisture content of 20—60 cm reached 5%.Biotic and abiotic variables could explain 70%—90%variation in tree mortality with RF model.Thus RF model can be usedin the tree mortality prediction and provide important information for the local forest management and plan.