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

不同滤波算法对反演叶面积指数的影响

梁勇奇 李明泽 杨瑞霞 耿同 李欢

梁勇奇, 李明泽, 杨瑞霞, 耿同, 李欢. 不同滤波算法对反演叶面积指数的影响[J]. 北京林业大学学报, 2020, 42(1): 54-64. doi: 10.12171/j.1000-1522.20180268
引用本文: 梁勇奇, 李明泽, 杨瑞霞, 耿同, 李欢. 不同滤波算法对反演叶面积指数的影响[J]. 北京林业大学学报, 2020, 42(1): 54-64. doi: 10.12171/j.1000-1522.20180268
Liang Yongqi, Li Mingze, Yang Ruixia, Geng Tong, Li Huan. Effects of different filter algorithms on deriving leaf area index (LAI)[J]. Journal of Beijing Forestry University, 2020, 42(1): 54-64. doi: 10.12171/j.1000-1522.20180268
Citation: Liang Yongqi, Li Mingze, Yang Ruixia, Geng Tong, Li Huan. Effects of different filter algorithms on deriving leaf area index (LAI)[J]. Journal of Beijing Forestry University, 2020, 42(1): 54-64. doi: 10.12171/j.1000-1522.20180268

不同滤波算法对反演叶面积指数的影响

doi: 10.12171/j.1000-1522.20180268
基金项目: 中科院先导专项地球大数据科学工程课题(XDA19030501),国家重点研发计划课题(2016YFC0503302)
详细信息
    作者简介:

    梁勇奇。主要研究方向:数字遗产。Email:yongqeeleeao@foxmail.com 地址:100094北京市海淀区西北旺镇邓庄南路9号中国科学院遥感与数字地球研究所

    责任作者:

    李明泽,博士,教授。主要研究方向:森林经理。Email:mingzelee@163.com 地址:150040黑龙江省哈尔滨市香坊区和兴路26号东北林业大学林学院

  • 中图分类号: S771.8

Effects of different filter algorithms on deriving leaf area index (LAI)

  • 摘要: 目的使用离散型激光雷达数据反演叶面积指数(LAI)的过程中,数据预处理的关键步骤为激光雷达滤波。穿透指数(LPI)作为反演过程中的重要变量,需要根据点云的类型计算,从而直接受到滤波精度的影响。因此,滤波算法的精度能间接影响到反演LAI的精度。虽然滤波算法不断改进,滤波精度逐渐提高,应用在越来越多的场景,但关于不同滤波算法对反演LAI精度影响的探讨较少。方法本文通过对机载LiDAR滤波算法历史、发展和现状的调研,最终选择混合滤波算法(Hybrid)、自适应不规则三角网滤波算法(ATIN)、形态学滤波算法(Morph)和基于坡度滤波算法(Slope)为研究对象;分别使用这4种算法,得到点云中的地面点;根据Beer-Lambert定律,反演帽儿山国家森林公园落叶松林和榆树林的LAI;以经过评估的精度更高的Hybrid算法为标准,计算另外3种算法的滤波精度和LPI偏差;对比分析LAI反演模型的平均精度;最后,通过分析不同误差来源的影响强度,确定了反演LAI时较好的滤波算法。结果在最佳的采样半径下,经过Hybrid、ATIN、Morph和Slope滤波算法处理,LAI反演模型的平均精度,在落叶松林,R2分别为:0.900 3、0.876 3、0.892 5、0.877 0;RMSE分别为:0.105 6、0.134 5、0.109 7、0.133 2;在榆树林,R2分别为:0.914 4、0.903 0、0.887 2、0.900 0;RMSE分别为:0.269 0、0.201 7、0.189 4、0.207 0。在落叶松林,I类误差较大的Morph算法,能保证较高的模型精度;而II类误差较大的Slope和ATIN算法对应的反演模型精度较低。结论经不同滤波算法处理得到的LAI反演模型精度存在差异,经混合滤波算法处理其对应的LAI反演模型精度更高,形态学滤波算法的滤波精度较低,对应的反演模型精度较高;滤波算法导致的I、II类误差中,II类误差对LAI反演模型的影响更大。

     

  • 图  1  LAI测站在样地内的分布

    Figure  1.  Distribution of LAI plots in the site

    图  2  固定样地的分布

                  TIN由10 m分辨率等高线生成TIN generated by 10 m resolution contour

    Figure  2.  Distribution of fixed sites

    图  3  LPI穿透界面效果

    Figure  3.  Effects of LPI penetrating interface

    图  4  样本集筛选结果示例

    Figure  4.  Example of selected samples

    图  5  最佳采样半径下模型精度和滤波算法的关系

    Figure  5.  Relation between models and filters under best sampling radius

    图  6  样地合并的滤波精度

    Figure  6.  Filtering accuracy of merged sites

    图  7  样地合并的I、II类误差

    Figure  7.  I and II error of merged sites

    图  8  落叶松林的LPI偏差

    Figure  8.  Deviation of LPI in larch

    图  9  榆树林的LPI偏差

    Figure  9.  Deviation of LPI in elm

    图  10  同一林型下样地的I类误差

    Figure  10.  I error of sites between different trees

    图  11  同一林型下样地的II类误差

    Figure  11.  II error of sites between different trees

    表  1  样地点云密度

    Table  1.   Point cloud density of sites

    森林类型
    Forest type
    样地编号
    Site No.
    点云密度/(点·m− 2
    Point cloud density/(point·m− 2
    落叶松林 L1 3.86
    Larix gmelinii L2 3.50
    L3 4.03
    榆树林 U1 3.53
    Ulmus pumila U2 3.62
    U3 6.66
    U4 6.69
    U5 7.02
    下载: 导出CSV

    表  2  样地实测LAI数据特征

    Table  2.   Characteristic of LAI plot data

    项目 Item落叶松 Larix gmelinii榆树 Ulmus pumila
    L1L2L3U1U2U3U4U5
    最小值 Min. 3.43 1.90 2.95 3.12 0.70 3.50 3.65 3.51
    最大值 Max. 5.99 5.31 5.18 6.37 6.39 8.34 6.27 6.82
    均值 Average 4.55 4.05 3.88 4.71 3.56 5.20 4.91 5.40
    方差 Variance 0.73 0.77 0.52 0.73 1.18 0.88 0.67 0.81
    下载: 导出CSV

    表  3  ISPRS对滤波算法的系统性评价(植被和不连续性部分)

    Table  3.   Systematic evaluation for filter algorithms by ISPRS in 2004 (vegetation and discontinuity part)

    项目 ItemElmqvistSohnRoggeroBrovelliWackAxelssonSitholePfeifer
    植被位置 Vegetation position 平坦 Flat place *** *** *** *** *** *** *** ***
    坡上 On slope *** *** ** ** ** ** ** ***
    低处 Low place *** ** ** ** *** ** ** ***
    不连续地形 Discontinuity landform 陡坡 Steep slope * * * * ** ** * **
    山脊 Sharp ridge * * * * ** * * *
    注:***表示该点云的滤除精度在90%以上,**为精度在50% ~ 90%,*表示精度在50%以下;该表引自参考文献[9];Elmqvist代表其于2001年提出的活动曲面法,Sohn代表其于2002年提出的正则法,Roggero代表其于2000年的形态学法,Brovelli代表其于2002年提出的分级样条插值法,Wack代表其于2002年提出的预定义局部最小值法,Axelsson代表其于1999年提出的自适应TIN法,Sithole代表其于2000年提出的预定义坡度法,Pfeifer代表其于2001年提出的分级稳定性插值法。Notes: the accuracy of filtering the point cloud up to 90%, between 50%−90%, and lower than 50%, are defined as ***, **, *, respectively. This table is cited from the Ref. [9]. Elmqvist represents the active surface method proposed in 2001, Sohn represents the regularization method proposed in 2002, Roggero represents its morphological method in 2000, Brovelli represents its hierarchical spline interpolation method proposed in 2002, Wack represents its predefined local minimum height method proposed in 2002, Axelsson represents its adaptive TIN method proposed in 1999, Sithole represents predefined local minimum slope method proposed in 2000. Pfeifer represents the hierarchical stability interpolation method proposed by Pfeifer in 2001.
    下载: 导出CSV

    表  4  误差计算

    Table  4.   Error calculation

    项目 Item待检验数据
    Data to be tested
    计算公式 Calculation formula
    地面点
    Ground point
    非地面点
    None-ground point
    TI=b/(a+b)Po=(a+d)/e
    标准数据
    Standard data
    地面点 Ground point a b TII=c/(c+d) Pc=((a+b) × (a+c)+(c+d) × (b+d))/e2
    非地面点 None-ground point c d TE=(b+c)/e kpa=(PoPc)/(1− Pc)
    注:TI为I类误差,TII为II类误差,TE为总误差,e为所有点的和,kpa为kappa系数,PoPc为计算kpa的中间变量。该表引自参考文献[11]。Notes: I error, II error and total error are defined as ‘TI’, ‘TII’, ‘TE’, respectively, and ‘e’ is the total amount of the points,Po and Pc are intermediate variables for calculating kpa. This table is cited from the Ref. [11].
    下载: 导出CSV

    表  5  样地滤波误差

    Table  5.   Filter error of sites

    样地类型
    Plot type
    滤波算法
    Filtering algorithm abbreviation
    样地编号
    Site No.
    TITIITEPoPcKappa
    落叶松林
    Larix gmelinii forest
    ATIN L1 0.362 0.004 0.082 0.912 0.700 0.707
    Morph 0.510 0.001 0.112 0.838 0.708 0.444
    Slope 0.362 0.003 0.081 0.913 0.701 0.709
    ATIN L2 0.264 0.007 0.046 0.951 0.768 0.790
    Morph 0.519 0.002 0.079 0.900 0.794 0.516
    Slope 0.265 0.004 0.043 0.954 0.770 0.801
    ATIN 0.330 0.006 0.046 0.950 0.809 0.741
    Morph L3 0.541 0.002 0.069 0.907 0.828 0.460
    Slope 0.331 0.006 0.046 0.950 0.809 0.740
    ATIN U1 0.473 0.002 0.095 0.835 0.722 0.405
    榆树林
    Ulmus pumila forest
    Morph 0.472 0.002 0.095 0.835 0.722 0.408
    Slope 0.493 0.004 0.101 0.888 0.736 0.575
    ATIN U2 0.492 0.001 0.032 0.959 0.907 0.559
    Morph 0.482 0.001 0.032 0.959 0.906 0.567
    Slope 0.323 0.001 0.022 0.976 0.898 0.764
    ATIN 0.324 0.019 0.046 0.951 0.849 0.674
    Morph U3 0.500 0.002 0.046 0.938 0.873 0.512
    Slope 0.349 0.002 0.032 0.963 0.863 0.732
    ATIN 0.381 0.010 0.117 0.872 0.629 0.655
    Morph U4 0.526 0.002 0.152 0.776 0.629 0.396
    Slope 0.389 0.007 0.116 0.871 0.631 0.650
    ATIN 0.268 0.011 0.048 0.950 0.775 0.777
    Morph U5 0.506 0.004 0.075 0.903 0.801 0.515
    Slope 0.271 0.007 0.045 0.953 0.778 0.788
    注:该表引自参考文献[11]。Note: this table is cited from the Ref. [11].
    下载: 导出CSV

    表  6  各滤波算法和采样半径下筛选的R2和RMSE

    Table  6.   R2 and RMSE under different filters and radius

    项目
    Item
    滤波算法
    Filtering algorithm
    abbreviation
    采样半径 Sampling radius/m
    指标 Index51015202530
    落叶松林样本集
    Sample set of larch
    ATIN R2 0.826 5 0.870 6 0.836 8 0.875 0 0.821 8 0.823 6
    RMSE 0.181 7 0.144 9 0.148 3 0.137 8 0.148 3 0.144 9
    n 30 28 33 30 32 32
    Hybrid R2 0.865 2 0.899 9 0.849 6 0.789 3 0.720 9 0.760 5
    RMSE 0.192 4 0.134 2 0.144 9 0.154 9 0.176 1 0.164 3
    n 30 31 29 29 32 29
    Morph R2 0.894 3 0.879 6 0.860 0 0.796 3 0.763 7 0.764 1
    RMSE 0.164 3 0.130 4 0.161 2 0.154 9 0.167 3 0.164 3
    n 30 29 31 29 30 29
    Slope R2 0.918 0 0.878 8 0.872 4 0.813 1 0.762 2 0.762 9
    RMSE 0.144 9 0.144 9 0.151 7 0.161 2 0.170 3 0.167 3
    n 30 29 28 30 32 30
    榆树林样本集
    Sample set of elm
    ATIN R2 0.853 2 0.863 5 0.907 6 0.884 5 0.863 9 0.869 0
    RMSE 0.426 2 0.380 7 0.385 1 0.371 3 0.385 1 0.380 7
    n 49 48 47 51 56 57
    Fusion R2 0.893 2 0.885 9 0.917 7 0.900 1 0.867 1 0.876 1
    RMSE 0.438 6 0.366 3 0.380 7 0.393 6 0.419 6 0.405 4
    n 53 45 50 54 56 53
    Morph R2 0.853 8 0.842 1 0.890 8 0.888 9 0.874 7 0.870 2
    RMSE 0.405 4 0.361 1 0.401 6 0.393 6 0.409 1 0.405 4
    n 51 53 53 57 57 56
    Slope R2 0.896 7 0.880 5 0.911 4 0.909 3 0.858 3 0.875 1
    RMSE 0.380 7 0.380 7 0.389 4 0.401 6 0.412 7 0.409 1
    n 57 58 52 57 56 57
    注:n为选定的样本集个数。Note: n is the number of the selected samples.
    下载: 导出CSV

    表  7  平均的模型精度

    Table  7.   Average accuracy of modelling and testing

    项目
    Item
    滤波算法
    Filtering algorithm
    abbreviation
    指标
    Index
    采样半径 Sampling radius/m
    51015202530
    落叶松林的模型精度
    Model accuracy of larch
    ATIN R2 0.822 4 0.876 3 0.825 9 0.857 2 0.767 1 0.798 2
    RMSE 0.165 8 0.134 5 0.140 3 0.114 5 0.135 7 0.131 7
    Hybrid R2 0.858 0 0.900 3 0.844 5 0.778 3 0.714 3 0.726 5
    RMSE 0.165 4 0.105 6 0.126 7 0.138 2 0.161 5 0.149 9
    Morph R2 0.869 3 0.892 5 0.849 6 0.789 7 0.738 5 0.728 9
    RMSE 0.159 8 0.109 7 0.126 7 0.135 0 0.155 2 0.151 6
    Slope R2 0.907 1 0.877 0 0.859 5 0.809 1 0.765 2 0.752 3
    RMSE 0.133 8 0.133 2 0.118 1 0.135 7 0.152 8 0.150 9
    榆树林的模型精度
    Model accuracy of elm
    ATIN R2 0.841 7 0.847 5 0.903 0 0.878 9 0.858 0 0.889 7
    RMSE 0.237 7 0.225 1 0.201 7 0.206 7 0.210 5 0.217 6
    Hybrid R2 0.889 7 0.875 0 0.914 4 0.900 3 0.869 8 0.871 8
    RMSE 0.217 6 0.220 0 0.269 0 0.158 3 0.178 5 0.185 9
    Morph R2 0.858 9 0.838 7 0.887 2 0.885 1 0.872 6 0.869 2
    RMSE 0.204 5 0.189 4 0.196 9 0.198 3 0.194 0 0.186 4
    Slope R2 0.894 5 0.877 1 0.900 0 0.851 2 0.867 2 0.881 5
    RMSE 0.232 3 0.207 0 0.211 7 0.268 3 0.160 5 0.154 2
    下载: 导出CSV
  • [1] Coppin P, Jonckheere I, Nackaerts K, et al. Digital change detection methods in ecosystem monitoring: a review[J]. International Journal of Remote Sensing, 2004, 25(9): 1565−1596. doi: 10.1080/0143116031000101675
    [2] You H, Wang T, Skidmore A K, et al. Quantifying the effects of normalisation of airborne LiDAR intensity on coniferous forest leaf area index estimations[J]. Remote Sensing, 2017, 9(2): 163−179. doi: 10.3390/rs9020163
    [3] Peduzzi A, Wynne R H, Fox T R, et al. Estimating leaf area index in intensively managed pine plantations using airborne laser scanner data[J]. Forest Ecology & Management, 2012, 270(4): 54−65.
    [4] Sumnall M J, Fox T R, Wynne R H, et al. Estimating leaf area index at multiple heights within the understorey component of Loblolly pine forests from airborne discrete-return LiDAR[J]. International Journal of Remote Sensing, 2016, 37(1): 78−99. doi: 10.1080/01431161.2015.1117683
    [5] 骆社周, 王成, 张贵宾, 等. 机载激光雷达森林叶面积指数反演研究[J]. 地球物理学报, 2013, 56(5):1467−1475. doi: 10.6038/cjg20130505

    Luo S Z, Wang C, Zhang G B, et al. Forest leaf area index (LAI) inversion using airborne LiDAR data[J]. Geophys, 2013, 56(5): 1467−1475. doi: 10.6038/cjg20130505
    [6] Chen T, Akciz S O, Hudnut K W, et al. Fault-slip distribution of the 1999 mw 7.1 hector mine earthquake, California, estimated from postearthquake airborne LiDAR data[J]. Bulletin of the Seismological Society of America, 2015, 105: 776−790. doi: 10.1785/0120130108
    [7] 黄作维, 刘峰, 胡光伟. 基于多尺度虚拟格网的LiDAR点云数据滤波改进方法[J]. 光学学报, 2017, 37(8):346−355.

    Huang Z W, Liu F, Hu G W. Improved method for LiDAR point cloud data filtering based on hierarchical pseudo-grid[J]. Acta Optic Sin, 2017, 37(8): 346−355.
    [8] Solberg S. Comparing discrete echoes counts and intensity sums from ALS for estimating forest LAI and gap fraction[C/OL]//International Conference on Silvilaser, Sept. 17−19, 2008: 247−256[2018−05−06]. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.591.1374&rep=rep1&type=pdf.
    [9] Sithole G, Vosselman G. Experimental comparison of filter algorithms for Bare-Earth extraction from airborne laser scanning point clouds[J]. ISPRS Journal of Photogrammetry & Remote Sensing, 2004, 59: 85−101.
    [10] Favorskaya M N, Jain L C. Handbook on advances in remote sensing and geographic information systems[M]. Cham:Springer International Publishing, 2017.
    [11] Pingel T J, Clarke K C, Mcbride W A. An improved simple morphological filter for the terrain classification of airborne LiDAR data[J]. ISPRS Journal of Photogrammetry & Remote Sensing, 2013, 77: 21−30.
    [12] Zhao X, Guo Q, Su Y, et al. Improved progressive TIN densification filtering algorithm for airborne LiDAR data in forested areas[J]. ISPRS Journal of Photogrammetry & Remote Sensing, 2016, 117: 79−91.
    [13] Polat N, Uysal M. Investigating performance of airborne LiDAR data filtering algorithms for DTM generation[J]. Measurement, 2015, 63: 61−68. doi: 10.1016/j.measurement.2014.12.017
    [14] Axelsson P. DEM generation from laser scanner data using adaptive TIN models[J]. International Archives of Photogrammetry & Remote Sensing, 2000(33): 110−116.
    [15] Zhang K Q, Chen S C, Whitman D, et al. A progressive morphological filter for removing non-ground measurements from airborne LiDAR data[C]. IEEE Transactions on Geoscience and Remote Sensing, 2003 (41): 872−882.
    [16] Vosselman G. Slope based filtering of laser altimetry data[C/OL]. Amsterdam: International Archives of Photogrammetry & Remote Sensing, 2000[2018−05−06]. https://www.researchgate.net/publication/228719860_Slope_based_filtering_of_laser_altimetry_data.
    [17] Zhao K, García M, Liu S, et al. Terrestrial LiDAR remote sensing of forests: maximum likelihood estimates of canopy profile, leaf area index, and leaf angle distribution[J]. Agricultural & Forest Meteorology, 2015, 209−210: 100−113.
    [18] Solberg S, Hill R, Suarez R. Mapping gap fraction, LAI and defoliation using various ALS penetration variables[J]. International Journal of Remote Sensing, 2010, 31(5): 1227−1244. doi: 10.1080/01431160903380672
    [19] Morsdorf F, Kötz B, Meier E, et al. Estimation of LAI and fractional cover from small footprint airborne laser scanning data based on gap fraction[J]. Remote Sensing of Environment, 2006, 104(1): 50−61. doi: 10.1016/j.rse.2006.04.019
    [20] Hyyppä J, Hyyppä H, Leckie D, et al. Review of methods of small-footprint airborne laser scanning for extracting forest inventory data in boreal forests[J]. International Journal of Remote Sensing, 2008, 29(5): 1339−1366. doi: 10.1080/01431160701736489
    [21] Deng S S, Shi W Z. Integration of different filter algorithms for improving the ground surface extraction from airborne LiDAR data[C/OL]. Proceedings of 8th International Symposium on Spatial Data Quality Implementation Science. Hong Kong: International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2013, XL-2/W1(2): 105−110[2018−05−06]. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.917.8968&rep=rep1&type=pdf.
  • 加载中
图(11) / 表(7)
计量
  • 文章访问数:  1066
  • HTML全文浏览量:  631
  • PDF下载量:  73
  • 被引次数: 0
出版历程
  • 收稿日期:  2018-08-21
  • 修回日期:  2018-10-19
  • 网络出版日期:  2019-10-16
  • 刊出日期:  2020-01-14

目录

    /

    返回文章
    返回