高级检索

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于随机森林算法的落叶松−云冷杉混交林单木胸径生长预测

欧强新 雷相东 沈琛琛 宋国涛

欧强新, 雷相东, 沈琛琛, 宋国涛. 基于随机森林算法的落叶松−云冷杉混交林单木胸径生长预测[J]. 北京林业大学学报, 2019, 41(9): 9-19. doi: 10.13332/j.1000-1522.20180266
引用本文: 欧强新, 雷相东, 沈琛琛, 宋国涛. 基于随机森林算法的落叶松−云冷杉混交林单木胸径生长预测[J]. 北京林业大学学报, 2019, 41(9): 9-19. doi: 10.13332/j.1000-1522.20180266
Ou Qiangxin, Lei Xiangdong, Shen Chenchen, Song Guotao. Individual tree DBH growth prediction of larch-spruce-fir mixed forests based on random forest algorithm[J]. Journal of Beijing Forestry University, 2019, 41(9): 9-19. doi: 10.13332/j.1000-1522.20180266
Citation: Ou Qiangxin, Lei Xiangdong, Shen Chenchen, Song Guotao. Individual tree DBH growth prediction of larch-spruce-fir mixed forests based on random forest algorithm[J]. Journal of Beijing Forestry University, 2019, 41(9): 9-19. doi: 10.13332/j.1000-1522.20180266

基于随机森林算法的落叶松−云冷杉混交林单木胸径生长预测

doi: 10.13332/j.1000-1522.20180266
基金项目: 国家自然科学基金(31870623)
详细信息
    作者简介:

    欧强新,博士。主要研究方向:森林生长模型与模拟。Email:jonsinou@foxmail.com 地址:100091 北京市海淀区香山路东小府2号中国林业科学研究院资源信息研究所

    责任作者:

    雷相东,博士,研究员。主要研究方向:森林生长模型与模拟。Email:xdlei@ifrit.ac.cn 地址:同上

  • 中图分类号: S758.5

Individual tree DBH growth prediction of larch-spruce-fir mixed forests based on random forest algorithm

  • 摘要: 目的单木生长受气候、林分等多种因子影响,需要利用适当的方法厘清气候以及林分中影响林木生长的主导因子。随机森林等机器学习方法提供了一种新的途径,需要检验利用随机森林算法分析气候和林分因子对林木生长影响的可靠性,为森林生长收获预估提供新的方法。方法以吉林省汪清林业局20块落叶松−云冷杉混交林固定样地25年(1986—2010年)间连续调查数据作为研究材料,候选气候和林分因子52个,利用随机森林算法建立了包含气候和林分的单木胸径生长模型,分析气候和林分因子对单木胸径年平均生长量的影响:基于52个超参数组合(决策树数目ntree = 1 000、决策树每个结点随机选择的预测变量个数mtry = {1, 2, ···, 52})构建了52个随机森林模型,利用10折交叉验证法分别训练和评估52个随机森林模型;基于完整数据集,利用最优随机森林模型分析自变量对单木胸径年平均生长量影响的相对重要性以及偏依赖关系。结果ntree = 1 000、mtry = 12所对应的模型是52个模型中具有最佳泛化能力的模型,该模型具有最大的交叉验证决定系数R2cvR2cv = 0.54),以及最小的交叉验证均方根误差RMSEcv、交叉验证平均绝对偏差MAEcv和交叉验证相对均方根误差rRMSEcv(RMSEcv = 0.14 cm、MAEcv = 0.10 cm、rRMSEcv = 50%)。单木胸径年平均生长量受林分因子的影响极大,相对重要性超过80.00%。8个林分因子中,大于对象木的林木断面积之和BAL对单木胸径年平均生长量影响最大,林分每公顷株数N对单木胸径年平均生长量影响最小,其他因子对单木胸径年平均生长量影响介于两者之间;单木胸径年平均生长量随BAL、林分每公顷断面积BA、N以及林分断面积平均胸径Dg的增加而下降,随对象木胸径与林分断面积平均胸径之比RD、林木期初胸径D0以及对象木胸径与林分中最大林木胸径之比DDM的增加而增加。单木胸径年平均生长量受气候因子的影响较小,相对重要性低于20.00%。44个气候因子对单木胸径年平均生长量的影响均较小(相对重要性均 < 1%),其中,生长季平均降水量(4—9月)与年均降水量之比Pratio、年总太阳辐射时长Asr、生长季平均降水量(4—9月)与生长季相对湿度(4—9月)之比Gspgsrh以及生长季太阳辐射时长(4—9月)Gssr是前4个相对重要的变量。结论随机森林模型能够较好地解析各变量与单木胸径年平均生长量之间复杂的关系,单木胸径年平均生长量受林分因子的影响极大,而受气候因子的影响较小。总体而言,在局部尺度上,林分因子是影响单木胸径生长的主导因子,而气候因子对单木胸径生长的解释能力有限。随机森林模型具有一定的泛化能力和统计可靠性,产生的变量重要性和偏依赖图具有合理的林学意义。

     

  • 图  1  不同mtry所对应的随机森林模型10折交叉验证评价指标

    图中黑色圆点为平均值,误差线为标准差;mtry表示树节点随机抽选的变量个数。The black dots in the graph are average, and the error bars are standard error;mtry is the number of predictive vaviables randomly sampled at each split.

    Figure  1.  Ten-fold cross validation evaluation index of different random forest models in accordance with mtry values

    图  2  mtry为12时所对应的随机森林模型在各折测试集上的模型表现

    图中参考线(虚线)为相应评价指标的平均值。The reference lines (dotted line) in the graph are average of corresponding evaluation index.

    Figure  2.  Performance of random forest model with the vaule of mtry as 12 based on each-fold test set

    图  3  基于两种重要性度量方法的各自变量对单木胸径年平均生长量影响的相对重要性得分

    TreeSpe_Code: 树种代码 Tree species code; Amaxt: 年最高气温 Maximum annual temperature; Amint: 年最低气温Minimum annual temperature; Amt: 年平均气温Mean annual temperature; Gsdd5: 生长季大于 5 ℃积温 (4—9月) The accumulated temperature is greater than 5 ℃ in growing season (April to September); Gsmaxt: 生长季最高气温 (4—9月) Maximum temperature in growing season (April to September); Gsmint: 生长季最低气温 (4—9月) Minimum temperature in growing season (April to September); Maxtwm: 最热月的最高气温 (7月) The highest temperature in the hottest month (July); Mmincm: 最冷月的最低气温 (1月) The lowest temperature of the coldest month (January); Mtcm: 最冷月的平均气温(1月) The mean temperature of the coldest month (January); Mtwm: 最热月的平均气温(7月) The mean temperature of the hottest month (July); Gsp: 生长季平均降水量(4—9月) Average precipitation in growing season (April to September); Map: 年均降水量 Mean annual precipitation; Msp: 月总降水量Monthly total precipitation; Sp: 夏季降水量(6—8月) Summer precipitation (June to August); Gsrh: 生长季相对湿度 (4—9月) Relative humidity in growing season (April to September); Marh: 年平均相对湿度 Annual mean relative humidity; Asr: 年总太阳辐射时长Annual total solar radiation duration; Gssr: 生长季太阳辐射时长 (4—9月) Solar radiation duration in growing season (April to September); Ahm: 1 000 × ((Amt + 10)/Map); Shm: Gsmaxt + Maxtwm; Amaxtmap: Amaxt/Map; Amtmap: Amt/Map; Gsmaxtgsrh: (Gsmaxt × Gsrh)/1 000; Gsmintgsp: Gsmint/Gsp; Gsmintgsrh: Gsmint/Gsrh; Gsmintmap: Gsmint/Map; Gspdd5: (Gsp × Gsdd5)/1 000; Gspgsrh: Gsp/Gsrh; Gspmtcm: (Gsp × Mtcm)/1 000; Gspgsmint: (Gsp × Gsmint)/1 000; Gsrhgsmint: (Gsrh × Gsmint)/1 000; Mapdd5: (Map × Gsdd5)/1 000; Mapgsmint: (Map × Gsmint)/1 000; Mapmtcm: (Map × Mtcm)/1 000; Maxtwmsp: Maxtwm/Sp; Mtcmgsp: Mtcm/Gsp; Mtcmmap: Mtcm/Map; Pratio: Gsp/Map.

    Figure  3.  Relative importance scores of each independent variable affecting individual tree DBH growth based on two methods of variable importance calculation

    图  4  林分因子(A ~ H)和部分气候因子(I ~ L)对单木胸径年平均生长量影响的偏依赖关系图

    Broadleaf 1: 慢阔Slow growing broadleaved tree; Broadleaf 2: 中阔Medium growing broadleaved tree; Fir: 冷杉Abies nephrolepis; Larch: 落叶松Larix olgensis; Pine: 红松Pinus koraiensis; Spruce: 云杉Picea jezoensis var. komarovii

    Figure  4.  Partial dependence plots of stand variables (A−H) and partial climatic variables (I−L) affecting individual tree DBH growth

    表  1  林分及单木因子统计

    Table  1.   Summary statistics of stand variables

    因子
    Variable
    平均值
    Mean
    标准差
    Standard deviation
    最小值
    Minimum
    最大值
    Maximum
    说明
    Description
    ΔD/cm 0.28 0.21 0.00 2.04 1986—2010年间任意5年间隔单木的胸径年均生长量
    Individual tree mean annual DBH increment within any 5-year intervals from 1986 to 2010
    D0/cm 17.44 6.51 5.00 51.00 林木期初胸径
    Initial tree DBH
    BAL/m2 17.70 8.29 0.00 37.29 大于对象木的林木断面积之和
    Sum of basal area larger than the subject tree
    RD 0.95 0.34 0.24 2.74 对象木胸径与林分断面积平均胸径之比
    Ratio of DBH of a subject tree to stand average DBH
    DDM 0.47 0.18 0.13 1.00 对象木胸径与林分中最大林木胸径之比
    Ratio of DBH of a subject tree to the maximal DBH
    N/(株·hm− 2
    N/(tree·ha− 1)
    1 017 245 395 1 585 林分每公顷株数
    Number of trees per hectare
    BA/(m2·hm− 2
    BA/(m2·ha− 1)
    26.26 5.42 14.22 37.37 林分每公顷断面积
    Stand basal area per hectare
    Dg/cm 18.31 2.13 13.01 22.95 林分断面积平均胸径
    Quadratic mean DBH
    下载: 导出CSV

    表  2  部分气候因子统计表

    Table  2.   Summary statistics of selected climate variables

    因子
    Variable
    平均值
    Mean
    标准差
    Standard deviation
    最小值
    Minimum
    最大值
    Maximum
    说明
    Description
    Pratio 0.84 0.009 7 0.82 0.85 生长季平均降水量(4—9月)与年均降水量之比
    Ratio of average precipitation in growing season (April to September) to mean annual precipitation
    Asr/h 2 381.32 34.57 2 316.51 2 442.43 年总太阳辐射时长
    Annual total solar radiation duration
    Gspgsrh/mm 0.069 2 0.004 1 0.062 8 0.074 5 生长季平均降水量(4—9月)与生长季相对湿度(4—9月)之比
    Ratio of average precipitation in growing season (April to September) to relative humidity in growing season (April to September)
    Gssr/h 1 241.32 28.08 1 190.61 1 278.48 生长季太阳辐射时长(4—9月)
    Solar radiation duration in growing season (April to September)
    Amt/℃ 3.09 0.28 2.42 3.47 年平均气温
    Mean annual temperature
    Map/mm 509.27 16.22 476.84 536.52 年均降水量
    Mean annual precipitation
    下载: 导出CSV
  • [1] Jiang X Y, Huang J G, Cheng J, et al. Interspecific variation in growth responses to tree size, competition and climate of western Canadian boreal mixed forests[J]. Science of the Total Environment, 2018, 631−632: 1070−1078. doi: 10.1016/j.scitotenv.2018.03.099
    [2] Clark J S, Bell D M, Kwit M C, et al. Competition-interaction landscapes for the joint response of forests to climate change[J]. Global Change Biology, 2014, 20(6): 1979−1991. doi: 10.1111/gcb.12425
    [3] Rollinson C R, Kaye M W, Canham C D. Interspecific variation in growth responses to climate and competition of five eastern tree species[J]. Ecology, 2016, 97(4): 1003−1011.
    [4] Matsushita M, Takata K, Hitsuma G, et al. A novel growth model evaluating age-size effect on long-term trends in tree growth[J]. Functional Ecology, 2015, 29(10): 1250−1259. doi: 10.1111/1365-2435.12416
    [5] Martínez-Vilalta J, López B C, Loepfe L, et al. Stand- and tree-level determinants of the drought response of Scots pine radial growth[J]. Oecologia, 2012, 168(3): 877−888. doi: 10.1007/s00442-011-2132-8
    [6] Messaoud Y, Chen H Y H. The influence of recent climate change on tree height growth differs with species and spatial environment[J]. Plos One, 2011, 6(2): e14691. doi: 10.1371/journal.pone.0014691
    [7] Subedi N, Sharma M. Climate-diameter growth relationships of black spruce and jack pine trees in boreal Ontario, Canada[J]. Global Change Biology, 2013, 19(2): 505−516. doi: 10.1111/gcb.12033
    [8] Castagneri D, Nola P, Motta R, et al. Summer climate variability over the last 250 years differently affected tree species radial growth in a mesic Fagus-Abies-Picea old-growth forest[J]. Forest Ecology and Management, 2014, 320: 21−29. doi: 10.1016/j.foreco.2014.02.023
    [9] Chen L, Huang J G, Stadt K J, et al. Drought explains variation in the radial growth of white spruce in western Canada[J]. Agricultural and Forest Meteorology, 2017, 233: 133−142. doi: 10.1016/j.agrformet.2016.11.012
    [10] Cortini F, Filipescu C N, Groot A, et al. Regional models of diameter as a function of individual tree attributes, climate and site characteristics for six major tree species in Alberta, Canada[J]. Forests, 2011, 2(4): 814−831. doi: 10.3390/f2040814
    [11] Latreille A, Davi H, Huard F, et al. Variability of the climate-radial growth relationship among Abies alba trees and populations along altitudinal gradients[J]. Forest Ecology and Management, 2017, 396: 150−159. doi: 10.1016/j.foreco.2017.04.012
    [12] Toledo M, Poorter L, Peña-Claros M, et al. Climate is a stronger driver of tree and forest growth rates than soil and disturbance[J]. Journal of Ecology, 2011, 99(1): 254−264. doi: 10.1111/j.1365-2745.2010.01741.x
    [13] Ashraf M I, Zhao Z Y, Bourque C P A, et al. Integrating biophysical controls in forest growth and yield predictions with artificial intelligence technology[J]. Canadian Journal of Forest Research, 2013, 43(12): 1162−1171. doi: 10.1139/cjfr-2013-0090
    [14] Breiman L, Friedman J H, Stone C J, et al. Classification and regression trees[M]. London: Chapman and Hall, 1984.
    [15] De'Ath G. Boosted trees for ecological modeling and prediction[J]. Ecology, 2007, 88(1): 243−251. doi: 10.1890/0012-9658(2007)88[243:BTFEMA]2.0.CO;2
    [16] Kuhn M, Johnson K. Applied predictive modeling[M]. New York: Springer, 2013.
    [17] Breiman L. Random forest[J]. Machine Learning, 2001, 45(1): 5−32. doi: 10.1023/A:1010933404324
    [18] 高若楠, 苏喜友, 谢阳生, 等. 基于随机森林的杉木适生性预测研究[J]. 北京林业大学学报, 2017, 39(12):36−43.

    Gao R N, Su X Y, Xie Y S, et al. Prediction of adaptability of Cunninghamia lanceolata based on random forest[J]. Journal of Beijing Forestry University, 2017, 39(12): 36−43.
    [19] 欧强新, 李海奎, 雷相东, 等. 基于清查数据的福建省马尾松生物量转换和扩展因子估算差异解析——3种集成学习决策树模型的比较[J]. 应用生态学报, 2018, 29(6):2007−2016.

    Ou Q X, Li H K, Lei X D, et al. Difference analysis in estimating biomass conversion and expansion factors of masson pine in Fujian Province, China based on national forest inventory data: a comparison of three decision tree models of ensemble learning[J]. Chinese Journal of Applied Ecology, 2018, 29(6): 2007−2016.
    [20] 余黎, 雷相东, 王雅志, 等. 基于广义可加模型的气候对单木胸径生长的影响研究[J]. 北京林业大学学报, 2014, 36(5):22−32.

    Yu L, Lei X D, Wang Y Z, et al. Impact of climate on individual tree radial growth based on generalized additive model[J]. Journal of Beijing Forestry University, 2014, 36(5): 22−32.
    [21] 沈琛琛. 气候敏感的长白落叶松立地指数模型研究[D]. 北京: 中国林业科学研究院, 2012.

    Shen C C. Climate-sensitive site index model of Larix olgensis Henry[D]. Beijing: Chinese Academy of Forestry, 2012.
    [22] 周志华. 机器学习[M]. 北京: 清华大学出版社, 2016.

    Zhou Z H. Machine learning[M]. Beijing: Tsinghua University Press, 2016.
    [23] Goldstein A, Kapelner A, Bleich J, et al. Peeking inside the black box: visualizing statistical learning with plots of individual conditional expectation[J]. Journal of Computational and Graphical Statistics, 2015, 24(1): 44−65. doi: 10.1080/10618600.2014.907095
    [24] 张瑞英, 庞勇, 李增元, 等. 结合机载LiDAR和LANDSAT ETM+数据的温带森林郁闭度估测[J]. 植物生态学报, 2016, 40(2):102−115. doi: 10.17521/cjpe.2014.0366

    Zhang R Y, Pang Y, Li Z Y, et al. Canopy closure estimation in a temperate forest using airborne LiDAR and LANDSAT ETM+ data[J]. Chinese Journal of Plant Ecology, 2016, 40(2): 102−115. doi: 10.17521/cjpe.2014.0366
    [25] Breiman L. Statistical modeling: the two cultures[J]. Statistical Science, 2001, 16(3): 199−231.
    [26] Alam S A, Huang J G, Stadt K J, et al. Effects of competition, drought stress and photosynthetic productivity on the radial growth of white spruce in western Canada[J/OL]. Frontiers in Plant Science, 2017, 8: 1−15 [2018−03−15]. https://www.frontiersin.org/articles/10.3389/fpls.2017.01915/full.
    [27] 孟宪宇. 测树学[M]. 3版. 北京: 中国林业出版社, 2006.

    Meng X Y. Forest mensuration edition)[M]. 3rd ed. Beijing: China Forestry Publishing House, 2006.
  • 加载中
图(4) / 表(2)
计量
  • 文章访问数:  1408
  • HTML全文浏览量:  1011
  • PDF下载量:  98
  • 被引次数: 0
出版历程
  • 收稿日期:  2018-08-15
  • 修回日期:  2019-01-12
  • 网络出版日期:  2019-08-26
  • 刊出日期:  2019-09-01

目录

    /

    返回文章
    返回